Engineering structure deformation monitoring gravity reference and horizontal displacement monitoring inverted plumb line intelligent agent

Abstract

The application discloses an engineering structure deformation monitoring gravity reference and horizontal displacement monitoring inverted plumb line intelligent agent, which comprises a monitoring and sensing layer, a cognitive decision-making layer and a control and execution layer; the monitoring and sensing layer is used for acquiring plumb line steel wire displacement, buoy state, steel wire limiting distance and environmental parameters; the cognitive decision-making layer is used for steel wire steady state discrimination, double measurement return quality control, multi-dimensional error correction, dynamic reference decoupling, deformation prediction and equipment health diagnosis; the control and execution layer is used for implementing buoyancy-tension interlocking, environmental regulation, emergency interruption and lens self-cleaning. The layers form a sensing, decision-making, execution and feedback closed loop through industrial Internet of Things, can reduce measurement errors caused by non-steady state data and environmental and local gravity field abnormalities, realize high-precision and unattended horizontal displacement monitoring, and are suitable for engineering structures such as dams, bridges, high-rise buildings and landslide bodies.

Description

Technical Field

[0001] This invention relates to the field of engineering structure safety monitoring technology, specifically to an intelligent agent for monitoring engineering structure deformation using a gravity reference and for monitoring horizontal displacement using an inverted plumb line. Background Technology

[0002] Large-scale water conservancy projects, bridges, high-rise buildings, and landslide control projects are prone to horizontal displacement and deep foundation deformation during long-term service due to factors such as water load, temperature changes, wind load, seismic action, foundation deformation, and time-varying effects of materials. These deformations are important indicators for evaluating the overall stability and operational safety of the engineering structure. Obtaining true and reliable absolute displacement requires establishing a long-term stable spatial reference unaffected by the deformation of the monitored structure itself. Plumb line monitoring technology, due to its intuitive principle, high accuracy, and good stability, has been widely applied in projects such as dams, bridge towers, high-rise buildings, and landslides. Existing plumb line devices typically fix anchor blocks deep in stable bedrock, use stainless steel wire as the reference carrier, and utilize the buoyancy of floats to tension the wire, keeping it vertical under the influence of gravity. The positional change of the engineering structure relative to the wire reference is then measured using a plumb line coordinate instrument. However, the actual working state of the plumb line is also affected by factors such as wire tension, float attitude, protective pipe boundaries, and the site environment.

[0003] Traditional plumb line measurements are primarily conducted manually. Before and after observation, numerous procedures are required, including instrument zero-point calibration, installation and leveling, adjusting the steel wire, waiting for the instrument to come to rest, reading records, and verification. This process is cumbersome and easily affected by factors such as personnel experience, reading parallax, recording errors, and on-site conditions like high humidity, dust, and dripping water. Manual data also requires offline processing and calculation, resulting in time lags in feedback and making it difficult to promptly detect sudden or gradual anomalies. Furthermore, observation points are often located near corridors, shafts, or boreholes, leading to a large workload for manual inspections, high maintenance costs, and risks associated with confined space operations.

[0004] While existing automated plumb line coordinate instruments can achieve automatic data acquisition and remote transmission, most systems still rely primarily on static instantaneous readings. They lack the ability to effectively assess the entire process of "movement-reset-restoration-retesting," failing to accurately identify conditions such as wire micro-vibration, jamming, near-wall friction, or incomplete stability, easily leading to invalid data entering subsequent analysis. Simultaneously, the immersion depth of the pontoon, horizontal sway, and the safe distance between the wire and the protective pipe typically lack real-time monitoring, resulting in buoyancy, tension, and structural boundaries remaining imperceptible. Furthermore, existing systems do not adequately consider the effects of temperature gradients, airflow disturbances, and changes in the local gravity field. Especially during significant fluctuations in reservoir water levels or landslide mass migration, slight changes in the local gravity direction can cause additional offsets in the wire or pontoon, which can overlap with the actual structural displacement. Existing processing methods, primarily based on statistical filtering and post-hoc gross error removal, are insufficient to identify interference at its physical source and may also mistakenly delete genuine anomalies under conditions such as earthquakes or sudden water level changes. In addition, lens contamination, equipment deterioration and environmental anomalies mostly rely on manual inspection and handling, and the frequency of observation is difficult to adapt to the working conditions.

[0005] Therefore, existing plumb line monitoring technologies still suffer from problems such as insufficient process quality control, incomplete baseline state perception, difficulty in quantifying and eliminating environmental interference, and passive and lagging operation and maintenance, making it difficult to meet the engineering monitoring requirements of high precision, high reliability, continuous operation, and unattended operation. To address these issues, a gravity baseline for monitoring engineering structural deformation and a plumb line intelligent agent for monitoring horizontal displacement are designed. Summary of the Invention

[0006] Based on the above description, this invention provides an intelligent agent for monitoring the deformation of engineering structures using a gravity reference and for monitoring horizontal displacement using an inverted plumb line, in order to solve the problems of existing inverted plumb line monitoring systems, such as the lack of effective quality control in the data acquisition process, insufficient perception of measurement reference and safety boundary status, difficulty in quantifying and eliminating interference from environmental factors and local gravity field anomalies, difficulty in real-time diagnosis of equipment operating status, and passive and lagging observation strategies and operation and maintenance methods.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0008] An intelligent agent for monitoring structural deformation using a gravity reference and for monitoring horizontal displacement using an inverted plumb line, comprising:

[0009] The system comprises a monitoring and sensing layer, a cognitive decision-making layer, and a control execution layer. The monitoring and sensing layer, located at the observation site of the plumb line device, includes a displacement sensing module, a state sensing module, a clearance sensing module, and an environmental sensing module. It is used to simultaneously acquire the spatial coordinates and horizontal displacement of the plumb line stainless steel wire, the working status of the float and optical measurement components, the safe clearance between the plumb line stainless steel wire and the protective tube, and observation environmental parameters. The cognitive decision-making layer, deployed on a remote cloud platform, includes a cognitive unit and a decision-making unit. It is used to perform steady-state discrimination, double-return consistency verification, and multi-dimensional object analysis on the multi-source data output by the monitoring and sensing layer. The system includes field correction, dynamic benchmark decoupling, structural deformation prediction, equipment health diagnosis, and maintenance decision-making. The control execution layer is located at the observation site and includes a buoyancy-tension safety interlocking actuator, an environmental adaptive control actuator, an emergency hard interruption protection mechanism, and a CCD lens cleanliness adaptive self-cleaning system. It is used to execute observation interlocking, buoyancy compensation, environmental control, emergency protection, and optical cleaning according to the control commands output by the cognitive decision layer. The monitoring and sensing layer, cognitive decision layer, and control execution layer form a closed-loop system with interconnected data flow, command flow, and status feedback flow through an industrial Internet of Things communication link.

[0010] Furthermore, the displacement sensing module includes an intelligent vertical coordinate instrument, which employs a dual orthogonal optical imaging structure, comprising a first optical imaging system positioned along the engineering axis, a second optical imaging system positioned perpendicular to the engineering axis, a high-resolution linear CCD sensor, and an embedded image processing unit. The transmitting and receiving ends of the first and second optical imaging systems are respectively arranged on opposite sides of the vertical stainless steel wire, used to illuminate the vertical stainless steel wire with a parallel beam and form a corresponding shadow image on the linear CCD sensor. The embedded image processing unit is configured to perform environmental background subtraction, adaptive exposure compensation, image enhancement, gradient extraction, and subpixel edge localization on the shadow image, and calculate its two-dimensional center coordinates and horizontal displacement components relative to the initial reference state or the previous observation period based on the edge coordinates on both sides of the vertical stainless steel wire. The intelligent vertical coordinate instrument retains its existing installation position and is connected to the cognitive decision layer to output displacement data without changing the original inverted vertical mechanical structure.

[0011] Furthermore, the state quantity sensing module includes a vertical non-contact tossing, resetting, and stability control unit. This control unit comprises a flexible electromagnetic drive mechanism, a torque detection sensor, a precision linear guide, and a PLC control module. The flexible electromagnetic drive mechanism is mounted on the precision linear guide and maintains a non-contact gap with the vertical stainless steel wire. The torque detection sensor is connected in series in the transmission link of the flexible electromagnetic drive mechanism. The PLC control module is communicatively connected to the flexible electromagnetic drive mechanism, the torque detection sensor, and the intelligent vertical coordinate instrument. The PLC control module is configured to control the flexible electromagnetic drive mechanism to apply a preset stroke and preset torque to the vertical stainless steel wire before effective displacement acquisition. Then, the disturbance force is removed, allowing the vertical stainless steel wire to freely reset under the buoyancy of the float and the damping of the liquid. The module determines whether the wire has reached a stability threshold based on the continuously output coordinate sequence from the intelligent vertical coordinate instrument. When the feedback torque exceeds the safety threshold, the PLC control module controls the flexible electromagnetic drive mechanism to stop and retract, and uses timing mutual exclusion logic to ensure that the tossing action, reset action, and effective sampling action do not occur simultaneously.

[0012] Furthermore, the state quantity sensing module also includes a real-time monitoring unit for the sway and wall-collision distance of the eddy current type float and a real-time monitoring unit for the immersion depth change of the magnetostrictive type float. The real-time monitoring unit for the sway and wall-collision distance of the eddy current type float includes four eddy current probes orthogonally and symmetrically arranged at 0°, 90°, 180°, and 270°, and a vector calculation module. The four eddy current probes are arranged circumferentially around the float and the oil tank to synchronously acquire four channels of gap data. The vector calculation module is used to calculate the two-dimensional sway position of the float based on the differential data relative to the eddy current probes. The system measures the displacement vector and the minimum micro-distance between the float and the oil drum wall, and generates a float near-wall alarm when the minimum micro-distance is lower than a preset alarm threshold. The real-time monitoring unit for the change in the immersion depth of the magnetostrictive float includes a waveguide wire, a float magnetic ring rigidly connected to the float and rising and falling synchronously with the float, and a pulse detection circuit. The pulse detection circuit is used to calculate the position change of the float magnetic ring and the immersion depth of the float based on the propagation time difference between the current pulse and the torsional wave, and outputs a buoyancy attenuation alarm when the immersion depth of the float is insufficient, providing state parameters for judging the tension state of the steel wire and correcting the displacement data.

[0013] Furthermore, the state quantity sensing module also includes a CCD lens surface cleanliness online monitoring unit and an intelligent vertical coordinate instrument working status visual monitoring unit. The CCD lens surface cleanliness online monitoring unit includes a high-contrast checkerboard target set outside the optical system window and a cleanliness evaluation unit that reuses the intelligent vertical coordinate instrument image processing unit. The cleanliness evaluation unit is used to acquire images of the checkerboard target during non-effective sampling periods, form a lens cleanliness evaluation value based on the gradient energy of the target image, and output an optical degradation warning and acquisition lockout signal when the lens cleanliness evaluation value is lower than the health threshold. The intelligent vertical coordinate instrument working status visual monitoring unit includes a low-light CMOS image sensor, an infrared fill light, and an edge preprocessing module. The edge preprocessing module is used to perform region of interest extraction and inter-frame difference on the optical lens and wire observation area of ​​the intelligent vertical coordinate instrument. When receiving a measurement data abnormality signal or a reset action signal, it switches to high-speed recording mode and saves video frames within a preset time window before and after the abnormal event. The video frames are bound to the corresponding displacement data, equipment status label, and warning information through event identifiers.

[0014] Furthermore, the clearance sensing module includes a four-dimensional capacitive steel wire tube internal position and clearance sensing unit. This unit comprises four capacitance probes arranged symmetrically at 90° angles on the inner wall of the critical section of the protective tube, a multi-channel capacitance detection circuit, and a spatial calculation module. The sensing electrodes of the four capacitance probes and the vertical stainless steel wires respectively constitute independent capacitance measurement channels. The multi-channel capacitance detection circuit is used to synchronously acquire the capacitance values ​​of each capacitance measurement channel and obtain the net distance data in four directions based on the pre-calibrated correspondence between capacitance values ​​and electrode spacing. The spatial calculation module is used to construct an overdetermined system of equations based on the net distance data in four directions. It calculates the two-dimensional center coordinates of the vertical stainless steel wire in the cross-section of the protective tube and the minimum net distance from the vertical stainless steel wire to the inner wall of the protective tube through circle center fitting. When the measurement residual of a single capacitive probe exceeds the abnormal rejection threshold, the data of that capacitive probe is rejected and the remaining data is used to re-perform circle center fitting. The spatial calculation module is also used to output a near-wall warning signal or a contact alarm signal based on the first-level near-wall warning threshold and the second-level contact alarm threshold, and to link the vertical non-contact toggle reset and stability control unit or the emergency hard interruption protection mechanism.

[0015] Furthermore, the environmental quantity sensing module includes a five-element weather station integrating a temperature sensor, humidity sensor, air pressure sensor, ultrasonic anemometer, and ultrasonic wind direction sensor. This five-element weather station is connected to the data bus of the monitoring and sensing layer via a digital interface, used to synchronously output temperature, humidity, air pressure, wind speed, and wind direction parameters. The ultrasonic anemometer and ultrasonic wind direction sensor adopt a solid-state structure without mechanical rotating parts, and calculate the two-dimensional wind speed component, instantaneous wind speed, and wind direction angle based on the time difference of sound wave propagation in orthogonal directions. The environmental quantity sensing module is configured to send temperature and air pressure parameters to the displacement and state sensing modules for environmental compensation of the original measurement data, and generate a disturbed state identifier and acquire a lockout signal based on the sliding average wind speed and instantaneous wind speed within a preset time window. The displacement, state, clearance, and environmental quantity sensing modules are connected to the same EtherCAT industrial bus and achieve timestamp alignment based on the PPS clock synchronization signal to form a structured multi-dimensional sensing data package containing wire coordinates, equipment status, boundary clearance, environmental parameters, and visual evidence indexes.

[0016] Furthermore, the cognitive unit includes a wire pre-decoupling and dual-return intelligent acquisition module, a multi-dimensional physical field real-time correction engine, an optical environment intelligent cognitive module, and an adaptive observation strategy engine. The wire pre-decoupling and dual-return intelligent acquisition module controls the vertical stainless steel wire to sequentially complete pre-disturbance, free reset, forced static placement, visual and coordinate dual steady-state determination, and two coordinate acquisitions. It outputs effective displacement data or a re-measurement command based on the comparison between the coordinate difference of the two re-measurements and the limit threshold. The multi-dimensional physical field real-time correction engine extracts ambient temperature, buoy immersion depth, buoy two-dimensional yaw displacement vector, and original effective displacement data. The steel wire tension is indirectly calculated by the immersion depth of the float, and the time-varying creep error of the steel wire and the sway coupling error of the float are progressively corrected. The optical environment intelligent cognition module is used to receive the lens cleanliness assessment value output by the state quantity sensing module, determine the data confidence level according to the lens cleanliness assessment value, and output a cleaning trigger command when the optical channel deteriorates. The adaptive observation strategy engine is used to adjust the observation period and output the working condition lockout signal or trigger the benchmark verification according to the minimum net distance of the steel wire output by the limit quantity sensing module, the wind speed, temperature change rate, load change rate, earthquake trigger signal and physical benchmark drift alarm output by the environmental quantity sensing module.

[0017] Furthermore, the decision-making unit includes a data storage and model evolution module, a safety boundary digital twin mapping module, a multimodal fusion prediction module, a dynamic benchmark decoupling module, and a predictive maintenance scheduling module. The data storage and model evolution module aggregates high-reliability correction displacement data, environmental factor data, equipment status data, visual evidence indexes, early warning tags, and action logs, and performs offline identification and online updates of the physical correction model parameters. The safety boundary digital twin mapping module maps the two-dimensional center coordinates of the vertical stainless steel wire, the cross-sectional coordinates of the protective pipe, and the inner wall radius of the protective pipe to the three-dimensional digital twin of the engineering structure. The twin model forms the full-section net distance distribution and risk section identification between the vertical stainless steel wire and the protective pipe; the multimodal fusion prediction module is used to predict the structural deformation trend and equipment health degradation trajectory based on displacement, environment, load, pontoon sway and equipment status characteristics; the dynamic benchmark decoupling module is used to establish a dynamic prediction benchmark based on ambient temperature, structural load, local gravity field anomaly parameters and historical displacement data, and separate structural anomaly residuals and physical benchmark drift from high-confidence corrected displacement data; the predictive maintenance scheduling module is used to generate maintenance work orders based on equipment health, remaining effective life and fault tags.

[0018] Furthermore, the buoyancy-tension safety interlock actuator includes a micro-injection pump, an oil tank, a filter, an electromagnetic control valve, and a sampling link hard-cutoff component. This is used to replenish transformer oil to the float when the float's immersion depth is insufficient and the hard-cutoff condition is not met, and to cut off the observation trigger circuit when the wire tension exceeds the limit or buoyancy compensation fails. The environmental adaptive control actuator includes a dehumidifying fan, a heating rod, and a semiconductor temperature control device, used to adjust the observation environment according to the environmental exceedance interlock signal. The emergency hard interruption protection mechanism includes an independent hardware control circuit, an overcurrent detection module, and a hardware relay, used to ensure that the vertical stainless steel wire contacts the protection tube or... In the event of overcurrent or overload, the power supply to non-core peripherals is cut off, and an emergency status is reported via a backup communication link. The CCD lens cleanliness adaptive self-cleaning system includes an oil-resistant flexible scraper driven by a stepper motor and an air curtain blowing component, which is used to perform air curtain blowing or scraping cleaning according to the degree of contamination, and triggers lens cleanliness re-evaluation after cleaning. Each actuator is set with control priority in the order of emergency hard interruption protection mechanism, buoyancy-tension safety interlock actuator, CCD lens cleanliness adaptive self-cleaning system, and environmental adaptive control actuator, and forms an execution closed loop through hardware interlock, status feedback, and monitoring and sensing layer verification.

[0019] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0020] 1. By actively decoupling the steel wire in advance, free reset, dual steady-state discrimination of vision and coordinates, and consistency verification of double measurement, the "pulling-reset-static-retesting" process in manual observation is transformed into an automated quality control process. It can identify abnormal states such as steel wire micro-vibration, jamming, instability and near-wall friction from the source of data acquisition, avoid invalid data from entering subsequent analysis, and improve the authenticity and reliability of monitoring data.

[0021] 2. Through multi-dimensional collaborative sensing of displacement, state, clearance and environmental quantities, the coordinates of the steel wire, the immersion depth and sway of the float, the net distance between the steel wire and the protective pipe, as well as information such as temperature, humidity, air pressure, wind speed and wind direction are obtained in real time. This overcomes the problem that the measurement benchmark, equipment body and safety boundary status of the traditional inverted plumb line system are not perceptible, and realizes comprehensive monitoring of the inverted plumb line operation status.

[0022] 3. Through real-time correction of multi-dimensional physical fields and decoupling of dynamic benchmarks, the local gravity field anomalies caused by wire creep, temperature changes, buoyancy fluctuations, buoy sway, and the rise and fall of reservoir water levels or the mass migration of landslide bodies are corrected in layers. This can effectively separate environmental interference, equipment errors and the actual displacement of engineering structures, improve the accuracy of horizontal displacement monitoring, and reduce the risk of misjudgment and missed judgment of abnormal deformation.

[0023] 4. By calculating the minimum net distance between the steel wire and the protective pipe in real time, monitoring the distance of the float hitting the wall, and digital twin mapping of the safety boundary, the spatial positional relationship between the steel wire, the float and the protective pipe can be reflected intuitively. Before the occurrence of near wall, contact, excessive sway or jamming, a graded early warning can be given, and corresponding protective measures can be implemented in conjunction to reduce steel wire wear, equipment damage and monitoring failure.

[0024] 5. Through adaptive observation strategies, buoyancy-tension safety interlocks, environmental control, emergency hard interruption and CCD lens self-cleaning functions, the system can automatically adjust the observation frequency, suspend invalid observations or perform on-site intervention according to working conditions such as flood season, earthquake, sudden load change, environmental over-limit and equipment abnormality. This enables equipment fault self-diagnosis, abnormal state self-protection and adaptive observation strategies, and improves the system's continuous operation capability and unattended operation level.

[0025] 6. By adopting the intelligent transformation method of "prioritizing the reuse of existing equipment and adding external components", the system can be upgraded without changing the original inverted plumb line mechanical structure such as drilling, anchor blocks, protective pipes, steel wires, pontoons and oil drums. This maintains the continuity and integrity of the original gravity benchmark, while reducing the workload of equipment replacement, on-site construction and manual inspection, and lowering the system construction and full life cycle operation and maintenance costs. This makes it easy to promote and apply in projects such as dams, bridges, high-rise buildings and landslides. Attached Figure Description

[0026] Figure 1This invention provides a three-layer closed-loop architecture diagram of an intelligent agent for monitoring structural deformation using a gravity reference and horizontal displacement using an inverted plumb line.

[0027] Figure 2 This is a schematic diagram of the inverted plumb line device provided in an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the vertical non-contact toggle reset mechanism provided in an embodiment of the present invention.

[0029] Figure 4 This is a schematic diagram of the multi-state integrated monitoring structure for floats provided in an embodiment of the present invention;

[0030] Figure 5 This is a cross-sectional structural diagram of the position monitoring inside the limit quantity sensing wire tube provided in an embodiment of the present invention;

[0031] Figure 6 This is a topology diagram of a four-dimensional sensing collaborative network provided in an embodiment of the present invention;

[0032] Figure 7 This is a logical architecture diagram of the internal modules of the cognitive decision-making layer provided in an embodiment of the present invention;

[0033] Figure 8 This is a logic block diagram of multi-mechanism linkage and priority in the control execution layer provided in an embodiment of the present invention. Detailed Implementation

[0034] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0036] Example 1:

[0037] refer to Figures 1-8 This invention relates to an intelligent agent for monitoring structural deformation using a gravity reference and for monitoring horizontal displacement using an inverted plumb line. It employs a three-layer closed-loop architecture of "monitoring and perception - cognitive decision-making - control and execution" (see appendix for details). Figure 1This architecture is driven by high-speed data flow and bidirectional command flow, and its core technologies include artificial intelligence algorithms, physical correction models, and hardware interlocking mechanisms. Through real-time information interaction, timing coordination, and closed-loop iteration within a three-layer architecture, it systematically addresses the inherent shortcomings of traditional plumb line monitoring systems, such as poor data reliability, blind spots in equipment status perception, rigid observation strategies, passive and lagging maintenance, and ineffective elimination of environmental and gravitational field interference. The three-layer architecture features clear division of labor, deep coupling, and closed-loop iteration. The specific layer functions and overall collaborative logic are as follows:

[0038] 1) Monitoring and Sensing Layer: Multi-dimensional holographic sensing to construct a high-precision spatiotemporal data foundation.

[0039] The intelligent sensing layer, acting as the system's "sensory sensing unit," is primarily used to construct a multi-source heterogeneous synchronous acquisition network. This completely resolves industry pain points in existing technologies, such as unclear measurement benchmarks and environmental boundary conditions, difficulty in quantifying and eliminating various interference factors, and lack of monitoring of the equipment's own operating status. This layer abandons the traditional single displacement acquisition mode and integrates four sensing modules: displacement, state, clearance, and environmental quantities. At key nodes of the inverted plumb line device, it forms a four-dimensional sensing network covering the monitoring structure, equipment body, safety boundary, and external environment. Through unified clock synchronization and high-speed bus transmission, it outputs time-aligned, dimensionally complete, and highly reliable raw multi-source sensing data, providing complete and accurate underlying data support for upper-level cognitive decision-making, thus filling the sensing blind spots and data deficiencies of traditional monitoring from the source.

[0040] 2) Cognitive Decision-Making Layer: Cloud-based global cognition enables intelligent decision-making and iterative optimization across the entire chain.

[0041] The cognitive decision-making layer, deployed on a remote cloud platform, serves as the "central decision-making brain" of the intelligent agent. It eliminates the need for on-site edge computing nodes and leverages big data analytics, artificial intelligence algorithms, digital twin modeling, and multi-factor coupling correction models to address the shortcomings of traditional monitoring systems, including lack of process quality control for data, coarse error correction models, rigid and inflexible observation strategies, lack of intelligent diagnosis of structural safety boundaries, black-box equipment health status, and passive and lagging operation and maintenance models. This layer comprises two core modules: the cognitive unit and the decision-making unit. The cognitive unit is responsible for real-time analysis, quality control verification, error decoupling correction, and adaptive observation strategy scheduling of the underlying sensing data. The decision-making unit is responsible for global situation assessment, deformation trend prediction, equipment health diagnosis, baseline drift detection, and predictive operation and maintenance scheduling. Simultaneously, it utilizes long-term historical data to achieve adaptive model evolution, continuously improving monitoring accuracy and decision reliability, completing a closed-loop cloud-based management system from raw data purification and intelligent cognition to global decision-making and iterative optimization.

[0042] 3) Control and execution layer: Precise closed-loop execution, achieving autonomous control and safety protection under all operating conditions.

[0043] The control execution layer is deployed at the monitoring site, serving as the system's "action execution terminal." It accurately receives various instructions from the cognitive decision-making layer, including observation strategies, operating condition control, equipment cleaning, and emergency protection. Simultaneously, it provides real-time feedback on equipment operation status and on-site conditions, forming a physical closed loop of "perception-decision-execution-feedback." This layer integrates four major execution mechanisms: buoyancy-tension safety interlock, environmental adaptive control, emergency hard interruption protection, and CCD lens adaptive self-cleaning. It specifically addresses the problems of traditional inverted plumb line monitoring, such as cumbersome and complex maintenance, poor environmental adaptability, delayed emergency response to abnormal conditions, susceptibility to equipment damage, and frequent invalid data. Through hardware interlocking, priority scheduling, and redundancy backup mechanisms, it ensures long-term stable, accurate, and safe operation of the system under complex and extreme conditions such as high humidity, dust, airflow disturbances, and gravitational field fluctuations.

[0044] 4) Three-layer collaborative closed-loop mechanism

[0045] The monitoring and perception layer, cognitive decision-making layer, and control execution layer do not operate independently. Instead, they achieve bidirectional communication, time synchronization, and closed-loop iteration of data flow, command flow, and status flow through the high-speed bus of the Industrial Internet of Things, forming a continuously optimized closed loop of "monitoring and perception → cognitive decision-making → control execution → re-intelligent perception." The perception layer collects four-dimensional holographic data in real time, completes low-level preprocessing and status identification, and uploads it to the cloud-based cognitive decision-making layer. The cognitive decision-making layer performs data quality control, error correction, situation analysis, and strategy optimization, outputting various commands for regulation, observation, operation and maintenance, and emergency response. The control execution layer accurately executes physical actions and provides feedback on the execution results. The perception layer then collects and verifies the action effects, achieving closed-loop verification. This mechanism enables the system to possess intelligent capabilities such as environmental self-adaptation, equipment self-diagnosis, strategy self-optimization, and fault self-protection, fundamentally solving core problems of traditional systems such as "false data analysis, prominent perception blind spots, passive and lagging operation and maintenance, and inability to eliminate interference," achieving a generational leap in engineering structure deformation monitoring technology.

[0046] I. Monitoring and Sensing Layer

[0047] The monitoring and sensing layer is used to construct a multi-source heterogeneous data acquisition network to address the problems in existing technologies, such as unclear measurement benchmarks and environmental boundary conditions, difficulty in quantifying and eliminating interference factors, and lack of monitoring of the equipment's own operating status. This layer includes a displacement sensing module, a state sensing module, a clearance sensing module, and an environmental sensing module, which are implemented through a plumb line device (see appendix for details). Figure 2 The above modules are integrated and deployed at key nodes to form a four-dimensional three-dimensional perception network covering the monitored object, the sensing entity, the security boundary and the external environment, providing the upper cognitive decision-making layer with time-synchronized raw multi-source data.

[0048] 1. Displacement sensing module (sensing object)

[0049] The displacement sensing module is the core sensing unit of the inverted plumb line intelligent agent. Its core function is to accurately acquire the spatial coordinates and displacement change data of the inverted plumb line steel wire, providing a highly reliable geometric data source for subsequent cognitive decision-making and error correction. This module uses an existing intelligent plumb line coordinate instrument as its core device, employing a reuse and retrofit approach. It requires no replacement of the original equipment and does not alter the original installation layout. The specific implementation method is as follows:

[0050] 1) Hardware Structure: The intelligent vertical coordinate instrument adopts a dual orthogonal optical imaging architecture, mainly composed of a first optical imaging system, a second optical imaging system, a high-resolution linear CCD sensor, and an embedded image processing unit arranged perpendicularly to each other. All components work together to ensure measurement accuracy and real-time performance. The specific structural design is as follows:

[0051] Optical Imaging System: The system adopts a parallel light projection design. The first optical imaging system (along the engineering axis) and the second optical imaging system (perpendicular to the engineering axis) are perpendicular to each other. Their transmitting and receiving ends are symmetrically arranged on opposite sides of the vertical steel wire, which strictly ensures that the parallel beam illuminates the surface of the steel wire perpendicularly. This makes the steel wire form a shadow with clear edges and uniform gray-scale gradient on the CCD sensor target surface, avoiding measurement errors caused by optical path offset and beam tilt from the source. The transmitting end uses a high-stability LED light source, and the light intensity can be adaptively adjusted according to the ambient light intensity to ensure stable shadow imaging quality.

[0052] High-resolution linear CCD sensor: It adopts a 16-bit grayscale acquisition specification, with a pixel size of no more than 5μm, an effective pixel count of no less than 2048 pixels, and a sampling frequency of up to 10Hz. It can meet the requirements of high-precision coordinate calculation and ensure the real-time data acquisition. Its dynamic range is no less than 60dB, which can adapt to the imaging requirements under different light intensity conditions in the observation environment.

[0053] Embedded Image Processing Unit: Integrates a high-performance embedded microprocessor (with a main frequency of not less than 1GHz), with built-in dedicated image preprocessing and coordinate calculation algorithms. It does not rely on external computing resources (such as edge devices other than the cognitive decision unit), and can realize real-time image processing and fast coordinate calculation, ensuring that the data processing delay does not exceed 50ms, thus meeting the core requirements of dynamic monitoring and real-time control of the system.

[0054] 2) Data Acquisition and Calculation Methods: The core function of the intelligent plumb line coordinate instrument is to capture the shadow of the steel wire through optical imaging. After image processing and algorithm calculation, the precise acquisition of the center coordinates and displacement of the steel wire is achieved. The specific process and calculation methods are as follows:

[0055] The first step, shadow imaging acquisition: During operation, the dual orthogonal optical imaging system synchronously emits a parallel beam of light, which is perpendicularly illuminated by a vertical steel wire. The steel wire blocks the beam, forming a clear shadow image on the target surface of the linear CCD sensor. To clarify the imaging rules, the light intensity distribution model is established as follows: Let the ideal light intensity distribution function of the parallel beam be... The diameter of the vertical steel wire is (In this embodiment, the existing ϕ1.5mm stainless steel wire is used, i.e.) Then, the actual light intensity distribution model on the CCD sensor target surface is: In the formula, This represents the actual light intensity distribution on the target surface of the CCD sensor. The ideal physical coordinates of the center of the vertical wire; For a rectangular window function, its value takes the following rules: when hour, This characterizes the complete occlusion effect of the wire shadow on the light beam; in other cases... It represents the light intensity distribution in the unobstructed area.

[0056] The second step is image preprocessing: Due to the complex environment of low light and high humidity in the observation corridor and vertical shaft observation room, images are prone to light decay and water vapor blurring, affecting the accuracy of edge extraction. Therefore, the image processing unit adopts 256-level optical path grayscale image scanning technology to sequentially perform environmental background subtraction, adaptive exposure compensation, and image enhancement algorithms on the acquired shadow images to eliminate environmental interference. The specific compensation formula is as follows: Let the first... The original grayscale matrix of the second sampling is By collecting the average grayscale value of the unobstructed area of ​​the CCD sensor, the estimated grayscale value of the environmental background is obtained. Then the grayscale value after adaptive exposure compensation for: 5; In the formula, This is the dynamic gain adjustment coefficient, ranging from 0.8 to 1.2, used for adaptive matching of different ambient light intensities: when the ambient light intensity is higher than 100 lux, Use a value between 0.8 and 1.0 to avoid overexposure of the image; when the ambient light intensity is below 50 lux, Automatically adjusts to 1.0~1.2, effectively improving image contrast and ensuring that shadow edges are clearly distinguishable.

[0057] The third step, sub-pixel edge extraction and center coordinate calculation: To overcome the precision limitations of physical pixels and achieve a spatial resolution better than 0.01mm, a sub-pixel edge extraction algorithm is used to locate the edges of the preprocessed grayscale image. First, the first-order spatial gradient of the grayscale image is calculated along the vertical radial direction, and a gradient vector is constructed. : The approximate location of the shadow edge is determined by using the gradient maxima, and then a 3×3 window Gaussian interpolation function is used to fit a curve to the neighborhood of the gradient maxima to accurately determine the sub-pixel edge coordinates. The coordinates of the two edges of the shadow in the horizontal direction (along the engineering axis) are: The coordinates of the two edges of the vertical (horizontal) shaded area are given. Based on these edge coordinates, the center coordinates of the vertical wire in the two-dimensional plane are calculated. The calculation formula is as follows: , .

[0058] Step 4, Displacement Calculation: Based on the initial reference state (coordinates collected after system installation and commissioning, without environmental interference, and during stable equipment operation) or the center coordinates of the plumb line measured in the previous observation period. Based on this, the center coordinates measured in the current observation period are: The horizontal displacement of the steel wire The calculation model is as follows: , In the formula, The horizontal displacement component is along the engineering axis (core monitoring direction); The displacement component is perpendicular to the engineering axis (lateral direction). The displacement unit is mm, and three decimal places are retained to ensure that the data accuracy meets the engineering monitoring requirements.

[0059] 3) Technical Effects: This displacement sensing module utilizes an existing intelligent vertical coordinate instrument. Through the synergistic effect of a dual orthogonal optical imaging system, adaptive image preprocessing, and sub-pixel resolution algorithms, it achieves high-precision, high-real-time steel wire displacement monitoring. Specific technical effects are as follows:

[0060] (1) Strong anti-interference capability: Through adaptive exposure compensation and image enhancement algorithms, it effectively overcomes interference such as light decay and water vapor blur caused by low light and high humidity in the observation environment, ensuring clear shadow imaging and guaranteeing the reliability of data acquisition; (2) High monitoring accuracy: It adopts a 5μm pixel size linear CCD sensor and sub-pixel edge extraction algorithm to achieve a spatial resolution better than 0.01mm, completely eliminating parallax error caused by manual reading, and meeting the core requirements of high-precision displacement monitoring of major engineering structures such as dams; (3) Excellent real-time performance: The embedded image processing unit realizes real-time image processing and coordinate calculation, with a data processing delay of no more than 50ms and a sampling frequency of up to 10Hz, which can capture the instantaneous displacement change of the steel wire and meet the dynamic monitoring and emergency response requirements of the system; (4) Good compatibility: It adopts the old-use transformation mode, without replacing the original intelligent vertical coordinate instrument, without changing the original installation layout, and directly connects to the cognitive decision unit to realize data transmission and control, reducing transformation costs and seamlessly adapting to the overall system architecture.

[0061] 2. State Variable Perception Module (Perception Subject)

[0062] The status sensing module is used to monitor the mechanical and mechanical working status of the plumb line system in real time, promptly detect equipment abnormalities, avoid invalid data collection, and ensure stable system operation. It includes the following five sub-units:

[0063] 1) Non-contact vertical wire repositioning and stability control unit: This unit is used to eliminate unsteady-state data caused by foreign object obstruction, oil film adhesion, or fatigue of the repositioning mechanism, ensuring that the collected data is the true displacement of the wire under steady-state conditions (see appendix for details). Figure 3 The specific implementation method is as follows:

[0064] (1) Hardware structure composition: including flexible electromagnetic drive mechanism, torque detection sensor, precision linear guide rail and control module. The flexible electromagnetic drive mechanism is fixed on the precision linear guide rail and is in the same measurement coordinate system as the intelligent vertical coordinate instrument to ensure that the turning direction is consistent with the monitoring direction; the torque detection sensor is connected in series in the transmission link, and the measurement accuracy is not less than 0.01 N·m, which is used to provide real-time feedback of the turning torque; the control module adopts a PLC controller, which is electrically connected to the torque detection sensor, flexible electromagnetic drive mechanism and intelligent vertical coordinate instrument respectively to realize action coordination and signal interaction.

[0065] (2) Pre-disturbance and stability determination method: 30 minutes before the start of automated data acquisition, the control module is configured to execute the following pre-processing logic to ensure that the steel wire is in an absolute steady state:

[0066] ① Pre-emptive flexible disturbance: 30 minutes before the start of the automated data acquisition task, control the flexible electromagnetic drive mechanism to operate at a preset stroke (0.5~1.0mm) and initial torque. (Preset value: 0.5~1.0 N·m) Apply a non-contact mechanical pulling force to the vertical steel wire to avoid damage caused by direct contact. Let the preset initial pulling torque be... The output excitation current of the electromagnetic driver satisfy: In the formula, This is the torque constant of the electromagnetic mechanism, specified by the manufacturer, with a value ranging from 0.1 to 0.5 N·m / A.

[0067] ② Precise reset: The flexible electromagnetic drive mechanism is controlled to remove the pulling force, so that the vertical steel wire can be freely reset under the action of buoyancy and damping of the float. The swing state of the steel wire is monitored in real time during the reset process.

[0068] ③ Static Discrimination: The coordinate measuring machine acquires the coordinate sequence in real time at a sampling frequency of 1Hz, calculates the oscillation amplitude, and determines that the vertical line has returned to a stable state when the oscillation amplitude decays to within a preset stability threshold. Let the real-time measured coordinate fluctuation range be... : In the formula, The reset start time, To determine the time window (default is 10 seconds). Let be the coordinates of the center of the steel wire at time t. When satisfying... ( When the preset stability threshold is 0.02 mm, a stability confirmation signal is output. If the conditions are not met, the judgment time window is extended until the conditions are met or an anomaly label is generated.

[0069] ④ Data Acquisition Trigger: Once the vertical line is determined to have returned to a stable state, the control module immediately triggers the automated data acquisition process; the coordinate measuring machine is then continuously acquiring data. Group( The mean of 10 sets of coordinate data (pre-set) is used as the effective observation value to reduce random error. Assuming continuous data acquisition... The set of coordinate data sequence is Its mean The calculation formula is: .

[0070] (3) Safety interlocking and yielding protection mechanism: In the preliminary "flexible disturbance" step, the real-time feedback value of the torque detection sensor is set to be When the feedback value exceeds the preset safety threshold When the value is 1.5 N·m (preset), the yield protection logic is triggered and the action is forcibly stopped, that is: Meanwhile, the control module incorporates timing mutual exclusion logic, defining the time state function of the coordinate instrument's effective sampling period as follows: (1 during sampling, 0 during non-sampling), the trigger signal for the flexible electromagnetic drive mechanism is... (1 when in action, 0 when at rest). When the order satisfies strict mutual exclusion constraints, its logical relationship is defined as: Ensure that the pre-processing non-contact mechanical disturbance and the formal automated data acquisition are strictly mutually exclusive on the time axis to avoid the disturbance process affecting the accuracy of data acquisition.

[0071] 2) CCD Lens Surface Cleanliness Online Monitoring Unit: This unit is used to monitor the cleanliness status of the optical lens of the intelligent vertical coordinate instrument in real time, avoiding image distortion and displacement measurement errors caused by lens contamination. The specific implementation method is as follows:

[0072] (1) Structure and Principle: It includes a built-in target and a control unit integrated on the outside of the optical system window. The built-in target is a high-contrast checkerboard target with a checkerboard size of 1mm×1mm to ensure the accuracy of image sharpness assessment. The control unit reuses the image processing unit, eliminating the need for additional hardware costs, and executes the monitoring logic of "online monitoring → status assessment → safety interlock" to monitor the cleanliness of the optical lens surface of the intelligent vertical coordinate instrument in real time. This unit does not contain mechanical cleaning execution components; it only achieves quantitative assessment of the health status of the optical sensing channel by analyzing the target image.

[0073] (2) State assessment and interlocking logic:

[0074] ① Quantitative Monitoring: During non-effective sampling periods (i.e., the intervals when the coordinate measuring machine is not acquiring displacement data), the image acquisition system is controlled to capture images of the built-in target. Based on an image sharpness evaluation algorithm, a lens surface cleanliness assessment value is quantitatively generated. Let the grayscale matrix of the target image be... The Tenengrad gradient function is used to calculate the gradient energy value of the image. As a cleanliness assessment indicator, its calculation formula is: In the formula, and These are the first-order gradients extracted from the image in the horizontal and vertical directions using the Sobel operator, respectively. The larger the gradient energy value, the higher the lens cleanliness and the clearer the image.

[0075] ② Data Acquisition Lockout and Early Warning: When the lens surface cleanliness assessment value is lower than the preset health threshold, the condition is met. ( The cleanliness threshold that triggers optical degradation is determined by calibration (usually 60% of the gradient energy value in a healthy state). This indicates that there are stains on the lens surface. The control unit generates an "optical degradation warning" signal and outputs an "acquisition lock signal" to the displacement sensing module to stop the acquisition of subsequent displacement data and avoid image degradation and data distortion caused by lens stains.

[0076] ③ Continuous Status Tracking: After generating the "Optical Deterioration Warning" signal, the control unit continuously acquires and evaluates target images according to a preset cycle (5 minutes / time), tracking the changes in lens surface cleanliness in real time until the evaluation value recovers to above the health threshold, thus satisfying the following conditions: At this point, the acquisition lockout is released, and the normal displacement acquisition process is resumed.

[0077] (3) Safety interlocking mechanism: When the "acquisition lock signal" is output, the signal simultaneously interlocks the cognitive decision-making layer and the control execution layer, forcibly locking all other mechanical actions such as the flexible electromagnetic drive mechanism, to prevent meaningless equipment wear or mechanical interference under invalid observation conditions; at the same time, the real-time lens surface cleanliness assessment value will be displayed. The data is written into the data packet and bound to the "Optical Degradation Warning" event label, so that the cognitive decision-making layer can reduce the confidence weight of the displacement data in this period and adjacent periods or directly remove it.

[0078] (4) Technical effect: The optical health status of the equipment is transformed from the traditional "black box" state into a quantifiable monitoring indicator, and the distortion data caused by dirt on the lens surface is blocked from being transmitted to the back end in real time from the data source; it provides accurate quantitative basis for the cognitive decision-making layer to judge the validity of the data and initiate manual cleaning and maintenance work orders, significantly improving the data reliability and self-sensing ability of the optical measurement system in complex and humid environments, and reducing the frequency of manual inspection.

[0079] 3) Real-time monitoring unit for eddy current type float yaw and wall-collision distance: This unit is used to monitor the horizontal yaw state of the float inside the oil drum and its micro-distance from the drum wall in real time. It is particularly sensitive to the impact of local gravity field anomalies caused by large fluctuations in reservoir water level on the float yaw, avoiding uneven stress on the steel wire and displacement measurement distortion caused by float yaw (including yaw induced by gravity field anomalies) and wall collisions. The specific implementation method is as follows:

[0080] (1) Structure and principle:

[0081] The system includes four eddy current probes fixed to a reference support on the outer perimeter of the oil drum in a 90° orthogonal symmetrical arrangement, along with a vector calculation module. Each probe incorporates a high-stability oscillation circuit with an oscillation frequency of 1MHz. By detecting real-time changes in the probe coil impedance, it calculates the gap distance between the probe end face and the outer wall of the oil drum, with a measurement range of 0-5mm and a resolution of 1μm. The vector calculation module receives four channels of gap data synchronously output from the four probes. Based on an orthogonal differential algorithm, it calculates the difference in gap values ​​between pairs of relative probes (i.e., 0° and 180° directions, 90° and 270° directions) to eliminate common-mode errors caused by radial thermal expansion and contraction due to temperature changes in the oil drum. Simultaneously, it captures the minute sway signal of the float caused by local gravitational field anomalies, accurately calculating the two-dimensional sway displacement vector (ΔX, ΔY) of the float-driven oil drum in the horizontal plane. This vector includes the sway component induced by the float's own force and the sway component induced by local gravitational field anomalies, comprehensively reflecting the true sway state of the float.

[0082] Let the real-time gap measurement values ​​of the probes at 0°, 90°, 180°, and 270° be respectively... , , , After eliminating the common-mode error and the influence of gravitational field interference, the two-dimensional yaw displacement components are: , Let the standard clearance value of the oil drum in its initial centered state be D0 (calibrated during installation, in mm). Combining this with the aforementioned yaw displacement, which includes components induced by gravitational anomalies, the minimum micro-distance between the float and the oil drum wall in the horizontal plane is calculated in real time. Its calculation model is as follows: By monitoring the changes in d_min in real time, when the anti-collision safety constraint is met, it is necessary to ensure that the real-time minimum distance is greater than the preset alarm threshold, that is: In the formula, The critical safety distance threshold affecting the accuracy of vertical observation is preset to 0.5mm. When d_min≤d_alarm, a near-wall alarm signal for the pontoon is generated. At the same time, the sway data is marked to distinguish between sway induced by gravity field anomalies and sway due to pontoon faults. This provides a clear basis for subsequent displacement data correction and avoids misjudging sway induced by gravity field anomalies as actual structural deformation.

[0083] (2) Technical effect: The displacement resolution reaches the micrometer level, which can accurately capture the horizontal micro-movement of the float (including the micro-sway induced by local gravity field anomalies) and output the micro-distance with the barrel wall in real time, effectively identifying the sway interference caused by gravity field anomalies; the probe is encapsulated with oil-resistant special polymer material, and the measurement principle is based on electromagnetic induction. It is not sensitive to oil, sludge and changes in the dielectric constant of the medium, realizing non-contact and wear-free measurement, and the service life is not less than 5 years; through the 90° orthogonal symmetrical layout combined with differential calculation, not only is the complete calculation of two-dimensional offset realized, but the measurement interference caused by the deformation of the oil barrel body and local gravity field anomalies is also effectively eliminated, avoiding the superposition of the float sway induced by gravity field anomalies into the displacement monitoring value and misjudging it as the actual deformation of the structure. At the same time, it avoids the adverse effect of the float touching the wall on the vertical observation accuracy, significantly improving the measurement robustness and system reliability under complex liquid environment and gravity field anomaly conditions, and providing a highly reliable float sway data source for the accurate correction of subsequent displacement data.

[0084] 4) Real-time monitoring unit for changes in the immersion depth of the magnetostrictive buoy: This unit is used to monitor the immersion depth of the buoy in real time, indirectly reflecting changes in wire tension, and avoiding wire bending and vibration due to insufficient buoyancy, which would affect measurement accuracy (see appendix for details). Figure 4 The specific implementation method is as follows:

[0085] (1) Structure and Principle: The system includes a waveguide wire, a float magnetic ring that floats synchronously with the float's liquid level, and a pulse detection circuit. The waveguide wire is made of 0.5mm diameter stainless steel wire and is fixed to a fixed support structure connected to the oil drum or external reference bracket. The float magnetic ring is rigidly connected to the float and rises and falls synchronously with it. The pulse detection circuit is integrated into the control module, with a sampling frequency of 1Hz. Based on the magnetostrictive effect, the pulse circuit generates a current pulse that propagates along the waveguide wire, generating a torsional wave at the float magnetic ring. By detecting the propagation time difference between the current pulse and the torsional wave, the absolute position of the float magnetic ring is calculated, thereby monitoring the vertical change in the float's immersion depth in real time. Let the emission time of the current pulse be... The pulse detection circuit receives the torsional wave at the moment when The speed of sound wave propagation in the waveguide wire is (The preset speed is 5800 m / s, which can be corrected through calibration), then the absolute position change of the float magnetic ring. The solution formula is: Based on the buoy immersion depth With absolute position change The geometric mapping relationship is used to calculate the current immersion depth in real time: In the formula, This represents the immersion depth of the pontoon when it is in its initial reference state, calibrated during installation, and expressed in mm. The control module calculates this in real time. Monitor the status of the floats. When the safety constraint that the floats must not protrude above the liquid surface is met, ensure that the real-time immersion depth is always greater than zero. ,when At that time, a buoyancy decay alarm signal is generated and uploaded to the cognitive decision-making layer.

[0086] (2) Technical effect: The time domain reflectance measurement technology is used to realize real-time monitoring of vertical changes in immersion depth with a resolution of 1μm; the whole machine has an IP68 protection level, and the key components are made of oil-resistant polymer materials, which can be immersed in transformer oil for a long time and adapt to humid and oily environments; the measurement results are not directly affected by changes in oil viscosity, fluctuations in medium density or temperature gradient, and the measurement error does not exceed ±0.1mm, which provides a reliable guarantee for the safety of the float liquid level of the inverted plumb line system and indirectly ensures the stability of the steel wire tension.

[0087] 5) Visual monitoring unit for the working status of the intelligent vertical coordinate instrument

[0088] This unit is used for all-weather visual monitoring of the working status of the intelligent vertical coordinate instrument, capturing abnormal equipment actions and external disturbances, and providing visual evidence for anomaly tracing. The specific implementation method is as follows:

[0089] (1) Structure and principle: This unit mainly consists of a low-light CMOS image sensor, an infrared fill light, and an edge preprocessing module. In order to adapt to the low-light environment inside the observation environment and accurately capture transient anomalies during the operation of the coordinate instrument, the low-light CMOS image sensor used has excellent low-light imaging performance: the minimum illumination is 0.1 lux, the maximum frame rate is not less than 120fps, and the pixel resolution is not less than 1080P; the infrared fill light uses the 850nm infrared band to avoid interference with the measurement optical path of the optical coordinate instrument, and can maintain high time resolution in extremely dark environments to avoid missing abnormal events.

[0090] The edge preprocessing module incorporates ROI (Region of Interest) extraction and moving target detection algorithms to perform focused analysis on the core working area of ​​the coordinate instrument (optical lens, wire observation area), reducing redundant data processing. Let the image matrix of the coordinate instrument's working area after ROI extraction be... ,in These are spatial pixel coordinates. If the time frame number is used, then its inter-frame difference model on the time axis is defined as: In the formula, This represents the inter-frame time interval, with a value of 1 / 120s (corresponding to a frame rate of 120fps). An adaptive dynamic background update strategy and a set moving target binarization threshold are used. (Values ​​range from 20 to 50, adaptively adjusted according to ambient brightness), the inter-frame difference results are evaluated, and feature regions of moving targets that can characterize sudden changes in the coordinate instrument's working state, external disturbances, or equipment malfunctions are extracted. : It enables precise location and feature extraction of abnormal changes in the coordinate measuring machine's working area.

[0091] (2) Linkage control method: The low-light high-speed vision unit establishes a linkage triggering mechanism with the intelligent vertical coordinate instrument body and the automatic reset unit to realize the full-process visual closed-loop monitoring of the coordinate instrument's working status. Under normal working conditions without abnormalities, the system collects, compresses and stores video streams in a loop according to the preset time interval (10min / time) and frame rate (10fps), saving storage resources; when the monitoring system receives an abnormal alarm signal of the coordinate instrument measurement data, or when the automatic reset unit executes an action trigger signal, it immediately switches to high-speed recording mode (frame rate 120fps) and automatically saves the complete video frames of the key time periods before and after the abnormal event, forming a traceable visual evidence chain.

[0092] Let the time of the exception or action be... The system's preset forward cache duration is (Default is 10s), the duration of subsequent continuous recording is (Default is 30s), then the video frame sequence time window of the evidence chain is finally stored in the storage unit for fault tracing and anomaly diagnosis. for: Through this linked storage and time window locking mechanism, complete, continuous and highly reliable visual evidence can be provided for the anomaly diagnosis, fault location, external disturbance analysis and operation status verification of the intelligent vertical coordinate instrument.

[0093] (3) Technical effects: It breaks through the limitation of "blind testing" of traditional vertical line monitoring equipment, transforming the black box operation status of optical coordinate instruments into concrete and traceable visual information; the low illumination of 0.1 lux combined with infrared supplementary lighting design ensures the quality of image acquisition while significantly reducing the system's dependence on high-end special image sensors, achieving excellent cost performance; the edge preprocessing algorithm based on ROI and inter-frame difference effectively reduces the amount of redundant video data while ensuring the accuracy of anomaly identification, alleviating the pressure of storage and transmission; through the linkage mechanism of "data / action anomaly trigger - time window locking recording", a complete diagnostic closed loop from the bottom sensor data to the top visual evidence is constructed, completely eliminating the controversy over the cause of anomalies caused by the lack of on-site process records, and significantly improving the maintainability, operation and maintenance decision efficiency and overall operational reliability of the intelligent vertical line system.

[0094] 3. Limit quantity sensing module (sensing safe distance)

[0095] The clearance sensing module is used to monitor the position of the vertical stainless steel wire inside the seamless steel pipe in real time, to obtain data on the distance change between the wire and the protective tube (hole wall), and to provide timely warnings of the risk of the wire approaching the wall or contacting the hole wall, thus avoiding displacement measurement distortion and wire damage caused by friction between the wire and the hole wall (see appendix for details). Figure 5 The specific implementation method of the four-dimensional capacitive steel wire tube position and clearance sensing unit is as follows:

[0096] 1) Hardware Structure: Includes four capacitance probes, a multi-channel capacitance detection circuit, and a spatial calculation module. For seamless steel pipe protection tubes with an outer diameter of 168mm and a wall thickness of 8mm (inner diameter of 152mm), commonly used in engineering, the four capacitance probes are orthogonally and symmetrically embedded in pre-set mounting holes on the inner wall of the key cross-section of the protection tube at 90° angles. The probe sensing surface is flush with or slightly protrudes into the inner wall of the protection tube (protrusion not exceeding 0.5mm) to avoid interference with the movement of the steel wire. The arc length distance between adjacent probes is approximately 120mm to avoid edge electric field crosstalk and ensure measurement independence. The probes use a stainless steel substrate, with a 1-2mm thick corrosion-resistant, highly insulating potting layer (such as PTFE or special epoxy resin) covering the side facing the steel wire. The insulation resistance is not less than 100MΩ. Electrical isolation is achieved between the probes and the metal body of the protection tube through the insulating layer. The overall protection level reaches IP68, suitable for humid and dusty environments inside holes.

[0097] 2) Data Acquisition and Calculation Method: The sensing electrodes of the four capacitance probes and the vertical stainless steel wire respectively constitute the two poles of the capacitance sensor, forming four independent capacitance measurement channels. When the steel wire is displaced relative to the seamless steel pipe, the change in the electrode spacing causes a corresponding change in the capacitance value of each probe. The multi-channel capacitance detection circuit synchronously acquires the four capacitance values ​​at a sampling frequency of 10Hz, with a measurement accuracy of not less than 0.1pF. Based on the pre-built "capacitance-electrode distance" calibration curve (calibrated during installation using a standard displacement stage), it converts the values ​​into four independent directional clearance data. The spatial calculation module receives the above four clearance data to construct an overdetermined set of equations and uses a least-squares circle fitting algorithm to calculate the two-dimensional center coordinates of the vertical stainless steel wire within the cross-section of the protective pipe. The minimum net distance between the steel wire and the pipe wall is calculated in real time. Let the theoretical radius of the inner wall of the protective pipe be... (For steel pipes with an inner diameter of 152mm, The distance from the center of the steel wire to the pipe wall obtained by solving the problem is the real-time net distance. ,satisfy: In the formula, For the first The center coordinates of each capacitive probe ( The minimum net distance is calculated by traversing the four directions. : When individual probes experience data jumps due to localized water film adhesion or contamination, the algorithm incorporates residual analysis, assuming the first... Measurement residuals of each probe for: When satisfied ( When the anomaly rejection threshold is set to 0.05mm, the anomaly point is automatically rejected, and the remaining effective net distance data is used to re-fit the center of the circle to ensure the continuity and accuracy of the wire position monitoring.

[0098] 3) Multi-level early warning and linkage mechanism: The spatial calculation module has a pre-stored first-level near-wall early warning threshold. (Preset to 2mm) and the secondary contact alarm threshold (Default is 0.5mm). When the minimum net distance value calculated in real time meets the requirements... When the minimum clearance value is met, a near-wall warning signal is generated; when the minimum clearance value is met... When the net distance change rate suddenly drops to zero (indicating physical contact), a contact alarm signal is generated. The near-wall warning signal is configured to be directly output to the vertical line automatic reset and data quality pre-control unit in the state quantity sensing module as a pre-linkage condition for triggering flexible towing intervention, actively eliminating the near-wall hazard of the steel wire; the contact alarm signal is uploaded to the cognitive decision layer to trigger emergency protection instructions.

[0099] 4) Technical Effects: This module boasts a micron-level net distance resolution, enabling precise real-time monitoring of the dynamic position of the stainless steel wire within the seamless steel pipe, with a measurement error not exceeding ±0.01mm. The 90° orthogonal layout ensures excellent symmetry and independence in the net distance calculations along the X and Y axes, guaranteeing the accuracy of the two-dimensional position calculations. Based on a 1-2mm thin insulation layer design, it maintains high sensitivity while ensuring pressure resistance and anti-creep performance in humid borehole environments. More importantly, the redundant observation system consisting of four probes, combined with a robust fitting algorithm, effectively overcomes data distortion caused by single-point local anomalies, achieving highly fault-tolerant and robust position monitoring in complex humid environments. Through the hard linkage between the early warning signal and the underlying mechanical execution unit, a physical closed loop is achieved, from safe distance monitoring to active anti-jamming intervention, avoiding damage and measurement errors caused by contact between the steel wire and the borehole wall.

[0100] 4. Environmental Sensing Module (Sensing the Environment)

[0101] The environmental sensing module is used to acquire multi-dimensional environmental element parameters of the working environment of the inverted plumb line agent, providing an input source for error correction of monitoring data and judgment of observation conditions, and eliminating the influence of environmental factors on measurement accuracy. The specific implementation method is as follows:

[0102] 1) Hardware Structure: The high-precision 5-element weather station integrates a temperature sensor, humidity sensor, barometric pressure sensor, ultrasonic anemometer, and ultrasonic wind direction sensor within its housing. The station boasts an overall IP66 protection rating and an operating temperature range of -40℃ to +70℃, adapting to extreme environments. The ultrasonic anemometer employs a solid-state design with no mechanical rotating parts, minimizing mechanical wear and ensuring a service life of at least 5 years. All sensors utilize digital output interfaces, directly connecting to the data acquisition bus to reduce signal interference.

[0103] 2) Data Acquisition and Processing Methods: Temperature, humidity, and air pressure sensors utilize MEMS technology to convert physical quantities into digital signals, with measurement accuracies of ±0.2℃ for temperature, ±3%RH for humidity, and ±0.1%FS for air pressure. The ultrasonic anemometer uses the ultrasonic time-of-flight method, calculating wind speed by measuring the time difference of sound wave propagation in orthogonal directions, with no starting wind speed limitation. Let the speed of sound in still air be... (Corrected by real-time temperature and air pressure parameters) , (Ambient temperature, in °C) The time differences of relative propagation in the orthogonal X-axis and Y-axis directions are respectively and The effective acoustic path between the ultrasonic probes is (Preset to 100mm), then the two-dimensional wind speed components The solution model is as follows: , In the formula, The round-trip propagation time of the ultrasonic wave in the X-axis direction is denoted as . The round-trip propagation time of the ultrasonic wave in the Y-axis direction. The synthesized instantaneous wind speed. with wind angle The expression is: , This high-precision 5-element weather station achieves high-precision synchronous acquisition of temperature, humidity, air pressure, wind speed, and wind direction, with a sampling frequency of 1Hz and a data update delay of no more than 100ms.

[0104] 3) Operating condition identification and linkage control method: The output data of the environmental quantity sensing module participates in the following system-level linkage logic:

[0105] ① Environmental effect parameter output: The real-time collected temperature and air pressure data are packaged into environmental compensation parameters and sent to the displacement sensing module and the state sensing module through the data bus. These parameters serve as input variables for the nonlinear error correction model of the original measurement data, and are used to eliminate measurement errors caused by temperature and air pressure changes.

[0106] ② Wind-induced disturbance interlock: Real-time calculation of instantaneous and continuous wind speeds, with a set time window. The moving average wind speed within the preset 10 seconds is The calculation formula is as follows: In the formula, The number of sampling points within the window ( ), For the first The instantaneous wind speed value at any given moment. When the instantaneous wind speed... (e.g., 1.0 m / s) or continuous average wind speed When a disturbance threshold condition of 0.5 m / s is reached, a "collection lockout signal" is generated and output to the control system to forcibly suspend the automated observation cycle. At the same time, a "disturbed state" label is added to the data packet for the current period, prohibiting the use of the data for that period as a valid reference value, thus avoiding the entry of unsteady data caused by wind-induced disturbances into the database.

[0107] 4) Technical Effects: Through the integrated high-precision 5-element meteorological station, the working environment status of the inverted plumb line intelligent body is fully perceived, significantly reducing the complexity of on-site wiring and installation, and realizing the synchronous acquisition and unified output of environmental parameters; based on the high-precision environmental parameter output and the sliding average wind speed threshold blocking mechanism, the interference of gusts and spikes is effectively filtered out, and the non-steady-state data caused by airflow excitation is blocked from entering the database from the physical source, providing accurate quantitative basis for the decoupling analysis of engineering structure deformation and environmental factors, and improving the accuracy and reliability of displacement measurement.

[0108] 5. Four-dimensional perception collaborative mechanism

[0109] The four sensing modules—displacement, state, limit, and environmental—are deeply interconnected via a high-speed data bus (using EtherCAT bus with a transmission rate of no less than 100Mbps) and a unified clock synchronization mechanism. This enables cross-module cross-verification and dynamic correction logic, breaking the limitations of traditional "island-style" sensor operation (see appendix for details). Figure 6 The specific collaboration mechanism is as follows:

[0110] 1) Clock synchronization and data bus architecture: Four sensing modules are connected to the same industrial bus and data is sampled based on a unified external clock synchronization signal (such as PPS second pulse). The clock synchronization accuracy does not exceed 1ms, ensuring that the displacement coordinates, float status, safety clearance, meteorological parameters and visual evidence output by each module are strictly aligned in timestamps, providing a spatiotemporal reference for subsequent data fusion of the cognitive decision layer and avoiding errors caused by time deviation.

[0111] 2) Cross-module cross-validation and correction logic

[0112] (1) Environmental parameter coupling correction: The temperature and air pressure parameters output by the environmental quantity sensing module (high-precision 5-element meteorological station) are used as independent variables and injected into the displacement quantity sensing module (coordinate calculation model) and the state quantity sensing module (buoyancy-tension calculation model) in real time to dynamically compensate the original readings of geometric and mechanical quantities, eliminate common mode errors caused by environmental drift, and improve measurement accuracy.

[0113] (2) Wind speed threshold condition interlock: The "acquisition interlock signal" generated by the environmental quantity sensing module based on the moving average wind speed algorithm overwrites the sampling enable terminal of the displacement sensing module through the bus. When the wind speed exceeds the limit, the generation path of displacement data is cut off in time sequence to avoid invalid data caused by wind-induced disturbance.

[0114] (3) Equipment status confidence identifier: The status quantity sensing module evaluates buoyancy fluctuation and reset deviation in real time. When the equipment is determined to be in an abnormal state (such as buoyancy decay or reset failure), an "abnormal status identifier" is generated. This identifier is bound and packaged with the displacement data at the same time in the bus layer, and the confidence weight of the displacement data of this frame is marked to the cognitive decision layer (the weight is set to 0.3 in the abnormal state and 1.0 in the normal state), which facilitates the upper layer algorithm to perform data quality control.

[0115] (4) Safety boundary early warning and intervention linkage: After the limit quantity sensing module calculates the minimum clearance value, it executes dual logic output: When the clearance is lower than the first-level early warning threshold, it outputs a "near wall early warning trigger signal" to the vertical automatic reset and data quality closed-loop control unit in the state quantity sensing module, actively starts flexible toggle intervention, and eliminates near wall hazards; when the clearance is lower than the second-level alarm threshold, it generates a "boundary contact mark" and binds it to the displacement data packet, so that the upper-level algorithm can eliminate false displacement steps caused by hole wall friction.

[0116] (5) Optical channel health coordination and data blocking: The online monitoring unit for the surface cleanliness of the CCD lens in the state quantity sensing module quantifies the lens cleanliness assessment value in real time. When the assessment value is lower than the health threshold, it outputs a "collection lock signal" to the displacement quantity sensing module to forcibly stop the displacement data acquisition, locks the underlying mechanical action, and binds the "optical degradation warning" event label with the real-time assessment value to the data packet, so that the cognitive decision layer can reduce the confidence weight of the displacement data in adjacent time periods or directly remove it.

[0117] (6) Visual evidence chain binding for abnormal working status: When the above-mentioned abnormal events such as "abnormal status indicator", "boundary contact indicator" or "optical degradation warning" are triggered, or when the reset unit action signal is received, the visual monitoring unit of the intelligent vertical coordinate instrument in the status quantity sensing module automatically locks the time window before and after the trigger. High-speed video recording is performed. The generated visual video evidence stream is strongly bound to the aforementioned multi-dimensional perception data package through a unique event ID, forming a composite evidence chain of "data anomaly alarm + underlying operating condition label + high frame rate video slice", providing complete evidence for anomaly tracing.

[0118] 3) Collaborative data packet construction and output: After the above cross-correction, interlocking control and identification binding, the underlying data streams of the four modules converge at the acquisition terminal. Valid sampled data that was not executed due to working condition interlocking is removed, and finally a structured multidimensional perception data packet containing "high-precision coordinates + equipment health tags + boundary net distance vector + environmental original parameters + visual evidence index" is constructed. The data format adopts JSON format and is uniformly imported into the cognitive decision layer to ensure the integrity, standardization and traceability of the data.

[0119] 4) Technical Effects: This collaborative mechanism breaks through the limitations of the "island-like" independent operation of each sensor in traditional monitoring. Through underlying hardware interlocking and data identification binding, environmental interference, equipment failure, boundary anomalies, and optical degradation are physically isolated or downweighted at the data source, avoiding the backward transmission of invalid errors to the cloud decision-making layer. It innovatively introduces visual perception as the final arbiter of abnormal operating conditions, making abstract data tags concrete into traceable video evidence, completely eliminating disputes over the causes of anomalies, and achieving a systematic leap in the overall measurement accuracy, environmental adaptability, and system maintainability of the sensing network.

[0120] II. Cognitive Decision-Making Level

[0121] The cognitive decision-making layer adopts a real-time cognition + global decision-making collaborative architecture (see Figure 7), and is centrally deployed on a remote cloud platform. The platform has two main functional modules: real-time cognition and global decision-making. These modules work collaboratively and in a closed-loop manner, relying on the Industrial Internet of Things (IIoT) and field terminals to achieve two-way data interaction, jointly constructing the core intelligent hub of the system. The input end of the cognitive decision-making layer connects to the monitoring and perception layer, uniformly receiving structured multi-dimensional sensing data output from the four-dimensional sensing collaborative mechanism; the control command output end links the control execution layer and various controlled mechanisms in the monitoring and perception layer, comprehensively completing the entire chain of management and control, including real-time data quality control, online error correction, cloud-based global diagnosis, situational analysis, and predictive maintenance. This effectively addresses the core shortcomings of traditional inverted plumb line monitoring systems in terms of data reliability, field status perception, maintenance system, and safety diagnostic capabilities. The specific composition and execution logic are as follows:

[0122] 1. Cognitive Unit

[0123] The cognitive unit has a built-in intelligent algorithm library for vertical line observation, deployed on a remote cloud platform. It collaborates with the decision-making unit to perform real-time analysis, quality control, and policy distribution of perceived data. This enables rapid processing and immediate intervention of on-site data, avoiding invalid data uploads and delayed responses. As the core of on-site perception analysis in the cognitive decision-making layer, it provides high-quality, highly reliable data support for subsequent decision-making units. Specifically, it includes the following modules:

[0124] 1) Wire pre-decoupling and dual-return intelligent acquisition module: This module works in deep collaboration with the automatic vertical line reset and data quality closed-loop control unit and the low-light high-speed visual monitoring unit in the state quantity sensing module. It proactively eliminates potential vertical line instability before automated data acquisition, achieving fully automated closed-loop control from decoupling, static setting, steady-state discrimination to dual-return self-consistent verification. Its specific execution logic is as follows:

[0125] Pre-emptive active decoupling: 30 minutes before the start of the preset automated data acquisition cycle, a pre-emptive decoupling trigger signal is output to the vertical line automatic reset and data quality closed-loop control unit to control the flexible electromagnetic drive mechanism to apply mechanical pulling force to the vertical line steel wire, so as to actively destroy and eliminate the non-free steady-state illusion caused by foreign object jamming, oil film adhesion or slowly changing environmental forces.

[0126] Free reset and long-term static setting: Remove the mechanical force to allow the vertical steel wire to reset freely, and then enter a forced static setting period of no less than 30 minutes to ensure that the steel wire fully attenuates the residual mechanical kinetic energy in the liquid damping environment and tends to the true physical equilibrium state.

[0127] Multi-source fusion steady-state discrimination: Before and after the 30-minute settling period, visual and coordinate dual feature extraction are simultaneously initiated: based on the edge computing platform, sub-pixel motion estimation is performed on the video stream output by the low-light high-speed visual monitoring unit to calculate the real-time vibration amplitude value A_vis(t); simultaneously, the real-time fluctuation range W(t) is calculated on the continuous coordinate sequence output by the intelligent vertical coordinate instrument. Steady-state discrimination is achieved if and only if both the visual vibration amplitude and the coordinate fluctuation range simultaneously satisfy the steady-state condition, i.e.: In the formula, The visual judgment threshold (e.g., 0.02mm). This is the threshold for the range of coordinate fluctuations. If the vertical wire reaches an absolutely static steady state, a valid sampling trigger signal is output to the intelligent vertical coordinate instrument; if the threshold is not met, the resting time is extended until the condition is met or a "long-term decoupling failure anomaly" label is generated.

[0128] A-round data acquisition: After receiving a valid sampling trigger signal, the intelligent vertical coordinate instrument is controlled to perform the first coordinate acquisition to obtain the A-round coordinate data. ;

[0129] Interval waiting and B-round data acquisition: After completing the A-round data acquisition, according to the preset time interval... (e.g., 30 seconds) Keep the system in standby mode, then execute the multi-source fusion steady-state discrimination logic again. After confirming that it is in an absolutely static steady state again, control the coordinate instrument to perform secondary coordinate acquisition and obtain the B-measured coordinate data. ;

[0130] Deviation calculation and validity determination: Calculate the difference between the coordinate data from measurement A and measurement B. : ;

[0131] Judge the difference Is it less than the preset tolerance threshold? (like ):

[0132] like If the double-return self-consistency is determined, the mean of the coordinate data from returns A and B is taken as the final valid displacement data. Output: ;

[0133] like If the data is abnormal, a retest instruction is generated, and the above "test A back to test B" process is repeatedly executed, with a preset maximum number of retests N (e.g., N=3).

[0134] Anomaly tracing flag: If the maximum number of retests N is reached, the time difference value... Still greater than or equal to the tolerance threshold (Right now Then, extract the minimum net distance value synchronously output by the limit quantity sensing module. If the minimum net distance value is less than the preset near-wall threshold (Right now If so, then the "near-wall friction anomaly" label will be bound to the final output data packet.

[0135] 2) Multi-dimensional Physical Field Real-time Correction Engine: The core function of this engine is to decouple and dynamically compensate for multi-source errors in the final effective displacement data. It focuses on the impact of local gravity field anomalies (caused by reservoir water level fluctuations and landslide displacement) on the buoy sway and wire tension, accurately eliminating various interference errors and outputting highly reliable corrected displacement data to provide reliable support for subsequent monitoring and analysis. Its specific execution logic is as follows: A progressive two-dimensional correction model is used to correct the original effective displacement data... Layer-by-layer corrections are performed, incorporating compensation and elimination of local gravitational field anomalies throughout the process to ensure correction accuracy.

[0136] (1) Parameter extraction: All core parameters required for calibration are extracted synchronously from the structured multidimensional sensing data package to provide complete data support for the two-layer calibration. Specifically, these parameters include: ambient temperature value T and buoy immersion depth value. The buoy yaw displacement vector (ΔX, ΔY) (including the yaw component induced by local gravity field anomalies) and the original effective displacement data. Among them, the buoy yaw displacement vector has been synchronously captured to capture the slight yaw of the buoy caused by the local gravity field anomaly, providing accurate interference parameters for subsequent buoyancy yaw coupling correction; all relevant parameters have been incorporated into the slight influence of the local gravity field anomaly, avoiding distortion of the final displacement data due to the failure to fully consider the local gravity field anomaly.

[0137] (2) Progressive two-dimensional correction process: The two-layer progressive logic of “time-varying creep error correction of steel wire → buoyancy yaw coupling correction” is adopted to eliminate interference layer by layer, ensuring the pertinence and accuracy of the correction, and taking into account the influence of local gravity field anomalies throughout the process.

[0138] ① Time-varying creep error correction layer for steel wire

[0139] Since tension meters are not directly deployed in the sensing layer, the tension of the steel wire is indirectly calculated using the immersion depth of the float. The impact of local gravity field anomalies on the stress on the float and the tension calculation is taken into account. The specific steps are as follows:

[0140] a. Indirect tension calculation: Based on the real-time submersion depth of the buoy output by the state quantity sensing module. Combined with the cross-sectional area of ​​the pontoon Liquid density ρ, gravitational acceleration g (which incorporates minute changes caused by local gravitational field anomalies), and reference immersion depth The formula for calculating the current real-time tension F is: In the formula, F_0 is the wire tension under the initial reference state; g incorporates the influence of local gravity field anomalies to ensure that the tension calculation fits the actual working conditions on site and avoids tension calculation deviations caused by local gravity field changes.

[0141] b. Creep Compensation Calculation: Extract the current time node t and the calculated real-time tension F, and use the improved Kelvin-Voigt rheological model to calculate the time-varying creep compensation of the steel wire. The formula is: In the formula, the first term is the elastic elongation, and the last two terms are the transient and steady-state creep; L is the effective length of the vertical line, E is the elastic modulus of the steel wire, and S is the cross-sectional area of ​​the steel wire; parameters A, B, C, and D are dynamically assigned based on the parameter update packet received from the decision-making unit, and can be flexibly adjusted according to the abnormal local gravity field conditions to improve the creep compensation accuracy and offset the creep deviation indirectly caused by the abnormal local gravity field.

[0142] c. Output the reading after creep correction The formula is: .

[0143] ② Buoyancy yaw coupling correction layer

[0144] To precisely eliminate coupling errors caused by buoy sway (including sway induced by local gravity field anomalies), the focus is on eliminating interference from local gravity field anomalies superimposed on the displacement data through buoy sway. The specific steps are as follows:

[0145] a. Extract the sway displacement vector (ΔX, ΔY) output by the real-time monitoring unit for the sway and wall collision distance of the eddy current pontoon. This vector fully includes the sway component induced by the pontoon's own force and the sway component induced by the local gravitational field anomaly, ensuring that no interference parameters are omitted.

[0146] b. Calculate the projection coupling error of the yaw displacement vector in the monitoring direction. To accurately eliminate interference from local gravity field anomalies superimposed on displacement data by buoy sway, the formula is: In the formula, To ensure that the angle between the monitoring direction and the X-axis is accurate and that the coupling error calculation meets the monitoring requirements, the sway interference induced by local gravitational field anomalies is precisely countered.

[0147] (3) Data output: Based on the above two-layer correction, the final high-confidence corrected displacement data is calculated. This data has completely eliminated multi-source interference such as wire creep and buoyancy yaw (including yaw induced by local gravity field anomalies), and can truly reflect the structural displacement state. The formula is: ;Will The data is transmitted upwards to the decision-making unit, providing high-precision and high-reliability displacement data support for subsequent dynamic benchmark decoupling and structural anomaly early warning. This effectively avoids interference such as buoy swaying and tension deviation induced by local gravity field anomalies, and reduces the risk of misjudging structural damage.

[0148] 3) Optical Environment Intelligent Cognition Module: This module communicates with the CCD lens cleanliness online monitoring unit in the state quantity sensing module. It is used to quantify the health status of the optical sensing channel, perform data weighting, and issue cleaning strategies. Its specific execution logic is as follows:

[0149] Cleanliness quantification assessment: During non-effective sampling periods, receive real-time lens cleanliness assessment values ​​output by the online CCD lens cleanliness monitoring unit. ;

[0150] Dynamic mapping of data confidence: establishing cleanliness assessment values The mapping relationship model between the displacement data confidence weight W and the displacement data is as follows: ;in, The cleanliness threshold that triggers optical degradation; when When the optical channel is deemed healthy, the confidence weight W is assigned a value of 1; when When the optical channel fails, the confidence weight W is set to 0.

[0151] Cross-module data weighting and interception: The confidence weight W is bound to the synchronously acquired displacement data packet in real time; when W=0, the acquisition failure lockout signal is output to the wire front decoupling and dual-return intelligent acquisition module and the adaptive observation strategy engine, forcibly interrupting the calculation and output of displacement data to prevent distorted data from being transmitted to the upper layer;

[0152] Adaptive cleaning strategy scheduling: When the confidence weight W=0, an adaptive cleaning trigger command is generated and sent to the CCD lens cleanliness adaptive self-cleaning unit in the control execution layer;

[0153] Cleaning performance review and cloud collaboration: After receiving the cleaning completion signal from the optical self-cleaning actuator, a preset stabilization time is allowed. Repeat the cleanliness quantification assessment steps to obtain the reassessment cleanliness assessment value. If the reassessed value meets the recovery criteria: If the acquisition failure blocking signal is released, the normal observation process will resume; if the evaluation value after re-evaluation still meets the failure conditions: If the error occurs, a "stubborn optical stain alarm" label and a maintenance work order request will be sent to the decision-making unit, which will then schedule the subsequent manual maintenance.

[0154] 4) Adaptive Observation Strategy Engine: This engine dynamically generates observation frequency adjustment commands and operating condition interlocking signals based on environmental parameters, boundary states, and structural operating conditions. It employs the following multi-factor discrimination logic:

[0155] Boundary-triggered encryption: Real-time parsing of the minimum clearance value output by the boundary quantity sensing module. When the minimum clearance value meets the near-wall warning condition At that time, a frequency increase command is generated, controlling the sampling period of the observation system to be increased from the first preset period. Shortened to the second preset cycle (Right now );

[0156] Environmental interlock control: Real-time analysis of the moving average wind speed output by the environmental sensing module. With instantaneous wind speed When the perturbation threshold condition is met: ; Generate a working condition interlock signal; The working condition interlock signal is transmitted to the intelligent vertical coordinate instrument and the automatic vertical reset and data quality closed-loop control unit, forcibly interrupting the current and subsequent sampling and mechanical actions until the wind speed drops below the disturbance threshold and remains stable for a preset time. Then, a working condition recovery signal is generated;

[0157] Normal maintenance logic: When the rate of change of the load on the engineering structure (such as the rate of change of the reservoir water level) Landslide displacement rate ), rate of change of temperature gradient and minimum net distance value of limit quantity When both normal safety constraints are met: ;(in The preset first rate threshold, The preset first temperature threshold, (As a safety margin threshold), output low-frequency sampling period command. (like );

[0158] Emergency encryption logic: When an earthquake intensity trigger signal is received from an external system... Or, if the rate of change of load or the rate of change of temperature exceeds the corresponding emergency threshold. , Or receive a "deformation rate mutation" tag from the decision-making unit. When the emergency triggering conditions are met: ; Generate high-frequency sampling period instructions (like The instruction has a higher priority than the boundary trigger encryption logic;

[0159] Benchmark verification triggered: When a "Physical benchmark suspected of drift" alarm command is received from the decision-making unit. hour: Generate continuous sampling commands to control the observation system to operate at a preset high frequency. Preset number of consecutive days (like The data for that period is then packaged and uploaded to the dynamic benchmark decoupling module of the decision-making unit.

[0160] 2. Decision-making unit

[0161] The decision-making unit, acting as the "central nervous system" of the entire integrated sensing intelligent system, is deployed on a remote cloud platform. It receives high-reliability correction displacement data, multi-source sensing data packets, and various status warning tags uploaded by the cognitive unit. Through big data analysis, artificial intelligence algorithms, and digital twin technology, it achieves global decision-making, intelligent prediction, and predictive operation and maintenance scheduling. Simultaneously, it issues model optimization parameters and maintenance work orders, linking the cognitive unit and the control execution layer to form a closed-loop management system. This unit completely solves the deep-seated problems of traditional plumb line monitoring systems, such as "lack of intelligent diagnosis of structural safety boundaries," "unknown equipment health status," "passive and lagging operation and maintenance mode," and "inability to identify baseline drift." It achieves a leap from single data monitoring to full-domain intelligent diagnosis, trend prediction, and predictive maintenance, specifically including the following modules:

[0162] 1) Data Storage and Model Evolution Module: This module serves as the data foundation and algorithm support core of the decision-making unit. It is responsible for the persistent storage of multi-source heterogeneous data, time-series feature mining, and adaptive iterative optimization of the underlying correction model, ensuring the continuous improvement of the correction accuracy of the cognitive unit. Its specific execution logic is as follows:

[0163] (1) Multi-source heterogeneous data aggregation and standardization: Receive and structure the full amount of data from the cognitive unit, including high-reliability corrected displacement time series data, environmental factors (temperature, humidity, air pressure, wind speed, wind direction) time series data, equipment status characteristic data (float immersion depth, yaw displacement, lens cleanliness, wire clearance, etc.), visual evidence flow index, various early warning labels and action logs fed back from the execution layer, to build a monitoring database with a unified spatiotemporal benchmark, and clean and complete abnormal data and missing data to ensure the integrity and consistency of the data.

[0164] (2) Dynamic parameter identification and iterative optimization: For key parameters in the real-time correction engine of the multidimensional physical field of the cognitive unit, including the nonlinear temperature correction term. And the parameter set in the time-varying creep error correction layer of steel wire The optimization adopts an "offline mining + online update" approach. First, historical displacement data is extracted from long-term historical data during periods when the ambient temperature is stable, the structure is free from sudden external load changes, and the equipment is in normal condition. This data serves as valid samples for parameter fitting. Then, the least squares method or particle swarm optimization algorithm is used to refit the parameters offline, establishing a historical displacement residual sequence. Objective optimization function: Where N is the total number of historical data samples selected. This represents the original valid displacement data for the i-th historical sample. For the corresponding feature input matrix, This is a forward model for physical correction that includes the parameters to be optimized.

[0165] (3) Parameter update determination and distribution: Calculate the determination coefficient of the new model after fitting. Historical coefficient of determination corresponding to the current cognitive unit operating parameters When the model improvement meets the distribution conditions: ;(in For the preset increase threshold (it is recommended to set a value between 0.05 and 0.1), the generator will contain the latest parameters. and The model parameter update package is sent to the multi-dimensional physical field real-time correction engine in the cognitive unit through the Industrial Internet of Things, so as to realize the adaptive evolution of the underlying correction model and continuously improve the accuracy and robustness of displacement measurement.

[0166] 2) Security Boundary Digital Twin Mapping Module

[0167] This module, based on limit quantity sensing data and 3D digital twin technology, constructs a spatial geometric constraint model of the inverted plumb line system, enabling visualized monitoring, risk warning, and traceability of the spatial positional relationship between the steel wire and the protective pipe (hole wall). Its specific execution logic is as follows:

[0168] (1) Real-time spatial coordinate mapping: Receives two-dimensional center coordinate data of the vertical steel wire uploaded by the cognitive unit. Absolute coordinates of the cross-section of the protective pipe The discrete data of the inner wall radius of the protective pipe are used to complete the real-time coordinate mapping between the steel wire and the protective pipe in the pre-constructed three-dimensional digital twin model of the engineering structure, thus restoring their spatial relative position relationship.

[0169] (2) Solution of net distance distribution across the entire cross section: Let the radius of the inner wall of the protective pipe in the j-th discretized angular direction be... Solve the dynamic net distance distribution field of the steel wire relative to the hole wall across the entire cross section. : Furthermore, the minimum net distance value in the dynamic net distance distribution field of the entire cross section is extracted. and its corresponding direction vector angle : , .

[0170] (3) Threshold comparison and visual early warning: the minimum net distance value Compared with the pre-stored two-level security threshold (first-level warning threshold) Level 2 alarm threshold Real-time comparison is performed to trigger the corresponding spatial topology highlighting logic. When any threshold is triggered, a spatial topology highlighting mark for the corresponding risk section is generated in the 3D digital twin model. At the same time, a visual warning command containing the section station number S, net distance vector, and risk level is output to the operation and maintenance terminal, which facilitates operation and maintenance personnel to quickly locate the risk location and judge the degree of risk.

[0171] (4) Historical trajectory tracing and trend analysis: Store historical time series data of the net distance distribution field of the entire cross section, support maintenance personnel to trace the evolution trajectory of the near wall risk of steel wire, and combine the output of the multi-modal fusion prediction module to predict the development trend of the near wall risk of steel wire, providing a basis for decision-making for proactive intervention.

[0172] 3) Multimodal Fusion Prediction Module: This module, based on a time-series machine learning model, integrates multi-dimensional features of displacement, environment, and equipment to achieve dual prediction of structural deformation trends and equipment degradation trends. It also completes anomaly tracing and decoupling, distinguishing between structural anomalies and anomalies caused by equipment failures. Its specific execution logic is as follows:

[0173] (1) Construction of multimodal feature vector: Extract the output displacement data, ambient temperature value, load change rate value, float sway vector value and equipment status feature data of the cognitive unit within the preset time window, use Z-score data standardization to eliminate the influence of dimensions, and construct multimodal joint feature vector.

[0174] (2) Structural deformation trend prediction: The multimodal joint feature vector is input into a pre-trained Long Short-Term Memory (LSTM) network model, which outputs the confidence interval of displacement evolution within a preset prediction period. When the prediction confidence interval exceeds the preset structural safety displacement threshold, a structural deformation warning is triggered.

[0175] (3) Extraction of equipment deterioration characteristics and prediction of trends:

[0176] ①Float jamming feature extraction: Extract the micro-vibration spectrum of the steel wire output by the low-light high-speed visual monitoring unit of the cognitive unit, calculate the ambiguity index of the fundamental frequency of vibration and the rate of change of damping ratio; when the increase of damping ratio relative to the historical benchmark value exceeds 30%, or when there is a sudden increase in the proportion of abnormal high-frequency harmonic energy in the spectrum energy distribution, generate the "float jamming risk" feature vector.

[0177] ② Steel wire cross-sectional loss feature extraction: Residual analysis is performed on the actual creep variable calculated by the multidimensional physical field real-time correction engine and the output value of the theoretical creep model to calculate the residual sequence. When the residual sequence shows a continuous unidirectional increasing trend, and the interference of buoyancy fluctuations and temperature factors is eliminated simultaneously, a "suspected steel wire corrosion and wear" feature vector is generated.

[0178] ③ Oil deterioration feature extraction: Extract the dielectric constant and turbidity time series data output by the oil sensor in the state quantity sensing module. When the deterioration boundary conditions are met, generate the "damping fluid failure" feature vector.

[0179] ④ Equipment health trend prediction: Input the above equipment degradation feature vectors into the preset gradient boosting tree (GBDT) model, and output the equipment health degradation trajectory containing the probability of specific fault categories.

[0180] (4) Anomaly source decoupling: When the structural deformation trend prediction result triggers an early warning and the synchronously received equipment health degradation trajectory is in the normal range, a “structural anomaly” label is generated; when the equipment health degradation trajectory synchronously triggers a deterioration early warning, a “equipment failure suspected to cause data anomaly” label is generated, providing a clear direction for subsequent operation and maintenance.

[0181] 4) Dynamic reference decoupling module

[0182] This module is used to eliminate structural elastic deformation and pontoon sway interference caused by environmental periodic loads and local gravity field anomalies, extract structural anomaly residuals that truly reflect structural damage, and detect physical reference drift to ensure the accuracy and reliability of monitoring data. Its specific execution logic is as follows:

[0183] (1) Dynamic benchmark fitting: Obtain the environmental temperature value sequence, structural load value sequence, local gravity field anomaly related parameter sequence (including reservoir water level fluctuation data and landslide displacement data) within a long-period historical time window, and the corresponding period's cognitive unit output historical displacement data sequence, where the displacement data already includes the relevant correction information for pontoon sway (including sway induced by local gravity field anomaly). Using multivariate nonlinear regression algorithm, support vector regression algorithm, or deep learning algorithm (such as long short-term memory network LSTM, convolutional neural network CNN), relying on the strong fitting ability of deep learning algorithm for nonlinear and multi-factor coupling relationship, local gravity field anomaly is included as the core influencing factor in the model, and a dynamic prediction benchmark function model between environmental factors, local gravity field anomaly and displacement response is established to accurately describe the structural elastic deformation and pontoon sway superposition interference jointly driven by environmental conditions and local gravity field anomaly, further improving the accuracy and generalization ability of benchmark fitting.

[0184] (2) Physical reference drift detection: The displacement residual sequence is extracted during the stable period of environmental factors with constant or low rate of change and local gravity field. This sequence has eliminated the interference of environmental load and local gravity field anomalies on the buoy sway and structural deformation, ensuring that the residual sequence only reflects the physical reference state and the actual damage to the structure. The mutation point detection algorithm is executed on the sequence to extract the step jump amplitude. When an irreversible step jump is detected in the displacement residual and the amplitude exceeds the preset drift threshold, a "physical reference suspected drift" alarm command is generated. This command is sent to the cognitive unit to trigger the reference verification logic and recalibrate the physical reference to avoid the superposition of buoy sway induced by reference drift and local gravity field anomalies, which would lead to the distortion of monitoring data.

[0185] (3) Structural anomaly residual calculation and early warning: The current environmental factors and local gravity field anomaly parameters are substituted into the dynamic prediction benchmark function model in real time to calculate the expected elastic deformation under the current working condition and the buoy sway interference induced by local gravity field anomaly. Then, the difference between the measured high-confidence corrected displacement data and the above expected value is calculated as the structural anomaly residual after removing the environmental coupling effect and the buoy sway interference induced by local gravity field anomaly. When the structural anomaly residual exceeds the preset statistical confidence interval envelope, a structural ill-conditioning displacement early warning signal is triggered, indicating that the structure may have damage or abnormal deformation. At the same time, the interference and misjudgment caused by local gravity field anomaly and buoy sway are eliminated to ensure the accuracy of the early warning.

[0186] 5) Predictive Maintenance Scheduling Module: This module, based on equipment health assessment results, early warning tags, and remaining effective life prediction, enables intelligent and proactive maintenance decisions, transforming "reactive maintenance" into "prevention," reducing operation and maintenance costs, and ensuring the long-term stable operation of the system. Its specific execution logic is as follows:

[0187] (1) System comprehensive health quantification assessment: Receive the optical alarm tag uploaded by the cognitive unit, the equipment health degradation trajectory quantification value output by the multimodal fusion prediction module, and the benchmark drift alarm status quantity output by the dynamic benchmark decoupling module. Calculate the comprehensive health score of the inverted plumb line monitoring system based on the preset weight allocation algorithm.

[0188] (2) Remaining effective life calculation: Based on the slope characteristics of the equipment health degradation trajectory, the Wiener process or exponential degradation model is used to calculate the remaining running time required to trigger the lower limit threshold of the comprehensive health score, which is used as the predicted value of the remaining effective life.

[0189] (3) Intelligent work order generation and issuance: When the remaining effective life prediction value is less than the preset time margin threshold, or when a set of hard fault signals is received, the system automatically matches the preset "fault characteristics-maintenance strategy" mapping table, generates a predictive maintenance work order containing fault source location, maintenance operation guidance and spare parts demand list, pushes it to the operation and maintenance terminal execution system, and synchronizes it to the cognitive unit to suspend unnecessary operations of the corresponding module to avoid the fault from expanding; after the operation and maintenance is completed, the system receives the maintenance confirmation signal fed back by the terminal, updates the equipment health score and the remaining effective life prediction value, and forms a maintenance closed loop.

[0190] 3. Cognitive decision-making level collaborative mechanism

[0191] Within the cognitive decision-making layer, cognitive units and decision-making units, as well as various modules, achieve deep collaboration through a data bus, forming a complete closed loop of "cognitive analysis - global decision-making - execution feedback": the cognitive unit is responsible for real-time analysis, quality control, and strategy distribution of perceived data, providing high-quality data support for the decision-making unit; the decision-making unit is responsible for model optimization, global early warning, trend prediction, and operation and maintenance scheduling, issuing parameter update and control instructions to the cognitive unit; various modules work together, with the data storage and model evolution module providing data support and algorithm optimization for all other modules; risk warning signals from the security boundary digital twin mapping module trigger specialized analysis by the multimodal fusion prediction module, and are simultaneously pushed to the predictive maintenance scheduling module; equipment degradation characteristics and structural deformation warnings from the multimodal fusion prediction module provide a basis for anomaly tracing by the dynamic benchmark decoupling module, and also provide fault information for work order generation by the predictive maintenance scheduling module; benchmark drift alarms from the dynamic benchmark decoupling module trigger benchmark review by the cognitive unit, and simultaneously update historical data in the data storage and model evolution module; the maintenance work order execution results from the predictive maintenance scheduling module update equipment health data and model parameters in reverse, continuously improving the accuracy and relevance of decision-making. The cognitive unit and decision-making unit are deployed collaboratively on a remote cloud platform. Through the Industrial Internet of Things, they achieve data interaction and command loop with the monitoring and perception layer and the control and execution layer, ensuring the efficient and accurate operation of the entire monitoring system.

[0192] Within the cognitive decision-making layer, cognitive units and decision-making units, as well as various modules, achieve deep collaboration through a data bus, forming a complete closed loop of "cognitive analysis - global decision-making - execution feedback": the cognitive unit is responsible for real-time analysis, quality control, and strategy distribution of perceived data, providing high-quality data support for the decision-making unit; the decision-making unit is responsible for model optimization, global early warning, trend prediction, and operation and maintenance scheduling, issuing parameter update and control instructions to the cognitive unit; various modules work together, with the data storage and model evolution module providing data support and algorithm optimization for all other modules; risk warning signals from the security boundary digital twin mapping module trigger specialized analysis by the multimodal fusion prediction module, and are simultaneously pushed to the predictive maintenance scheduling module; equipment degradation characteristics and structural deformation warnings from the multimodal fusion prediction module provide a basis for anomaly tracing by the dynamic benchmark decoupling module, and also provide fault information for work order generation by the predictive maintenance scheduling module; benchmark drift alarms from the dynamic benchmark decoupling module trigger benchmark review by the cognitive unit, and simultaneously update historical data in the data storage and model evolution module; the maintenance work order execution results from the predictive maintenance scheduling module update equipment health data and model parameters in reverse, continuously improving the accuracy and relevance of decision-making. The cognitive unit and decision-making unit are deployed collaboratively on a remote cloud platform. Through the Industrial Internet of Things, they achieve data interaction and command loop with the monitoring and perception layer and the control and execution layer, ensuring the efficient and accurate operation of the entire monitoring system.

[0193] III. Control Execution Layer

[0194] The control execution layer is deployed at the observation site, and its command input terminal is connected to the cognitive decision-making layer (including cognitive units and decision-making units) for communication (see appendix for details). Figure 8 This layer receives observation strategy instructions, cleaning scheduling instructions, and maintenance work orders. Its status feedback terminal communicates with the monitoring and perception layer, providing real-time feedback on the action execution status to form a closed-loop control. The control execution layer, acting as the "action output unit" of the intelligent agent, transforms the intelligent instructions from the cognitive decision-making layer into precise physical actions. This addresses the pain points of traditional plumb line monitoring, such as "cumbersome operation and maintenance," "poor environmental adaptability," and "delayed emergency response," ensuring long-term stable operation of the system under complex working conditions. Its specific composition and execution logic are as follows:

[0195] 1. Buoyancy-tension safety interlock actuator

[0196] The buoyancy-tension safety interlock actuator communicates deeply with the cognitive unit of the cognitive decision layer and the state quantity perception module of the monitoring and perception layer. It specifically executes observation interruption, active compensation, and safety interlocking logic for abnormal mechanical states of the inverted plumb line system (tension exceeding limits, buoyancy attenuation), ensuring the inverted plumb line wire remains in a stable mechanical state and preventing measurement data distortion or equipment damage due to mechanical imbalance. Its specific execution logic is as follows:

[0197] 1) Abnormal Status Reception and Analysis: The system receives the "Tension Exceedance Abnormality" and "Buoyancy Decline Alarm" tags output by the cognitive unit in real time. At the same time, it reads the real-time tension value, tension deviation, and buoy immersion depth data fed back by the status quantity sensing module to complete the accurate analysis of the abnormality type, deviation degree, and abnormality cause, providing data support for subsequent action execution.

[0198] 2) Active Compensation Sub-Logic: When a "buoyancy attenuation alarm" is received but the hard cut-off threshold has not been reached, the required buoyancy increment is calculated based on the tension deviation, float cross-sectional area, and liquid density parameters. A micro-injection pump is then controlled to precisely inject transformer oil into the inverted plumb line float. The liquid level change is read in real-time from the precision measurement unit of the magnetostrictive float immersion depth. The actual buoyancy increment is dynamically calculated. When the actual buoyancy increment offsets the tension deviation and the steel wire is in a steady state (swing amplitude ≤ 0.02mm), the oil injection action stops, and a "mechanical compensation complete" signal is output to the cognitive unit, resuming the normal observation process.

[0199] 3) Hard cut-off of sampling link: When the “tension over-limit abnormal” tag is received, or the alarm tag is not cleared even when the oil injection volume reaches the preset maximum limit value for a single time (preset to 10% of the float volume), the hard cut-off trigger condition is met. The trigger enable circuit of the intelligent vertical coordinate instrument and the automatic vertical reset and data quality closed loop control unit is cut off through the hardware relay, and the current and subsequent automated observation cycles are forcibly interrupted to avoid invalid data acquisition and further damage to the equipment.

[0200] 4) State maintenance and recovery: During the interruption, the current system lock state is maintained, and any observation and mechanical action are prohibited; until the "fault clearance and process recovery" instruction is received from the cloud decision layer, the enable circuit of the hardware relay is closed again to restore normal operation, and a "state recovery" signal is fed back to the cognitive unit at the same time.

[0201] 2. Environmental adaptive control actuator

[0202] The environmental adaptive control actuator communicates bidirectionally with the environmental quantity sensing module and cognitive unit of the monitoring and sensing layer. This allows for the adaptive start and stop of local micro-environment control equipment based on on-site environmental conditions (wind speed, humidity, temperature), establishing stable observation conditions and eliminating the impact of environmental interference on measurement accuracy. Its specific execution logic is as follows:

[0203] 1) Control command reception and parsing: Real-time reception of the "environmental exceedance lockout signal" and the corresponding environmental parameter deviation vector (including the difference between wind speed, humidity, temperature and standard operating conditions) issued by the cognitive unit's adaptive observation strategy engine, to clarify the control target and priority.

[0204] 2) Adaptive start-stop logic: Based on the vector matching of environmental parameter deviations, a preset control strategy is matched to achieve precise intervention: When the wind speed exceeds the limit (instantaneous wind speed > 1.0 m / s or 10-second moving average wind speed > 0.5 m / s), the vent damper actuator is controlled to close, blocking the disturbance of external airflow to the vertical steel wire; when the humidity exceeds the limit (relative humidity > 85%), the dehumidifying fan and heating rod are activated to reduce the ambient humidity and prevent water vapor condensation from affecting optical measurements; when the temperature exceeds the limit (deviating from the standard observation temperature ± 5℃), the output power of the semiconductor temperature control device is adjusted, and a proportional-integral (PI) control algorithm is used for zero steady-state error adjustment to maintain a stable ambient temperature.

[0205] 3) Recovery Judgment and Closed-Loop Verification: Read the latest environmental parameters output by the environmental quantity sensing module in real time. When all parameters fall back to the preset effective observation condition range and continue to reach the preset stable duration (preset to 5 minutes), generate a "condition recovery ready" signal and upload it to the cognitive unit as a prerequisite for unlocking the observation lockout and restoring the normal observation cycle.

[0206] 3. Emergency Hard Interruption Protection Mechanism

[0207] The emergency hard interruption protection mechanism is the highest priority actuator in the control execution layer. It communicates independently with the limit quantity sensing module and other mechanisms within the control execution layer. It is used to perform hardware-level emergency protection actions under extreme physical conditions (contact between the steel wire and the hole wall, equipment overcurrent and overload) to minimize the risk of equipment damage. Its specific execution logic is as follows:

[0208] 1) Emergency trigger signal analysis: The independent monitoring limit quantity sensing module directly sends out the "contact alarm signal" (minimum clear distance between steel wire and hole wall ≤ 0.5mm), and the control execution layer internal "severe overcurrent / overload signal" (power supply current exceeds the rated value by 1.5 times). Define the emergency trigger comprehensive Boolean variable S_emergency. When either signal is triggered, S_emergency=1.

[0209] 2) Instantaneous physical protection action: When the emergency trigger variable S_emergency=1, the conventional scheduling logic of the cognitive unit is bypassed, and a hardware-level cut-off control signal is output through an independent hardware control loop to instantly cut off the power supply to non-core peripherals such as the flexible electromagnetic drive mechanism and the environmental control fan; if a mechanical locking module is configured, the locking mechanism is synchronously driven to lock the relative position of the vertical steel wire to prevent the steel wire from shaking out of control and causing further damage (such as steel wire wear and coordinate instrument optical path offset).

[0210] 3) Independent status reporting: While executing the hard interrupt action, a backup communication link independent of the main communication bus (such as a 4G / 5G backup module) is invoked to send an emergency message to the cloud decision-making layer, which includes the trigger type, action execution status and current device parameters, ensuring that the cloud can still accurately obtain the actual situation of underlying physical intervention when the main system is paralyzed, so as to provide a basis for subsequent operation and maintenance disposal.

[0211] 4. CCD lens cleanliness adaptive self-cleaning system (including actuator)

[0212] This system is in deep coordination with the CCD lens cleanliness online monitoring unit in the state quantity perception module of the monitoring perception layer and the optical environment intelligent cognition module of the cognition unit. It is used to perform adaptive physical cleaning, action execution and effect re-evaluation when the optical sensing channel is degraded, so as to realize closed-loop control from instruction issuance to cleaning completion, ensure the cleanliness of the optical measurement channel, and avoid measurement errors caused by image distortion. The specific implementation manners are as follows:

[0213] 1) Hardware structure composition: It includes a control unit and an adaptive self-cleaning actuator integrated on the outer side of the optical system window. The adaptive self-cleaning actuator includes a stepper motor-driven wiping component (using oil-resistant flexible wiper blade) and an air curtain purging component (high-pressure micro air pump). The control unit is integrated with an embedded processor, which is responsible for instruction parsing, action scheduling and status feedback. The overall protection level reaches IP68, which is suitable for the harsh working conditions of high humidity, dust and oil pollution in the observation environment.

[0214] 2) Closed-loop control logic: The control unit strictly executes the following logic process to ensure that the cleaning effect and measurement accuracy do not interfere with each other: ① Trigger reception and working condition interlock: When receiving the "acquisition interlock signal" and the lens cleanliness evaluation value E_clean issued by the cognition unit, the sampling function of the displacement perception module is forcibly interlocked, and coordinate data acquisition is prohibited; ② Strength calculation and adaptive cleaning: Match the preset "pollution degree-action parameter" mapping function based on the cleanliness evaluation value E_clean, calculate the cleaning action strength parameter vector (purging pressure, wiping times, action rate), and control the actuator to perform targeted decontamination according to the parameters; ③ Delay stabilization and effect re-evaluation: After the cleaning is completed, enter a delay waiting period (preset to 30s) to eliminate the interference of residual vibration of the actuator on the measurement, and then send a re-evaluation trigger instruction to the cleanliness monitoring unit; ④ Re-evaluation decision and status release: Receive the latest cleanliness evaluation value E_re after re-evaluation, compare it with the health threshold E_lim (60% of the gradient energy value in the healthy state). If E_re ≥ E_lim, release the acquisition interlock and resume normal observation; if E_re < E_lim, upload the label "Stubborn stain requires manual intervention" and the cleaning log to the decision unit, and wait for manual operation and maintenance.

[0215] 3) Safety interlock mechanism: The control unit is equipped with hardware timing interlock logic. During the cleaning operation, an interlock signal is output to forcibly prevent the flexible electromagnetic drive mechanism from starting, so as to avoid the cleaning operation and the steel wire tossing operation. At the same time, the cleaning operation parameters, cleaning duration, re-evaluation results and other information are written into the operation log and bound to the corresponding time period data packet for operation and maintenance traceability and effect analysis.

[0216] 4) Specific action logic of the actuator: When the contamination is light (E_lim>E_clean≥0.8E_lim), only the air curtain is activated (blowing pressure 0.3~0.5MPa, blowing time 10~15s) to avoid the scraping parts from damaging the lens; when the contamination is heavy (E_clean<0.8E_lim), a combination of "blowing pretreatment + scraping + synchronous blowing" is adopted. First, the floating dust is blown away, and then the scraping parts are controlled to scrape back and forth at a rate of 5mm / s 3~5 times. At the same time, the air curtain is activated to remove the stains generated by the scraping in time and ensure the cleaning effect.

[0217] 5) Action status feedback and safety protection: Real-time detection of the position of the wiping component and the drive current. If jamming occurs (drive current exceeds 1.2 times the rated value) or abnormal stroke occurs, the action will stop immediately and the abnormal log will be uploaded to the cognitive unit. After cleaning, the air curtain will be maintained at low power (0.1~0.2MPa) for 30 seconds to further remove residual dust, and then enter a stabilization cycle to ensure that the optical window is free from vibration and stains.

[0218] 6) Technical effects: Through the integrated design of the control unit and actuator, the cleaning strategy can be adaptively adjusted and the action can be precisely executed; through strict timing lockout and vibration elimination design, interference with high-precision displacement sampling is avoided, the optical channel measurement capability is restored from the physical source, and the lens contamination problem caused by high humidity, dust and oil in the observation environment is effectively solved, supporting the long-term stable operation of the system without human intervention.

[0219] 5. Control Execution Layer Coordination Mechanism

[0220] The various mechanisms at the control execution layer achieve command coordination and status linkage through a fieldbus (EtherCAT bus, with a transmission rate of no less than 100Mbps), constructing a complete control link of "command reception - action execution - status feedback - closed-loop verification" to ensure orderly, safe, and efficient actions. The specific coordination mechanism is as follows:

[0221] 1) Command Priority Sequencing: A strict descending order of command priority is defined for each mechanism: Emergency Hard Interrupt > Buoyancy-Tension Interlock > Optical Self-Cleaning > Environmental Adaptive Control. When a high-priority command is triggered, the execution of low-priority commands is automatically interrupted, prioritizing equipment safety and core measurement functions to avoid equipment damage or data distortion caused by command conflicts.

[0222] 2) Status cross-feedback: Each actuator feeds back the action status vector (execution status, action parameters, abnormal information) to the cognitive unit and the monitoring and perception layer in real time. The perception layer synchronously collects environmental and equipment status data after the action and uploads it to the cognitive unit for closed-loop verification to ensure that the execution action achieves the expected effect and forms a closed-loop linkage of "decision-execution-perception-review".

[0223] 3) Hardware interlock protection: Set up a global hardware interlock matrix to achieve mutual exclusion of multiple mechanism actions: lock the flexible electromagnetic drive mechanism when cleaning action or emergency interruption occurs; cut off the power supply to all non-core equipment when emergency triggering occurs; prohibit violent action of environmental control mechanism when buoyancy-tension interlock action is executed, and avoid interference of multiple mechanism linkage on the steady state of steel wire.

[0224] 4) Redundancy backup design: Core actuators (such as emergency hard interrupt circuits and micro-injection pumps) are equipped with redundant power supply and control links. When the main link fails, it automatically switches to the backup link through a redundancy fault-tolerant switching mechanism (switching time ≤ 100ms), which significantly improves the reliability and anti-interference capability of the control and execution layer system and ensures that the protection action does not fail under extreme working conditions.

[0225] Example 2:

[0226] This embodiment takes the inverted plumb line intelligent agent of a large hydropower station dam as an example. It strictly follows the three-layer closed-loop architecture of "monitoring and perception - cognitive decision-making - control execution" proposed in this invention. It carries out intelligent transformation on the basis of the existing inverted plumb line device, and fully reproduces the implementation details of the entire process of system hardware deployment, intelligent observation process, multi-dimensional error correction, equipment health diagnosis and cloud collaboration.

[0227] I. System Hardware Deployment

[0228] The hardware deployment in this embodiment follows the principle of "prioritizing the reuse of existing equipment and adding intelligence." No structural modifications are made to the existing plumb line mechanical devices (drilling, protective pipes, anchor blocks, steel wires, pontoons, etc.) on site. Only the newly added intelligent sensing and control execution modules are integrated into their periphery and key nodes to achieve non-intrusive intelligent upgrades.

[0229] 1. Reused components: Drilling, protective pipes, and anchor block assemblies

[0230] 1) Drilling and Protective Pipe: The existing stainless steel protective pipes installed in the foundation rock boreholes of the dam will be used. In this embodiment, the drilling depth is 100m, and the protective pipe is a seamless steel pipe with a diameter of ϕ168mm (wall thickness 8mm, inner diameter 152mm), with a smooth inner wall and an effective pipe diameter of not less than 100mm.

[0231] 2) Anchor block assembly: The absolute gravity reference of the plumb line is stably provided by the original anchor block and steel wire system. The monitoring and sensing layer of the present invention reads the relative displacement of the steel wire in a non-contact manner without changing the physical reference of the anchor block itself.

[0232] 3) Steel wire and pontoons: The existing ϕ1.5mm stainless steel wire and constant force pontoons will be used.

[0233] 4) Intelligent vertical coordinate instrument: Reuse the existing intelligent vertical coordinate instrument on site without replacement or parallel installation. Use its original installation location and fixing method, and directly connect it to the cognitive decision unit to realize data transmission and control.

[0234] 2. New intelligent transformation component: integration of sensing and control modules

[0235] Based on the existing equipment, the following modules were added to the observation room and inside and outside the protective tube. All modules were installed independently and had no mechanical interference with the original equipment, enabling online construction (without disassembling the original equipment).

[0236] 1) New modules added to the monitoring and sensing layer

[0237] (1) Non-contact toggle reset and stability control unit: It is installed on the side of the coordinate instrument. The flexible electromagnetic drive mechanism is fixed to the independent bracket through a precision linear guide rail. Its toggle direction is consistent with the monitoring direction. The gap between the end and the steel wire is kept at 5~10mm to realize non-contact mechanical toggle. It integrates a torque detection sensor (accuracy 0.01N·m) and a PLC control module.

[0238] (2) Magnetostrictive buoy immersion depth monitoring unit: externally installed on the outside of the buoy. The waveguide wire (0.5mm diameter stainless steel wire) is fixed to the fixed support structure connected to the oil drum or external reference support through an insulating bracket. The float magnetic ring is rigidly connected to the buoy and rises and falls synchronously with the buoy. The pulse detection circuit is directly integrated into the cognitive decision unit and connected through a coaxial cable. The whole machine has an IP68 protection rating and is suitable for oily environments. The outer cylinder of the buoy can be made of stainless steel. The monitoring unit is externally installed. The float magnetic ring is rigidly connected to the buoy through a clamp, without damaging the buoy's sealing structure.

[0239] (3) Real-time monitoring unit for eddy current type float yaw and wall-collision distance: 4 eddy current probes (oscillation frequency 1MHz, range 0~5mm) are installed symmetrically at 90° orthogonal on the inner side of the oil drum and the gap between the float and the oil drum through corrosion-resistant insulating brackets, and are completely immersed in transformer oil. The brackets are made of insulating material and have no contact with the oil drum or the float. They are adapted to the stainless steel material characteristics of the oil drum and the float, effectively avoiding the interference of metal materials on the eddy current monitoring signal and ensuring monitoring accuracy. The distance between the probe end face and the stainless steel outer wall of the float is preset to 2.5mm (midpoint of the range). It can simultaneously collect the two-dimensional yaw displacement vector of the float and the wall-collision distance between the float and the inner wall of the oil drum in real time. The vector calculation module is directly integrated into the cognitive decision unit. The probe is connected to the cognitive decision unit through the sealed interface of the oil drum through a waterproof sealed cable, without damaging the sealing performance of the oil drum. The monitoring data is directly used to correct the observation values ​​of the intelligent vertical coordinate instrument.

[0240] (4) Four-dimensional capacitive steel wire pipe internal position and clearance sensing unit: Four capacitive probes (stainless steel substrate, surface covered with 1~2mm polytetrafluoroethylene insulation layer) are embedded in the stainless steel protective pipe opening through an elastic expansion insulating bracket. The sensing sensor is only deployed at the pipe opening and does not extend into the protective pipe. The bracket adopts a nail-free fixing method (such as spring expansion ring or adhesive) to avoid drilling or structural damage to the protective pipe opening and has no contact with the inner wall of the protective pipe. The multi-channel capacitive detection circuit (sampling frequency 10Hz, accuracy 0.1pF) and spatial calculation module are directly integrated into the cognitive decision unit and connected to the pipe opening sensor through a sealed cable to realize accurate monitoring of the internal position and clearance status of the steel wire pipe. It is suitable for the installation scenario of the protective pipe opening and takes into account both monitoring accuracy and device integrity.

[0241] (5) CCD lens surface cleanliness online monitoring unit: The built-in target (high contrast checkerboard, 1mm×1mm) is integrated on the outside of the optical window of the intelligent vertical coordinate instrument, reuses the coordinate instrument image processing unit, and has no independent mechanical cleaning parts.

[0242] (6) Visual monitoring unit for the working status of the intelligent vertical coordinate instrument: It is mounted on an independent bracket, with a low-light CMOS image sensor (minimum illumination 0.1 lux, maximum frame rate 120fps, resolution 1080P) and an 850nm infrared fill light aligned with the core working area of ​​the coordinate instrument (optical lens, steel wire observation area). The edge preprocessing module is directly integrated into the cognitive decision unit.

[0243] (7) Environmental quantity sensing module: Deployed in the observation corridor or vertical shaft observation room, the high-precision 5-element meteorological station (integrating temperature ±0.2℃, humidity ±3%RH, air pressure ±0.1%FS, ultrasonic wind speed and direction) is directly connected to the cognitive decision unit bus through the digital interface, adapting to the installation scenario of the observation corridor or vertical shaft observation room, ensuring the accuracy and stability of environmental parameter collection.

[0244] 2) New mechanisms added to the control execution layer

[0245] (1) Buoyancy-tension safety interlock actuator: including a micro-injection pump (flow rate 0.1~1mL / min), an oil tank (5L), a filter and an electromagnetic control valve. The outlet of the injection pump is connected to the original float's reserved oil replenishment port (or a temporarily installed sealed interface) through a hose to realize automatic buoyancy compensation without changing the original sealing structure of the float, and is directly controlled by the cognitive decision unit.

[0246] (2) Environmental adaptive control actuator: including dehumidifying fan and heating rod, semiconductor temperature control device (attached to coordinate instrument chassis), all of which are directly controlled by cognitive decision unit through relay or PID module, without the need to set electric damper actuator, adapting to the environmental control needs of observation corridor and vertical shaft observation room, ensuring stable working environment of equipment.

[0247] (3) Emergency hard interruption protection mechanism: Independent hardware control loop, including overcurrent detection module (monitoring power supply current) and hardware relay (cutting off power supply to non-core peripherals). Mechanical locking module is optional, but not mandatory, and its trigger signal is directly output by the cognitive decision unit.

[0248] (4) CCD lens cleanliness adaptive self-cleaning system: integrated on the outside of the coordinate instrument's optical window, including a stepper motor-driven oil-resistant flexible scraper and a high-pressure micro air pump (air curtain purging). The control unit integrates an embedded processor and communicates directly with the cognitive decision unit via an independent cable.

[0249] Important Note: All mechanical components of the original plumb line device (steel wire, floats, anchor blocks, protective pipes, oil drums, etc.) remain unchanged without any structural modifications or drilling. All intelligent functions of this invention are implemented through externally added modules. Each module connects directly to the cognitive decision-making unit, eliminating the need for an edge computing gateway. Construction can be carried out without interrupting the original monitoring system, and a seamless switch to intelligent mode is possible after the upgrade. Data transmission and control command issuance for all newly added sensors and actuators are directly handled by the cognitive decision-making unit, ensuring multi-source data timestamp accuracy ≤1ms. The cognitive decision-making unit communicates with the remote cloud platform via a dedicated fiber optic network.

[0250] II. Implementation Process of Intelligent Observation and Data Quality Control Closed-Loop

[0251] This embodiment details a complete automated observation process, strictly following the closed-loop logic of "monitoring and perception - cognitive decision-making - control execution," corresponding to... Figure 4 The process of pre-decoupling of steel wire, steady-state discrimination, and double-test self-verification.

[0252] 1. Step One: Environmental Prediction and Adaptive Strategy Loading (Cognitive Unit)

[0253] Cognitive decision-making unit timed wake-up system, the adaptive observation strategy engine first reads the real-time wind speed (instantaneous value Vw, 10s sliding average value ṼT) of the environmental quantity perception module, as well as the temperature and humidity, and the minimum clear distance value dmin of the limit quantity perception module.

[0254] Working condition discrimination and execution:

[0255] If Vw>1.0m / s or ṼT>0.5m / s is satisfied, the engine generates an "acquisition blocking signal" to forcibly interrupt the process; meanwhile, a "disturbed state" identifier is added to the data packet.

[0256] If dmin<dwarn (preset to 2mm), the engine generates a "frequency increase instruction" to temporarily shorten the subsequent observation period from the default 24 hours to 1 hour.

[0257] If the temperature and humidity exceed the effective observation range (for example, temperature < -10°C or > 50°C, humidity > 90%), the system records "environmental over-standard" and postpones the observation, and simultaneously starts the dehumidification fan, the heating rod, and the semiconductor temperature control device to perform adaptive adjustment on the observation environment, and resumes the normal observation process only after the temperature and humidity return to the effective observation range.

[0258] After confirming that the environmental working condition meets the observation conditions, proceed to the next step.

[0259] 2. Step 2: Pre-positive active decoupling and multi-source fusion steady-state discrimination (cognitive unit + control execution layer)

[0260] Pre-positive active decoupling: The cognitive decision-making unit issues a "pre-decoupling trigger signal" to the non-contact toggle reset and stability control unit. The control module drives the flexible electromagnetic drive mechanism according to a preset logic (initial toggle torque T0=0.5N⋅m, stroke 0.8mm) to apply a non-contact mechanical toggle force to the steel wire, actively destroying the non-free steady-state false image that may be caused by foreign matter jamming, oil film adhesion or slowly changing environmental force. During the toggling process, the torque detection sensor feeds back the torque value to the cognitive decision-making unit in real time, and immediately gives way and reports an abnormality if the value exceeds the safety threshold Tsafe=1.5N⋅m.

[0261] Free reset and long-term standing: Remove the toggle force, the steel wire resets freely under the buoyancy and damping action of the buoy, and the system enters a mandatory standing waiting period of no less than 30 minutes, so as to ensure that the steel wire fully attenuates the residual mechanical kinetic energy in the liquid damping environment.

[0262] Multi-source fusion steady-state discrimination: Before and after the end of the standing period, start the vision-coordinate dual feature extraction. The cognitive decision-making unit processes synchronously:

[0263] Vision data: Perform sub-pixel motion estimation on the video stream of the working state visual monitoring unit, and solve the real-time vibration amplitude Avis(t).

[0264] Coordinate data: Calculate the real-time fluctuation range W(t) of the continuous coordinate sequence output by the intelligent vertical coordinate instrument.

[0265] The steel wire is considered to have reached an absolutely static steady state if and only if Avis(t) < εvis (preset 0.02 mm) and W(t) ≤ εstatic (preset 0.02 mm), and a valid sampling trigger signal is output to the coordinate instrument. If the condition is not met, the resting time is extended (by 5 minutes each time) until the condition is met, or if the condition is not met after a total resting time of more than 60 minutes, a "long-term decoupling failure anomaly" label is generated.

[0266] 3. Step Three: Fully Automated Dual-Regression Data Acquisition and Self-Consistency Verification (Cognitive Unit + Perception Layer)

[0267] A. Acquisition: After the steady-state discrimination is passed, the cognitive decision-making unit triggers the intelligent vertical coordinate instrument to perform the first acquisition.

[0268] Image acquisition: The optical imaging system emits parallel light, and the linear CCD acquires the image of the wire shadow.

[0269] Adaptive enhancement: The image processing unit performs adaptive exposure compensation (dynamic gain adjustment coefficient α is 0.8~1.2) to eliminate light decay and water vapor interference caused by low ambient light and high humidity.

[0270] Subpixel calculation: The Sobel operator is used to extract the gradient, and the Gaussian interpolation curve is used to fit the subpixel level edge coordinates to accurately calculate the center coordinates (XA, YA) of the wire with a resolution better than 0.01mm. The data is directly transmitted to the cognitive decision unit.

[0271] Interval waiting and B-test: After completing the A-test, at an interval of ΔtAB=30 seconds, repeat the “pulling-resetting-steady-state discrimination” process in step two. After confirming again that the steel wire is in an absolutely static steady state, obtain the B-test coordinate data (XB, YB) and transmit it to the cognitive decision unit.

[0272] Deviation calculation and validity determination:

[0273] The cognitive decision-making unit calculates the two-test error Δ = √[(XA - XB)]. ² +(YA-YB) ² If |Δ| < δ (preset tolerance threshold δ = 0.15 mm), then the double-retest self-consistency is deemed valid, and the output mean Dnew = ((XA + XB) / 2, (YA + YB) / 2).

[0274] 4. Step Four: Independent Secondary Observation and Data Integration (Optional, for scenarios requiring high precision)

[0275] For critical sections (such as deformation-sensitive areas of dam foundations), the system can be configured to execute an enhanced mode of "dual measurement + independent secondary observation": After completing the above dual measurement acquisition, the system enters a sleep waiting mode, with a timer of 15 minutes to ensure the steel wire is completely stable and free from environmental disturbance. After the timer expires, "step two + step three" is completely repeated to obtain the second set of independent observation values ​​Rind=(Xind, Yind) and transmit them to the cognitive decision unit. The cognitive decision unit calculates the difference between the two sets of independent observation values. If the difference is ≤0.2mm, the weighted average of the two sets of observation values ​​(each with a weight of 0.5) is taken as the original valid displacement data for this observation; if the difference is >0.2mm, a manual review process is triggered.

[0276] III. Correction of Observations Based on Real-time Monitoring Unit for Yaw and Collision Distance of Eddy Current Float

[0277] The core function of the real-time monitoring unit for eddy current type float yaw and collision distance (formerly the eddy current type float yaw monitoring unit) is to collect the two-dimensional yaw displacement of the float and the collision distance between the float and the inner wall of the oil drum in real time. The monitoring data is directly used to correct the observation values ​​of the intelligent vertical coordinate instrument, focusing on eliminating the interference of float offset (including yaw induced by local gravity field anomalies) on the steel wire displacement monitoring, improving the accuracy and reliability of the monitoring data, and ensuring that the final output displacement data can truly reflect the deformation state of engineering structures such as dams, providing high-quality data support for subsequent dynamic benchmark decoupling, structural anomaly early warning and digital twin mapping.

[0278] 1. Parameter Extraction

[0279] The cognitive decision-making unit simultaneously reads two sets of core monitoring data output from the real-time monitoring unit for the sway and wall-collision distance of the eddy current float, ensuring complete consistency with the timestamps of the data observed by the intelligent plumb line coordinate instrument. This provides data synchronization assurance for accurate correction. The specific extracted parameters are as follows:

[0280] Two-dimensional yaw displacement vector (Unit: mm): Reflects the horizontal displacement of the buoy caused by buoyancy fluctuations, installation deviations, environmental disturbances, and local gravity field anomalies (resulting from reservoir water level fluctuations and landslide displacement), accurately capturing all types of horizontal micro-movements of the buoy;

[0281] Real-time contact distance between the float and the inner wall of the oil drum (Unit: mm): Used to help determine the rationality of the buoy's offset and whether there is a risk of jamming, while also providing a basis for triggering the correction logic and calculating errors.

[0282] 2. Determination of the reasonableness of the offset

[0283] To avoid using abnormal yaw data caused by pontoon jamming or hitting the wall for correction, which would affect the correction accuracy, the correction is first based on the real-time wall-hitting distance. To determine whether the pontoon offset is within the normal operating range, a graded judgment rule is formulated based on abnormal local gravity field conditions, as follows:

[0284] Normal offset determination: The minimum safe distance between the preset float and the inner wall of the oil drum during normal operation. ,like The float was determined to be normally offset (including slight swaying induced by local gravity field anomalies), and its sway displacement data can be used to correct the observation values ​​of the intelligent vertical coordinate instrument.

[0285] Excessive offset warning: If If it is determined that the buoy has excessive offset or slight jamming, the observation value correction will not be performed for the time being. At the same time, an "excessive buoy offset" warning signal is generated and uploaded to the cloud decision unit to prompt the operation and maintenance personnel to check whether there are problems such as abnormal increase in local gravity field or failure of buoy damping.

[0286] Handling of jamming abnormalities: If If it is determined that the float is stuck against the inner wall of the oil drum, the float jamming diagnosis process is immediately triggered. The data observed by the intelligent vertical coordinate instrument is marked as abnormal and will not be included in the correction or any subsequent calculations to avoid abnormal data interfering with the structural deformation analysis.

[0287] 3. Coupling error calculation

[0288] For float offsets determined to be within the normal range, the projection coupling error of the two-dimensional yaw displacement vector in the monitoring direction of the intelligent vertical coordinate instrument is calculated. This error is the interference of float offset (including yaw induced by local gravity field anomalies) on the steel wire displacement observation value. This ensures that the error calculation is consistent with the actual monitoring scenario, as detailed below:

[0289] Assume the monitoring direction of the intelligent vertical coordinate instrument is the horizontal X-axis (along the river direction), and the coupling error... The calculation formula is: In this embodiment, the monitoring direction is completely consistent with the X-axis, that is... ( , Therefore, the coupling error simplifies to: This simplified value precisely corresponds to the deviation of the intelligent vertical coordinate instrument observation value caused by the buoy offset (including the sway induced by local gravity field anomalies), providing accurate error parameters for subsequent correction.

[0290] 4. Observation correction execution

[0291] The cognitive decision-making unit adopts the "error cancellation" principle to calculate the coupling error caused by the buoy offset between the effective displacement data after double-measurement verification (or the data after integration of independent secondary observations) and the buoy offset. This completes the accurate correction of the intelligent vertical coordinate instrument observations and outputs the final high-reliability corrected displacement data. The specific execution process is as follows:

[0292] (1) Take the mean effective displacement after double measurement verification (or the displacement data after integration of independent secondary observations) as the correction benchmark value. ;

[0293] (2) The final high-confidence corrected displacement data are calculated using the following formula. : In the formula, This is the coupling error caused by the buoy's offset (including the sway induced by local gravity field anomalies);

[0294] (3) The final corrected data is encapsulated in a standardized JSON format, and the encapsulated content includes timestamps, original observations from the intelligent vertical coordinate instrument, and buoy yaw displacement vectors. Real-time wall collision distance Key information such as the intermediate value of the coupling error correction term and the equipment operating status label (normal / warning / abnormal);

[0295] (4) The final corrected data after packaging is directly uploaded to the cloud decision unit by the cognitive decision unit to ensure linkage with other monitoring data and ensure data consistency and traceability.

[0296] 5. Supplementary procedures for handling abnormal situations

[0297] To further ensure the reliability of the corrected data, a coupling error anomaly handling mechanism is added: if the calculated coupling error... (Including errors caused by swaying induced by local gravity field anomalies) The cognitive decision unit automatically marks the corrected data as "to be reviewed" and pushes it to the cloud simultaneously, prompting maintenance personnel to further verify it in conjunction with the pontoon status diagnosis results (such as jamming, abnormal sealing, abnormal fluctuations in local gravity field, etc.), so as to completely avoid the distortion of monitoring data caused by excessive coupling error and ensure that the corrected data can truly reflect the actual deformation state of the engineering structure.

[0298] IV. Implementation Details of Float Submersion Status Monitoring and Automatic Oil Injection Control

[0299] The cognitive decision-making unit monitors the buoy's immersion status in real time. Only when the buoy is not completely submerged in transformer oil will the automatic oil injection function of the buoyancy-tension safety interlock actuator be activated to ensure stable buoyancy and guarantee monitoring accuracy.

[0300] 1. Submersion status monitoring and criteria

[0301] Parameter extraction: The cognitive decision unit reads in real time the real-time immersion depth Hsub (mm) of the buoy output by the magnetostrictive buoy immersion depth monitoring unit, and simultaneously extracts the preset reference depth Hfull for full buoy immersion (determined by installation calibration, Hfull=100mm in this embodiment).

[0302] Immersion state criterion: When the real-time immersion depth Hsub < Hfull, it is determined that the buoy is not fully immersed in the transformer oil, the buoyancy of the buoy is insufficient, which may affect the steady state of the steel wire and the monitoring accuracy, and the cognitive decision unit immediately triggers the automatic oil injection process; when Hsub ≥ Hfull, it is determined that the buoy is fully immersed, no oil injection is required, and the normal monitoring state is maintained.

[0303] 2. Automatic oil injection control execution

[0304] Oil injection instruction issuance: The cognitive decision unit issues an automatic oil injection instruction to the buoyancy-tension safety interlock actuator to control the start of the micro oil injection pump, the oil injection pump injects transformer oil into the buoy at a preset flow rate (0.1~1mL / min), and the change of Hsub is monitored in real time during the oil injection process.

[0305] Oil injection stop condition: When Hsub fed back by the magnetostrictive buoy immersion depth monitoring unit is greater than or equal to Hfull, the cognitive decision unit immediately issues an oil injection stop instruction, turns off the oil injection pump and the electromagnetic control valve, stops oil injection, ensures that the buoy is fully immersed in the transformer oil, and maintains stable buoyancy.

[0306] Abnormal handling: If Hsub still does not reach Hfull after 5 minutes of continuous oil injection, the cognitive decision unit generates an "oil injection abnormality" alarm, uploads it to the cloud decision unit synchronously, and pushes a maintenance work order to prompt operation and maintenance personnel to check the oil injection pump, oil circuit and buoy sealing state, so as to avoid the insufficient buoyancy of the buoy affecting the normal operation of the monitoring system.

[0307] Data recording: The cognitive decision unit automatically records the time of each oil injection, the duration of oil injection, and the Hsub values before and after oil injection, forms an oil injection log, and uploads it to the cloud for storage synchronously to provide data support for subsequent operation and maintenance.

[0308] V. Implementation of equipment health diagnosis and predictive maintenance

[0309] Based on the equipment state data collected by the state quantity perception module, combined with the built-in algorithm of the cognitive decision unit and the analysis of the cloud decision layer, the active operation and maintenance of the equipment in the whole life cycle is realized. All data processing and instruction issuance are completed by the cognitive decision unit, and no edge computing gateway is provided.

[0310] 1. Intelligent diagnosis and handling of buoy jamming

[0311] Diagnostic principle: Under normal working conditions, the vibration of the steel wire after being toggled and reset approximates simple harmonic vibration, with a single frequency spectrum and a high fundamental frequency (approximately 0.5~2Hz) energy proportion; if the buoy rubs against or is stuck by the wall of the oil barrel, the vibration waveform is distorted, the fundamental frequency energy proportion decreases, and high-frequency harmonics may appear.

[0312] Implementation steps: During the preposed toggling and reset process of "Step 2", the cognitive decision-making unit collects the full-process vibration waveform x(t) (the sub-pixel motion displacement sequence from the low-illumination high-speed visual monitoring unit) at a sampling frequency of 120fps for a duration of 30s. The cognitive decision-making unit performs FFT transformation on x(t), and calculates the fundamental frequency energy proportion η=Ef0 / Etotal, where Ef0 is the energy within ±0.2Hz of the center frequency, and Etotal is the total energy from 0 to 20Hz.

[0313] Criteria and disposal: If η≥0.8, the cognitive decision-making unit determines that the state of the buoy is normal.

[0314] If 0.6≤η<0.8, it is determined as "slight jamming risk", and the cognitive decision-making unit automatically issues an instruction to the non-contact toggling reset and stability control unit to perform "jamming releasing operation" (adding one more small-amplitude toggle, with the torque increased to 0.8N·m).

[0315] If η<0.6 or the jamming releasing operation is invalid for 3 consecutive times, combined with the Dwall data of the distance to the barrel wall output by the eddy current type real-time monitoring unit for buoy deflection and barrel wall collision, if Dwall<Dsafe (preset as 5mm) or Dwall=0, it is further confirmed that the buoy is severely jammed, and the cognitive decision-making unit generates the "buoy severe jamming" feature vector, uploads it to the cloud multi-modal fusion prediction module, triggers the predictive maintenance scheduling module, and automatically pushes a maintenance work order containing the operation guideline of "inspecting the inner wall of the oil barrel and the buoy guiding mechanism".

[0316] 2. Adaptive closed-loop management of CCD lens cleanliness

[0317] Monitoring: During non-effective sampling periods (such as the interval between two observations), the intelligent optical environment cognition module built in the cognitive decision-making unit receives the cleanliness evaluation value Eclean from the CCD lens cleanliness online monitoring unit (calculated based on the Tenengrad gradient function, with a typical value of about 8000~12000 in the healthy state).

[0318] Decision and execution: When Eclean<Elim (the health threshold, preset as 60% of the gradient energy value in the healthy state, and Elim=5000 in this embodiment), the cognitive decision-making unit generates an "acquisition interlock signal" and issues it to the displacement perception module to forcibly interrupt the displacement data solution and output; at the same time, the CCD lens cleanliness adaptive self-cleaning system is triggered.

[0319] Light pollution (5000≤Eclean<8000): The cognitive decision-making unit controls to start only air curtain purging (purging pressure 0.4MPa, duration 12s).

[0320] Heavy pollution (Eclean<5000): The cognitive decision-making unit controls to adopt the combined action of "purging pretreatment + scraping + simultaneous purging". First, perform air curtain purging for 5s to remove floating dust, then control the stepper motor to drive the wiper blade to perform reciprocating scraping 3 times at a rate of 5mm / s, and simultaneously turn on the air curtain purging during the scraping process to timely remove the stains generated by scraping.

[0321] Re-evaluation and feedback: After cleaning is completed, the system delays for 30s to eliminate residual vibration, and the cognitive decision-making unit controls to re-acquire the target image and calculate the re-evaluation cleanliness assessment value Ere.

[0322] If Ere≥Elim, the cognitive decision-making unit releases the acquisition lock, restores the normal observation process, and records the cleaning action log.

[0323] If Ere<Elim, the cognitive decision-making unit outputs an "optical stubborn stain alarm" and a maintenance work order request to the cloud decision-making unit, suggesting manual lens cleaning.

[0324] VI. Cloud Decision-Making and Digital Twin Collaboration

[0325] As the "central brain" of the entire inverted plumb line intelligent monitoring system, the cloud decision-making unit realizes in-depth direct collaboration with the cognitive decision-making unit, and is mainly responsible for the functions of global diagnosis, model evolution and visual management and control. The two establish a high-speed direct communication link through 4G / 5G wireless network or optical fiber private network, without any edge computing gateway, to ensure the real-time, integrity and security of data transmission, and realize two-way efficient interaction of monitoring data and control instructions.

[0326] 1. Dynamic reference decoupling and structural abnormality early warning

[0327] Relying on the high-trust data uploaded by the cognitive decision-making unit, the cloud decision-making unit focuses on the influence of local gravity field anomalies (caused by reservoir water level fluctuation and landslide mass displacement) on the buoy pendulum, and completes the double elimination of environmental interference and local gravity field anomaly interference and structural abnormality identification through the dynamic reference decoupling module, so as to ensure the accuracy of early warning. The specific implementation process is as follows:

[0328] 1) Reference fitting: The dynamic reference decoupling module calls the long-period historical monitoring data of the past 365 days, which is a group of high-trust corrected displacements synchronously uploaded by the cognitive decision-making unit every day (the error caused by the buoy pendulum induced by local gravity field anomalies has been eliminated), and synchronously matches the ambient temperature of the corresponding period , reservoir water level Load factors such as parameters related to local gravity field anomalies (e.g., reservoir water level fluctuations, landslide displacement) are considered. A multivariate nonlinear regression algorithm (integrating deep learning algorithms to optimize fitting accuracy and enhance the fitting ability of the coupling relationship between local gravity field anomalies and environmental factors) is used to establish a dynamic prediction benchmark model. This model accurately describes the structural elastic deformation and the superimposed disturbance of pontoon sway, jointly driven by environmental conditions and local gravity field anomalies. The model expression is as follows: In the formula, The predicted value of structural elastic deformation is driven by both environmental conditions and local gravity field anomalies. This is a seasonal factor used to adapt to the different effects of environmental loads in different seasons. These are parameters of local gravity field anomalies, used to quantify the impact of local gravity field anomalies on the elastic deformation of the structure and the sway of the pontoon.

[0329] 2) Residual calculation: The cloud-based decision unit receives the current high-reliability correction displacement uploaded by the cognitive decision unit in real time. (Correction of float sway error induced by local gravity field anomaly has been completed.) Based on the elastic deformation prediction value output by the dynamic prediction benchmark model, the structural anomaly residual is calculated in real time. This residual, after eliminating environmental interference, local gravity field anomaly interference, and float sway effects, truly reflects the core parameters of the structural deformation. The residual data is fed back to the cognitive decision-making unit in real time. The residual calculation formula is: ;

[0330] 3) Anomaly warning: When structural anomalies occur in the residuals... When the displacement exceeds the preset statistical confidence interval envelope (e.g., 3 standard deviations), the cloud-based decision-making unit immediately triggers a "structural pathological displacement early warning" signal and simultaneously sends it to the cognitive decision-making unit. Upon receiving the signal, the cognitive decision-making unit simultaneously issues a local alarm, indicating that the engineering structure may have damage or abnormal deformation, providing early warning support for early hazard handling, and eliminating misjudgments caused by interference from local gravity field anomalies and pontoon sway. Simultaneously, the cognitive decision-making unit monitors irreversible step jumps in residual data in real time, focusing on distinguishing whether the step jump is caused by physical reference drift or induced by sudden fluctuations in the local gravity field. If the jump amplitude exceeds the preset physical reference drift threshold (e.g., 0.3mm), and the influence of local gravity field anomalies is ruled out, an "physical reference suspected drift" alarm command is immediately generated and uploaded to the cloud-based decision-making unit. At the same time, it triggers its built-in reference verification logic, initiating a reference calibration process to avoid the superposition of reference drift and local gravity field anomaly interference leading to data distortion, ensuring the accuracy and reliability of the early warning.

[0331] 2. Digital twin visualization of security boundaries

[0332] The cloud-based decision-making unit integrates a digital twin mapping module for security boundaries, which collaborates with the cognitive decision-making unit to achieve real-time visual control of the status of steel wires and protective pipes, intuitively presenting safety risks. The specific implementation process is as follows:

[0333] 1) Real-time mapping: The cognitive decision-making unit maps the two-dimensional center coordinates of the steel wire calculated by the limit quantity sensing module. absolute coordinates of the protective pipe cross-section The data is uploaded to the cloud-based decision-making unit in real time. The safety boundary digital twin mapping module, based on a pre-built 3D digital twin scene of the engineering structure, accurately maps the position information of the steel wire and protective pipe in real time, restoring their relative spatial positions. The inner wall radius of the protective pipe is preset to 76mm (corresponding to an inner diameter of 152mm). The cognitive decision-making unit stably uploads the steel wire position data to the cloud every 5 minutes to ensure the real-time nature of the mapping.

[0334] 2) Full-section net distance distribution: The cognitive decision unit synchronously calculates the full-section dynamic net distance distribution field of the steel wire relative to the wall of the protective pipe, and extracts the minimum net distance value of the full section in real time. and its corresponding direction vector angle The core data will be uploaded to the cloud-based decision-making unit to provide accurate information for security risk assessment.

[0335] 3) Tiered risk warning and intervention: Cloud-based decision-making units are based on minimum net distance values. Set two-level early warning thresholds to achieve tiered risk management: ① When Triggering Level 1 warning threshold ( When the risk vector and direction are simultaneously marked, the corresponding section in the cloud-based digital twin model is highlighted in orange, and the cloud-based decision-making unit sends an early warning signal to the cognitive decision-making unit to alert maintenance personnel; ② When When the net distance change rate suddenly drops to zero, it is judged as a high-risk state. The corresponding section in the cloud-based digital twin model will flash red to trigger an alarm, and the cognitive decision unit will simultaneously issue a local audible and visual alarm. Maintenance personnel can use the cloud-based digital twin platform to click on the corresponding section to view the historical trajectory of the steel wire position and the net distance change trend curve, quickly locate the root cause of the risk, and formulate proactive intervention plans (such as adjusting the alignment of the protective pipe, cleaning debris from the borehole wall, etc.). The intervention command is directly issued to the cognitive decision unit through the cloud-based decision unit, which controls the corresponding execution mechanism to complete the intervention operation, forming a closed-loop management system of "monitoring-early warning-decision-intervention".

[0336] 3. Model Evolution and Parameter Distribution

[0337] The cloud-based decision-making unit, through its data storage and model evolution module, enables iterative optimization and online updates of the cognitive decision-making unit's built-in parameters, ensuring continuous high-precision monitoring. The specific execution process is as follows:

[0338] 1) Iterative Optimization: The data storage and model evolution module extracts high-quality historical monitoring data accumulated monthly. This data has been synchronously uploaded by the cognitive decision-making unit, and abnormal labeled data and data from environmentally closed periods have been pre-filtered to ensure data validity. The cognitive decision-making unit uses built-in parameters related to buoy immersion monitoring (such as baseline immersion depth) to optimize the data. Taking parameters such as oil injection flow rate threshold as the optimization object, the least squares method or particle swarm optimization algorithm is used to construct the objective optimization function, and offline refitting of the parameters is achieved. The optimization function is: In the formula, The set of parameters to be optimized For historical data sample size, For the first Group of historical immersion depth monitoring residuals.

[0339] 2) Parameter update determination: After the model iterative optimization is completed, the determination coefficient of the newly fitted model is calculated. And the determination coefficients corresponding to the current cognitive decision-making unit's operating parameters. Compare the results. When the model improvement meets the following conditions... (in When the preset improvement threshold (valued at 0.05) is set, the new model is deemed to have a significant optimization effect, and the cloud-based decision unit generates a model containing the latest parameters. The model parameter update package.

[0340] 3) Online Updates: The cloud-based decision-making unit directly sends model parameter update packages to the cognitive decision-making unit via the industrial IoT communication link. Upon receiving the update package, the cognitive decision-making unit automatically verifies the integrity and validity of the parameters. Before the end of the current observation period and the start of the next, it loads new parameters for the pontoon immersion monitoring and oil injection control logic, enabling online parameter evolution without manual on-site operation. This ensures continuous high-precision monitoring and adapts to parameter drift issues caused by environmental changes and equipment aging during long-term operation.

[0341] VII. Emergency Response Cases for Abnormal Situations (High Water Level Scenario during Flood Season)

[0342] This case study addresses complex disturbances caused by a rapid rise in reservoir water levels during the flood season, including localized gravity field anomalies and water flow impacts. It demonstrates the system's emergency response and handling capabilities, ensuring reliable monitoring data and timely early warnings. Details are as follows:

[0343] 1. Triggering conditions

[0344] The cloud-based decision-making unit received the reservoir water level rise rate pushed by the hydrological system. Upon receiving the signal and the dynamic benchmark decoupling module detecting a continuous increasing trend in displacement residuals (including residual fluctuations induced by local gravity field anomalies), the cloud-based decision-making unit immediately issued an emergency response instruction to the cognitive decision-making unit, initiating the flood season emergency response process.

[0345] 2. System Response

[0346] (1) Adaptive adjustment of observation frequency: After receiving the “emergency encrypted observation” instruction, the cognitive decision unit adjusts the observation frequency from the default 24 hours / time to 1 hour / time, focusing on capturing the subtle changes in structural displacement under the conditions of high water level and abnormal local gravity field during the flood season, ensuring that abnormal deformation is not missed.

[0347] (2) Control execution layer on standby: The cognitive decision-making unit controls the environment adaptive regulation mechanism to strengthen dehumidification and temperature control, maintain the stability of the indoor environment of the observation room, and avoid high humidity and excessive temperature difference from affecting the accuracy of the equipment; at the same time, the control emergency hard interruption protection mechanism improves the sensitivity of overcurrent monitoring, prevents sudden risks such as lightning strikes and power supply fluctuations during the flood season, and ensures the continuous operation of the system.

[0348] 3. Fault Tolerance Measures

[0349] Due to the turbulent airflow caused by the flood season, the environmental sensing module detected an instantaneous wind speed exceeding 1.0 m / s. The cognitive decision unit automatically executed a "data acquisition lockout," temporarily halting the observation process. Simultaneously, it did not consider closing the ventilation openings using damper actuators (to avoid sudden changes in indoor temperature and humidity, which could further affect equipment stability). Once the environmental sensing module detected the wind speed returning to a safe threshold (≤1.0 m / s), the cognitive decision unit restarted the observation process after a 5-minute delay. This ensured that the observation data was unaffected by airflow disturbances and avoided measurement errors caused by the superposition of local gravity field anomalies and airflow interference.

[0350] 4. Early Warning and Response

[0351] In a subsequent encrypted observation, the cognitive decision-making unit received structurally anomalous residuals from the cloud. The value exceeded the preset envelope of 3 standard deviations (0.30 mm), reaching the warning threshold. The cognitive decision unit immediately initiated the anomaly tracing process, checked the data from the real-time monitoring unit for the eddy current float yaw and wall-collision distance, and confirmed that the float yaw displacement was normal and the wall-collision distance was within acceptable limits. (5mm) to eliminate observation errors caused by float offset (including sway induced by local gravity field anomalies); at the same time, check the environmental quantity sensing data (wind speed 0.3m / s) and optical environment intelligent cognition data (lens cleanliness qualified) to confirm that there are no other environmental and equipment interferences.

[0352] After confirming that the anomaly was structural deformation, the cognitive decision-making unit immediately sent a Level 3 alarm to maintenance personnel via SMS and APP. The alarm information was standardized and complete, specifically including: "Measurement point ID: PLZ-03, deformation rate: 0.12mm / h, abnormal residual: 0.45mm, environmental and equipment interference has been ruled out (wind speed 0.3m / s, lens cleanliness qualified, float status normal), immediate on-site verification is recommended." At the same time, the cognitive decision-making unit simultaneously uploaded the low-light high-speed video evidence from the previous 30 minutes (clearly showing no abnormal shaking of the steel wire, and the intelligent plumb line coordinate instrument working normally) to the cloud digital twin platform. The cloud decision-making unit highlighted the measurement point in the 3D twin scene for remote expert consultation to quickly determine the nature of the anomaly.

[0353] 5. Post-event analysis: After the flood season, the cognitive decision-making unit uploaded the displacement residual sequence (including parameters related to local gravity field anomalies) for the entire flood season to the cloud. The cloud-based dynamic benchmark decoupling module combined long-term historical data for retrospective analysis. After eliminating interference factors such as local gravity field anomalies and environmental loads, it identified an irreversible step jump of about 0.2 mm at the measuring point. It was finally determined to be plastic deformation of the dam foundation rock mass under the above-design water level. This data provided key basis for subsequent dam safety assessment and structural reinforcement optimization, fully demonstrating the monitoring reliability and data value of the system under complex working conditions.

[0354] VIII. Summary of Implementation Results

[0355] The practical application of this invention in a large hydropower station demonstrates that by adopting the inverted plumb line intelligent agent system (modified from existing reusable equipment, without setting up an edge computing gateway, with each module directly connected to the cognitive decision-making unit), the pain points of traditional systems, such as low data reliability, passive operation and maintenance, and insufficient accuracy, are effectively solved. Simultaneously, it also considers adaptability to complex operating conditions such as local gravity field anomalies, achieving the following significant results:

[0356] 1. Significantly improved data reliability: The pass rate of monitoring data has increased from 85% in the traditional system to 99.8%, and the success rate of dual-test self-consistent verification has reached 98.5%. Through multiple mechanisms such as pre-decoupling of steel wire, optical environment control, and buoy sway correction (including sway induced by local gravity field anomalies), "false data for true analysis" caused by unstable steel wire, lens stains, environmental interference, and local gravity field anomalies has been completely eliminated. The direct connection mode without edge computing gateway reduces intermediate links in data transmission, further improving the stability and timeliness of data transmission, and ensuring that each set of monitoring data can truly reflect the structural deformation state.

[0357] 2. Monitoring accuracy reaches sub-millimeter level: the system's overall error is ≤ ±0.1mm, and the spatial resolution is better than 0.01mm. The intelligent vertical coordinate instrument's observations are precisely corrected through a real-time monitoring unit for the eddy current type pontoon sway and wall impact distance, effectively eliminating interference from pontoon offset (including sway induced by local gravity field anomalies). The cognitive decision unit directly completes data processing and correction, avoiding additional errors from the edge gateway and further improving correction accuracy. In practical applications, it successfully captured 0.12mm of instantaneous deformation of the dam body under high water levels during the flood season and 0.08mm of residual displacement after an earthquake, enabling sensitive identification of early, minute structural deformations.

[0358] 3. Significantly improved operational efficiency and substantial cost reduction: The frequency of manual on-site inspections decreased from twice a week to once a quarter, reducing workload by 92%. Predictive maintenance provided early warnings four months in advance of slow jamming caused by aging float seals, preventing data interruptions; and two months in advance of persistent dirt on optical lenses, allowing for planned cleaning and completely changing the traditional reactive "reactive maintenance" model. Overall lifecycle operational costs were reduced by 35%, while eliminating the procurement, installation, and maintenance costs of edge computing gateways, further reducing operational investment and adapting to long-term engineering maintenance needs.

[0359] 4. A qualitative leap in intelligent capabilities: The system achieves fully unattended operation, with monitoring strategies adaptively adjusted according to operating conditions (every 3 days during stable periods, every hour during flood season). It possesses automatic equipment health status diagnosis and predictive maintenance capabilities. For complex operating conditions such as localized gravity field anomalies, the system can automatically adjust monitoring strategies and optimize correction parameters without manual intervention. The retrofit construction did not affect the normal operation of the original plumb line system, achieving a smooth transition between the old and new systems. The design without an edge computing gateway simplifies the hardware deployment process, reduces construction difficulty, and shortens the retrofit cycle.

[0360] 5. Outstanding economic benefits and significant promotional value: Since it reuses existing main mechanical components such as boreholes, protective pipes, anchor blocks, steel wires, and pontoons, and does not require the installation of edge computing gateways, the cost of intelligent transformation of a single system is only 35% of that of a brand-new intelligent plumb line system, which greatly reduces the investment cost of intelligent upgrades and has extremely high economic efficiency and promotional value.

[0361] In summary, this invention deeply integrates engineering structure monitoring theory with IoT, AI, and digital twin technologies, achieving a comprehensive intelligent upgrade without altering existing plumb line devices or setting up edge computing gateways. This simplifies the hardware architecture, reduces costs and maintenance complexity, and effectively adapts to complex conditions such as local gravity field anomalies, redefining the standard for plumb line observation. This invention provides an efficient, accurate, intelligent, and low-cost solution for the safety monitoring of engineering structures such as dams, bridges, high-rise buildings, and landslides, demonstrating significant practicality and broad application value. When used in conjunction with a normal plumb line intelligent body, it can monitor the overall vertical displacement and deflection of engineering structures, further expanding the system's monitoring range and enhancing the comprehensiveness and systematic nature of engineering structure safety monitoring.

[0362] Important Note

[0363] 1) All intelligent upgrades in this embodiment are based on independent external installations without any structural modifications to the original plumb line mechanical devices (drilling, seamless steel pipe protection pipe, top fixing device, plumb line stainless steel wire, float, oil drum, bottom anchoring device, etc.). This ensures the continuity and integrity of the original gravity monitoring benchmark, avoids changes in steel wire tension and float attitude deviation caused by structural modifications, guarantees the original monitoring accuracy and long-term stability of the device, and is compatible with the transformation and upgrading of various existing plumb line systems.

[0364] 2) All newly added sensing modules and control actuators are directly connected to the cognitive decision unit without adding intermediate edge gateways. The synchronization accuracy of data transmission and command issuance is ≤1ms, which fully meets the core requirements of high-precision and long-term continuous monitoring of engineering structures such as water conservancy and landslides. At the same time, the hardware architecture is simplified, interference and delay in the data transmission link are reduced, the complexity of system deployment and operation and maintenance is reduced, and the reliability of data transmission is improved.

[0365] 3) The core logic of local gravity field anomaly detection, multi-source error correction (including float yaw, wire creep, temperature interference, etc.), safety early warning triggering, equipment health diagnosis, and adaptive scheduling of observation strategies are all centrally completed by the cognitive decision-making unit or cloud decision-making unit. The monitoring and perception layer is only responsible for multi-source data acquisition and status feedback. It strictly follows the three-layer closed-loop architecture design of "monitoring and perception - cognitive decision-making - control execution" to ensure the efficiency, reliability and scalability of the system operation.

[0366] 4) The float level adjustment and wire tension compensation are only implemented in an automatic compensation mode using a micro-injection pump for precise oil injection when the buoyancy decay has not reached the hard cut-off threshold. When the oil injection volume reaches the preset limit (10% of the float volume) and the alarm is not eliminated, or when serious abnormalities such as tension exceeding the limit occur, the sampling link is immediately cut off and manual intervention is triggered. No other automatic control mechanisms such as automatic float position adjustment are set up to ensure the safety and accuracy of float attitude and wire tension adjustment, and to avoid equipment damage or monitoring data distortion caused by improper automatic control.

[0367] 5) All new modules and actuators are designed with IP68 or higher protection level to adapt to the harsh working conditions of high humidity, dust and oil in the observation site; the core actuators (emergency hard interruption circuit, micro oil injection pump) are equipped with redundant power supply and control links, with a switching time of ≤100ms, to ensure that the protection action does not fail under extreme working conditions and to adapt to the long-term unattended operation requirements of the inverted plumb line.

[0368] 6) All observation processes, data quality control standards, early warning thresholds, and maintenance strategies of the system strictly follow the industry standards for engineering structure deformation monitoring. The core experience of manual observation is algorithmized and the process is automated, which not only eliminates individual differences in manual operation, but also ensures the standardization and comparability of monitoring data, providing a compliant and reliable data source for engineering structure safety analysis and decision-making.

[0369] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent agent for monitoring structural deformation using a gravity reference and for monitoring horizontal displacement using a plumb line, characterized in that, It includes: The system comprises a monitoring and sensing layer, a cognitive decision-making layer, and a control execution layer. The monitoring and sensing layer, located at the observation site of the plumb line device, includes a displacement sensing module, a state sensing module, a clearance sensing module, and an environmental sensing module. It is used to simultaneously acquire the spatial coordinates and horizontal displacement of the plumb line stainless steel wire, the working status of the float and optical measurement components, the safe clearance between the plumb line stainless steel wire and the protective tube, and observation environmental parameters. The cognitive decision-making layer, deployed on a remote cloud platform, includes a cognitive unit and a decision-making unit. It is used to perform steady-state discrimination, double-return consistency verification, and multi-dimensional object analysis on the multi-source data output by the monitoring and sensing layer. The system includes field correction, dynamic benchmark decoupling, structural deformation prediction, equipment health diagnosis, and maintenance decision-making. The control execution layer is located at the observation site and includes a buoyancy-tension safety interlocking actuator, an environmental adaptive control actuator, an emergency hard interruption protection mechanism, and a CCD lens cleanliness adaptive self-cleaning system. It is used to execute observation interlocking, buoyancy compensation, environmental control, emergency protection, and optical cleaning according to the control commands output by the cognitive decision layer. The monitoring and sensing layer, cognitive decision layer, and control execution layer form a closed-loop system with interconnected data flow, command flow, and status feedback flow through an industrial Internet of Things communication link.

2. The intelligent agent for monitoring engineering structure deformation using a gravity reference and horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The displacement sensing module includes an intelligent vertical coordinate instrument, which employs a dual orthogonal optical imaging structure. This structure includes a first optical imaging system positioned along the engineering axis, a second optical imaging system positioned perpendicular to the engineering axis, a high-resolution linear CCD sensor, and an embedded image processing unit. The transmitting and receiving ends of the first and second optical imaging systems are respectively arranged on opposite sides of the vertical stainless steel wire, used to illuminate the wire with a parallel beam and form a corresponding shadow image on the linear CCD sensor. The embedded image processing unit is configured to perform environmental background subtraction, adaptive exposure compensation, image enhancement, gradient extraction, and sub-pixel edge localization on the shadow image, and calculate its two-dimensional center coordinates and horizontal displacement components relative to the initial reference state or the previous observation period based on the edge coordinates on both sides of the vertical stainless steel wire. The intelligent vertical coordinate instrument retains its existing installation position and is connected to the cognitive decision layer to output displacement data without altering the original vertical mechanical structure.

3. The intelligent agent for monitoring engineering structure deformation using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 2, is characterized in that: The state quantity sensing module includes a vertical non-contact toggle reset and stability control unit, which includes a flexible electromagnetic drive mechanism, a torque detection sensor, a precision linear guide rail, and a PLC control module. The flexible electromagnetic drive mechanism is mounted on the precision linear guide rail and maintains a non-contact gap with the vertical stainless steel wire. The torque detection sensor is connected in series in the transmission link of the flexible electromagnetic drive mechanism. The PLC control module is communicatively connected to the flexible electromagnetic drive mechanism, the torque detection sensor, and the intelligent vertical coordinate instrument. The PLC control module is configured to control the flexible electromagnetic drive mechanism to apply a flexible disturbance with a preset stroke and preset torque to the vertical stainless steel wire before effective displacement acquisition, and then remove the disturbance force to allow the vertical stainless steel wire to freely reset under the buoyancy of the float and the liquid damping. The module then determines whether the wire has reached a stability threshold based on the coordinate sequence continuously output by the intelligent vertical coordinate instrument. When the feedback torque exceeds the safety threshold, the PLC control module controls the flexible electromagnetic drive mechanism to stop and retract, and ensures that the toggle action, reset action and effective sampling action do not occur at the same time through timing mutual exclusion logic.

4. The intelligent agent for monitoring engineering structure deformation using a gravity reference and horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The state quantity sensing module further includes a real-time monitoring unit for the sway and wall-collision distance of the eddy current type float and a real-time monitoring unit for the immersion depth change of the magnetostrictive type float. The real-time monitoring unit for the sway and wall-collision distance of the eddy current type float includes four eddy current probes orthogonally and symmetrically arranged at 0°, 90°, 180°, and 270°, and a vector calculation module. The four eddy current probes are circumferentially arranged around the float and the oil tank to synchronously acquire four channels of gap data. The vector calculation module is used to calculate the two-dimensional sway displacement vector of the float based on the differential data relative to the eddy current probes. The system measures the minimum micro-distance between the float and the oil drum wall, and generates a float near-wall alarm when the minimum micro-distance is lower than a preset alarm threshold. The real-time monitoring unit for the change in the immersion depth of the magnetostrictive float includes a waveguide wire, a float magnetic ring rigidly connected to the float and rising and falling synchronously with the float, and a pulse detection circuit. The pulse detection circuit is used to calculate the position change of the float magnetic ring and the immersion depth of the float based on the propagation time difference between the current pulse and the torsional wave, and outputs a buoyancy attenuation alarm when the immersion depth of the float is insufficient, providing state parameters for judging the tension state of the steel wire and correcting the displacement data.

5. The intelligent agent for monitoring engineering structure deformation using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 2, is characterized in that: The state quantity sensing module also includes a CCD lens surface cleanliness online monitoring unit and an intelligent vertical coordinate instrument working status visual monitoring unit. The CCD lens surface cleanliness online monitoring unit includes a high-contrast checkerboard target set outside the optical system window and a cleanliness assessment unit that reuses the intelligent vertical coordinate instrument image processing unit. The cleanliness assessment unit is used to acquire images of the checkerboard target during non-effective sampling periods, form a lens cleanliness assessment value based on the gradient energy of the target image, and output an optical degradation warning and acquisition lockout signal when the lens cleanliness assessment value is lower than the health threshold. The intelligent vertical coordinate instrument working status visual monitoring unit includes a low-light CMOS image sensor, an infrared fill light, and an edge preprocessing module. The edge preprocessing module is used to perform region of interest extraction and inter-frame difference on the optical lens and wire observation area of ​​the intelligent vertical coordinate instrument. When receiving a measurement data abnormality signal or a reset action signal, it switches to high-speed recording mode and saves video frames within a preset time window before and after the abnormal event. The video frames are bound to the corresponding displacement data, equipment status label, and warning information through event identifiers.

6. The intelligent agent for monitoring the deformation of engineering structures using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 3, is characterized in that: The clearance sensing module includes a four-dimensional capacitive steel wire tube internal position and clearance sensing unit. This unit comprises four capacitance probes orthogonally and symmetrically arranged at 90° on the inner wall of the critical section of the protective tube, a multi-channel capacitance detection circuit, and a spatial calculation module. The sensing electrodes of the four capacitance probes and the vertical stainless steel wire each form an independent capacitance measurement channel. The multi-channel capacitance detection circuit synchronously acquires the capacitance values ​​of each capacitance measurement channel and obtains the net distance data in four directions based on a pre-calibrated correspondence between capacitance values ​​and electrode spacing. The spatial calculation module... The module is used to construct an overdetermined system of equations based on the net distance data in four directions. It solves the two-dimensional center coordinates of the vertical stainless steel wire in the cross-section of the protective tube and the minimum net distance from the vertical stainless steel wire to the inner wall of the protective tube by center fitting. When the measurement residual of a single capacitive probe exceeds the anomaly rejection threshold, the data of that capacitive probe is rejected and the remaining data is used to re-fit the center. The spatial calculation module is also used to output a near-wall warning signal or a contact alarm signal based on the first-level near-wall warning threshold and the second-level contact alarm threshold, and to link the vertical non-contact toggle reset and stability control unit or the emergency hard interruption protection mechanism.

7. The intelligent agent for monitoring engineering structure deformation using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The environmental quantity sensing module includes a five-element weather station integrating a temperature sensor, humidity sensor, air pressure sensor, ultrasonic anemometer, and ultrasonic wind direction sensor. This five-element weather station is connected to the data bus of the monitoring and sensing layer via a digital interface, used to synchronously output temperature, humidity, air pressure, wind speed, and wind direction parameters. The ultrasonic anemometer and ultrasonic wind direction sensor adopt a solid-state structure without mechanical rotating parts and calculate the two-dimensional wind speed component, instantaneous wind speed, and wind direction angle based on the time difference of sound wave propagation in orthogonal directions. The environmental quantity sensing module is configured to send temperature and air pressure parameters to the displacement and state sensing modules for environmental compensation of the raw measurement data, and to generate a disturbed state identifier and acquire a lockout signal based on the sliding average wind speed and instantaneous wind speed within a preset time window. The displacement, state, clearance, and environmental quantity sensing modules are connected to the same EtherCAT industrial bus and time-stamp aligned using a PPS clock synchronization signal to form a structured multi-dimensional sensing data package containing wire coordinates, equipment status, boundary clearance, environmental parameters, and visual evidence indexes.

8. The intelligent agent for monitoring the deformation of engineering structures using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The cognitive unit includes a wire pre-decoupling and dual-return intelligent acquisition module, a multi-dimensional physical field real-time correction engine, an optical environment intelligent cognitive module, and an adaptive observation strategy engine. The wire pre-decoupling and dual-return intelligent acquisition module controls the vertical stainless steel wire to sequentially complete pre-disturbance, free reset, forced static placement, visual and coordinate dual steady-state determination, and two coordinate acquisitions. It outputs effective displacement data or a re-measurement command based on the comparison between the coordinate difference of the two re-measurements and the limit threshold. The multi-dimensional physical field real-time correction engine extracts ambient temperature, buoy immersion depth, buoy two-dimensional yaw displacement vector, and original effective displacement data, and then... The immersion depth of the pontoon indirectly calculates the wire tension and progressively corrects the time-varying creep error of the wire and the pontoon yaw coupling error. The optical environment intelligent cognition module receives the lens cleanliness assessment value output by the state quantity sensing module, determines the data confidence level based on the lens cleanliness assessment value, and outputs a cleaning trigger command when the optical channel deteriorates. The adaptive observation strategy engine adjusts the observation period and outputs a working condition lockout signal or triggers a benchmark review based on the minimum wire clearance output by the limit quantity sensing module, the wind speed, temperature change rate, load change rate, earthquake trigger signal, and physical reference drift alarm output by the environmental quantity sensing module.

9. The intelligent agent for monitoring the deformation of engineering structures using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The decision-making unit includes a data storage and model evolution module, a safety boundary digital twin mapping module, a multimodal fusion prediction module, a dynamic benchmark decoupling module, and a predictive maintenance scheduling module. The data storage and model evolution module aggregates high-reliability correction displacement data, environmental factor data, equipment status data, visual evidence indexes, early warning tags, and action logs, and performs offline identification and online updates of the physical correction model parameters. The safety boundary digital twin mapping module maps the two-dimensional center coordinates of the vertical stainless steel wire, the cross-sectional coordinates of the protective pipe, and the inner wall radius of the protective pipe to the three-dimensional digital twin model of the engineering structure. The system forms a full-section net distance distribution and risk section identification between the vertical stainless steel wire and the protective pipe; the multi-modal fusion prediction module is used to predict the structural deformation trend and equipment health degradation trajectory based on displacement, environment, load, pontoon sway and equipment status characteristics; the dynamic benchmark decoupling module is used to establish a dynamic prediction benchmark based on ambient temperature, structural load, local gravity field anomaly parameters and historical displacement data, and separate structural anomaly residuals and physical benchmark drift from high-confidence corrected displacement data; the predictive maintenance scheduling module is used to generate maintenance work orders based on equipment health, remaining effective life and fault tags.

10. The intelligent agent for monitoring the deformation of engineering structures using a gravity reference and for monitoring horizontal displacement using a plumb line, as described in claim 1, is characterized in that: The buoyancy-tension safety interlock actuator includes a micro-injection pump, oil tank, filter, electromagnetic control valve, and sampling link hard-cut-off component. It is used to replenish transformer oil to the float when the immersion depth is insufficient and the hard-cut-off condition is not met, and to cut off the observation trigger circuit when the wire tension exceeds the limit or buoyancy compensation fails. The environmental adaptive control actuator includes a dehumidifying fan, heating rod, and semiconductor temperature control device, used to adjust the observation environment according to the environmental exceedance interlock signal. The emergency hard interruption protection mechanism includes an independent hardware control circuit, overcurrent detection module, and hardware relay, used to prevent overcurrent when the vertical stainless steel wire contacts the protection tube or equipment. In case of overcurrent or overload, the power supply to non-core peripherals is cut off, and an emergency status is reported through a backup communication link. The CCD lens cleanliness adaptive self-cleaning system includes an oil-resistant flexible scraper driven by a stepper motor and an air curtain blowing component, which is used to perform air curtain blowing or scraping cleaning according to the degree of contamination, and trigger lens cleanliness re-evaluation after cleaning is completed. Each actuator is set with control priority in the order of emergency hard interruption protection mechanism, buoyancy-tension safety interlock actuator, CCD lens cleanliness adaptive self-cleaning system, and environmental adaptive control actuator, and forms an execution closed loop through hardware interlock, status feedback and monitoring and sensing layer verification.

Images