Remote control and supervision system for infrastructure machinery based on operation condition analysis

Abstract

The invention relates to the technical field of infrastructure machinery supervision, and discloses an infrastructure machinery remote control and supervision system based on operation condition analysis. The system comprises an operation data real-time acquisition module, a working condition characteristic dynamic extraction module, an abnormal state identification module, a control strategy self-adaptive generation module and a supervision report comprehensive generation module. The operation data real-time acquisition module is used for acquiring a mechanical position coordinate sequence, vibration spectrum distribution and load pressure fluctuation by virtue of a heterogeneous sensor network; the working condition feature dynamic extraction module outputs feature vectors containing time domain stability, frequency domain energy distribution and load change gradient through multi-scale fusion processing; the abnormal state recognition module performs multi-dimensional comparison on the vector and a reference working condition model to generate an abnormal state indication matrix; the control strategy self-adaptive generation module configures remote control parameters accordingly; and the supervision report comprehensive generation module integrates the data to generate a structured report and assists infrastructure machinery supervision optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrastructure machinery supervision, in particular to an infrastructure machinery remote control and supervision system based on operation condition analysis. BACKGROUND

[0002] In the field of infrastructure engineering, the stable operation and effective supervision of machinery are directly related to the progress of the project and the safety of the operation. With the continuous expansion of infrastructure projects, the operation scene of machinery is becoming increasingly complex, and the demand for remote control and supervision is becoming more urgent. At present, the monitoring method of infrastructure machinery mainly relies on single sensor data collection, which can only obtain position information or simple operation parameters, and it is difficult to fully reflect the operation condition of the machinery. For example, some systems only track the position of the machinery through GPS positioning, and cannot capture key data such as vibration spectrum distribution and load pressure fluctuation, resulting in one-sidedness in judging the operation state of the machinery. At the same time, the existing technology has a single means of extracting working condition characteristics, mostly staying at the time domain analysis level, lacking comprehensive consideration of frequency energy distribution and load change gradient, and being difficult to accurately capture subtle changes in the operation process of the machinery, and thus unable to timely discover potential abnormal hidden dangers. In the abnormal state recognition link, the traditional method usually uses a single threshold for judgment, and cannot be compared with the preset baseline working condition model in multiple dimensions, which is easy to cause misjudgment or omission, affecting the accuracy of supervision. The current configuration of remote control parameters is mostly in a fixed mode, which cannot be adaptively adjusted according to the actual abnormal state of the machinery. When the machinery appears different types of abnormalities, it is difficult to take targeted control measures, resulting in poor control effect. The generation of supervision reports is also limited to the simple listing of data, lacking integrated analysis of abnormal states and control parameter execution, and being unable to provide comprehensive and structured reference information for subsequent project management. These problems have restricted the improvement of the level of remote control and supervision of infrastructure machinery. SUMMARY

[0003] The purpose of the present application is to provide an infrastructure machinery remote control and supervision system based on operation condition analysis to solve the problems raised in the background.

[0004] To achieve the above purpose, the present application provides an infrastructure machinery remote control and supervision system based on operation condition analysis, which comprises: An operation data real-time acquisition module, based on a heterogeneous sensor network deployed on the infrastructure machinery, acquires operation condition data in real time, including a sequence of mechanical position coordinates, vibration spectrum distribution and load pressure fluctuation; An operation condition characteristic dynamic extraction module inputs the operation condition data into a multi-scale fusion processing unit to output an operation condition characteristic vector containing time domain stability features, frequency energy distribution features and load change gradient; An abnormal state identification module compares the working condition feature vector with a preset reference working condition model in multiple dimensions to generate an abnormal state indication matrix; A control strategy adaptive generation module configures remote control parameters, including enabling a high-precision positioning tracking mode and applying dynamic load adjustment testing, based on the abnormal state indication matrix; A regulatory report comprehensive generation module integrates the abnormal state indication matrix and control parameter execution logs to generate a structured regulatory report.

[0005] Preferably, the operation data real-time acquisition module specifically includes: A sensor data synchronization unit coordinates the sampling frequencies of the heterogeneous sensor network through an edge computing node to ensure that the timestamps of the mechanical position coordinate sequence, the vibration frequency spectrum distribution and the load pressure fluctuation are aligned; A data quality enhancement unit performs environmental noise suppression processing on the real-time acquired operation working condition data to eliminate measurement deviations caused by temperature and humidity interference; A preprocessed data output unit converts the processed operation working condition data into a standardized data stream and inputs it to the working condition feature dynamic extraction module.

[0006] Preferably, the working condition feature dynamic extraction module specifically includes: A time-domain stability feature calculation unit performs window sliding analysis on the mechanical position coordinate sequence in the standardized data stream to calculate the position drift accumulation and the speed mutation frequency, and generates a time-domain stability feature vector; A frequency domain energy distribution feature extraction unit performs a frequency band decomposition algorithm on the vibration frequency spectrum distribution in the standardized data stream to quantify the energy density ratio in the preset sensitive frequency band and construct a frequency domain energy distribution feature vector; A load change gradient evaluation unit calculates the slope distribution and interval dispersion of the load amplitude mutation point based on the load pressure fluctuation in the standardized data stream, and generates a load change gradient vector; A feature fusion processing unit tensor splices the time-domain stability feature vector, the frequency domain energy distribution feature vector and the load change gradient vector, and outputs the working condition feature vector through normalized weighted fusion.

[0007] Preferably, the abnormal state identification module specifically includes: A reference model construction unit trains a reference working condition model based on historical normal operation working condition data, and the reference working condition model includes a position stability threshold, a vibration energy reference spectrum and a load pressure standard range; The multi-dimensional comparison execution unit aligns the working condition feature vector with the benchmark working condition model by dimension, calculates the position drift cumulative deviation of the time domain stability feature, the energy density deviation of the frequency domain energy distribution feature, and the slope distribution difference of the load change gradient, and generates an initial abnormal index set; The abnormal state indication matrix generation unit performs importance weighting processing on the initial abnormal index set to eliminate the dimension influence, and outputs a three-order abnormal state indication matrix with dimensions of [machine number x timestamp x abnormal type].

[0008] Preferably, the control strategy adaptive generation module specifically includes: The high-precision positioning tracking mode configuration enables the high-precision positioning tracking mode when the position drift cumulative deviation in the abnormal state indication matrix exceeds the preset deviation threshold, and improves the position sampling frequency to 3-5 times the original frequency. The dynamic load adjustment test execution applies a multi-stage load pressure disturbance sequence to the mechanical node with the largest slope distribution difference of the load change gradient in the abnormal state indication matrix. The control parameter output unit encapsulates the configured remote control parameters into a control instruction set and inputs it to the remote execution interface.

[0009] Preferably, the dynamic load adjustment test execution specifically includes: The load disturbance injection unit generates a multi-stage load pressure disturbance sequence including stepwise pressure increments and pulsed pressure fluctuations through a controllable actuator. The response data capture unit records the mechanical vibration response spectrum and position offset after the load pressure disturbance sequence is injected. The abnormal correlation determination unit compares the measured vibration response spectrum with the theoretical response spectrum to calculate the load abnormal correlation factor.

[0010] Preferably, the regulatory report comprehensive generation module specifically includes: The data integration unit extracts the abnormal type distribution data in the abnormal state indication matrix and aligns it with the mode switching records and test results in the control parameter execution log on the time axis. The structured template mapping unit maps the integrated data to the preset report template fields, including the abnormal type distribution field, the control execution frequency field, and the test result field. The regulatory report generation unit generates the structured regulatory report containing a multi-dimensional data table based on the mapping result.

[0011] Preferably, the regulatory report comprehensive generation module further includes: The risk level marking unit calculates a comprehensive risk score of each mechanical node according to the abnormal type distribution data in the abnormal state indication matrix, and divides a risk level interval; The report optimization unit performs information level sorting processing on the structured supervision report based on the comprehensive risk score.

[0012] Preferably, the system further comprises a remote execution feedback module: The control instruction transmission unit sends the control instruction set output by the control strategy adaptive generation module to the infrastructure mechanical control terminal; The execution state monitoring unit captures mechanical response data in the control instruction execution process in real time; The feedback data backflow unit inputs the mechanical response data into the working condition characteristic dynamic extraction module.

[0013] Preferably, the remote execution feedback module specifically comprises: The response data analysis unit performs feature extraction processing on the captured mechanical response data to generate a response feature vector; The strategy adjustment triggering unit triggers the reconfiguration of the control strategy adaptive generation module when the response feature vector deviates from the expected response model by more than a preset adjustment threshold.

[0014] Compared with the prior art, the present application has the following advantages: By deploying a heterogeneous sensor network to realize real-time collection of operating condition data, compared with the traditional single sensor collection method, multi-dimensional data such as mechanical position coordinate sequence, vibration frequency spectrum distribution and load pressure fluctuation can be obtained at the same time, which can more comprehensively reflect various state information in the mechanical operation process, so that the supervisor has a more complete understanding of the mechanical operating condition, and avoids the supervision blind spot caused by data loss. In the aspect of working condition characteristic extraction, the system adopts a multi-scale fusion processing unit to output a working condition characteristic vector containing time domain stability features, frequency domain energy distribution features and load change gradient, breaking through the limitation of traditional single time domain analysis and realizing multi-dimensional mining of working condition characteristics. This multi-scale fusion extraction method can accurately capture the subtle changes in the mechanical operation process, whether it is stability fluctuation in the time domain, energy distribution difference in the frequency domain, or gradient trend of load change, which can be effectively extracted, thereby providing more abundant and accurate feature basis for subsequent abnormal state recognition.

[0015] The abnormal state recognition module compares the working condition characteristic vector with a preset reference working condition model in multiple dimensions to generate an abnormal state indication matrix, which changes the mode of traditional single threshold judgment. The multi-dimensional comparison mode can comprehensively consider the difference between the mechanical running state and the reference model from multiple dimensions, effectively reduces the probability of misjudgment and missed judgment, makes the abnormal state recognition more accurate and reliable, and facilitates the supervisor to timely and accurately master whether the machine has an abnormality and the specific type and degree of the abnormality. The control strategy adaptive generation module configures remote control parameters based on the abnormal state indication matrix, including enabling a high-precision positioning tracking mode and applying a dynamic load adjustment test, which breaks the shackles of the traditional fixed control parameter mode. According to different abnormal state indication information, the system can automatically adjust the control parameters and take corresponding control measures for different types of abnormalities, so that the remote control is more targeted and flexible. When the machine has an abnormality, the control parameters can be adjusted quickly and effectively to alleviate or solve the abnormal problem, thereby ensuring the stable operation of the machine. The supervision report comprehensive generation module integrates the abnormal state indication matrix and the control parameter execution log to generate a structured supervision report. Compared with the traditional simple data list, the structured report can clearly present the specific situation of the abnormal state, the execution process and effect of the control parameters, and organically combine the two to provide more comprehensive and orderly supervision information for the engineering management personnel, facilitate the management personnel to clearly understand the overall situation of the mechanical operation supervision, and provide more valuable reference for engineering management decision-making, thereby further optimizing the management process of the infrastructure project. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A timing diagram of the infrastructure machine remote control and supervision system based on running working condition analysis according to the present application; Figure 2 A flowchart of the running data real-time acquisition module; Figure 3 A flowchart of the working condition characteristic dynamic extraction module; Figure 4 A flowchart of the dynamic load adjustment test execution work; Figure 5 A flowchart of the supervision report comprehensive generation module. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Please refer toFigure 1 The application provides a capital construction machinery remote control and supervision system based on operation condition analysis, which comprises the cooperative work of an operation data real-time acquisition module, a working condition characteristic dynamic extraction module, an abnormal state recognition module, a control strategy adaptive generation module and a supervision report comprehensive generation module.

[0019] The operation data real-time acquisition module acquires operation condition data such as mechanical position coordinate sequence, vibration frequency spectrum distribution and load pressure fluctuation in real time through a heterogeneous sensor network deployed on the capital construction machinery. The working condition characteristic dynamic extraction module receives these data and outputs a working condition characteristic vector containing time domain stability characteristics, frequency domain energy distribution characteristics and load change gradient through a multi-scale fusion processing unit. The abnormal state recognition module compares the working condition characteristic vector with a preset reference working condition model in multiple dimensions to generate an abnormal state indication matrix. The control strategy adaptive generation module configures remote control parameters such as enabling a high-precision positioning tracking mode or applying a dynamic load adjustment test based on the matrix. The supervision report comprehensive generation module integrates the abnormal state indication matrix and control parameter execution logs to generate a structured supervision report. The entire system realizes a closed-loop process of data acquisition, feature extraction, abnormality recognition, control adjustment and report generation through modular design, ensuring comprehensive monitoring and remote management of the operation state of the capital construction machinery.

[0020] Embodiment 1 : refer to Figure 2 Focusing on the technical implementation of the operation data real-time acquisition module, the module completes the capture and preprocessing of raw working condition data through a heterogeneous sensor network distributed in key parts of the capital construction machinery. In implementation, multiple types of sensors are deployed on the main structure, power unit and work execution mechanism of each capital construction machinery, and specific configurations include a high-dynamic positioning module, a three-axis vibration sensor array and a distributed hydraulic pressure sensor. The positioning module uses real-time dynamic carrier phase difference technology and is fixed at the center of the mechanical rotation to output high-precision coordinate data of latitude and longitude at a frequency of 5-10 times per second; the vibration sensor array is arranged at the engine base, transmission shaft support point and work arm joint, and each channel captures wideband vibration waveform at a sampling rate of not less than 20 kHz; the hydraulic pressure sensor is embedded in the inlet and return oil pipelines of the execution cylinder to record millisecond-level pressure fluctuations through an industrial strain gauge. All sensor nodes are connected to the edge computing node of the mechanical body through an industrial Ethernet or CAN bus.

[0021] The edge computing node is built-in multi-channel data acquisition card and time synchronization coprocessor, which implements the function of sensor data synchronization unit. The time synchronization protocol is based on IEEE1588 precision clock synchronization standard to build master-slave clock architecture, and the edge node acts as the master clock to broadcast time synchronization frame to all sensors regularly. When the vibration sensor triggers sampling at a certain time, the positioning module automatically delays or advances the sampling period by no more than 0.5 ms to ensure timestamp alignment. The data packet is marked with coordinated universal time by hardware timestamp mechanism, and the linear interpolation algorithm is used to complete the missing points of time series for load pressure fluctuation data. The hardware level is equipped with a constant temperature crystal oscillator module to maintain the clock stability within ±0.1 ppm, eliminating the time reference offset caused by temperature drift.

[0022] The data quality enhancement unit runs in the real-time operating system task of the edge computing node. When the original data stream enters the double-buffer queue, multi-level noise suppression processing is performed in parallel: for positioning coordinate data, an improved adaptive Kalman filter is used to dynamically adjust the process noise covariance matrix parameters according to the mechanical motion state. When the machine is in straight line motion, the constraint weight of the state transition model is strengthened, and when the machine is turning, the prediction error tolerance range is expanded. Wavelet packet decomposition and reconstruction method is used for vibration data processing to reduce noise, db8 wavelet basis is selected to decompose the original waveform for 8 layers, and threshold quantization processing is performed on the sub-band coefficients containing mechanical characteristic frequency band to eliminate electromagnetic interference and structural noise above 10 kHz. The hydraulic pressure data is corrected by the environment compensation unit, which real-time collects the readings of the environment temperature sensor and humidity sensor and inputs them into the compensation model. The model establishes a nonlinear mapping relationship between pressure drift and temperature and humidity based on polynomial regression, and directly applies the compensation coefficient to the original pressure value.

[0023] The preprocessed data output unit establishes a standardized data pipeline mechanism. The cleaned working condition data is converted into a structure array format, each data point contains a timestamp field (nanosecond level precision ISO8601 format), a device identifier, and a joint structure containing three types of working condition data. Positioning coordinates are stored as double-precision floating-point planar rectangular coordinates or geodetic coordinates; vibration frequency spectrum distribution is stored in 16-bit integer array form for 4096-point FFT results; load pressure fluctuation is stored in differential encoding for millipascal level difference value sequence. The check mechanism is designed as two levels of redundancy: each frame of data is attached with CRC-32 check code, and a data block check package containing SHA-256 digest is output every 10 seconds. The data stream interface uses a serialized transmission protocol, which is pushed to the working condition feature dynamic extraction module through the ZeroMQ publish-subscribe mode, and automatically switches to Thrift binary compression format transmission when the bandwidth resource is limited.

[0024] The hardware platform selects an industrial-grade embedded system, carries a quad-core ARM Cortex-A72 processor and configures a real-time performance optimization kernel patch. The heat dissipation scheme adopts a fanless heat pipe design, and the operating temperature range covers the industrial environment requirements of-40°C to 85°C. The power management unit supports mechanical battery direct connection, integrates overvoltage protection and surge current elimination circuit, and cooperates with the super capacitor to realize 20 millisecond accidental power-off data cache protection. All algorithm firmware is deployed in a static allocated real-time task partition, which ensures the completion of the whole process of sensor data collection to output within 20 milliseconds. The hardware interface protection reaches IP67 level, and the signal input port is configured with a TVS diode array to resist 2kV electric fast transient pulse interference.

[0025] The physical deployment of the sensor network follows the optimization of modal analysis results. In a large excavator prototype, vibration sensors are installed on the stress concentration areas of the boom and stick through magnetic bases. The sensor arrangement density is determined based on the work load distribution map, and collection points are set every 0.5 meters in areas with large stress gradient changes. The positioning module installation reference point obtains the centroid coordinates through mechanical design drawings, and uses a laser assisted centering instrument to calibrate the installation deviation to within ±3mm. The hydraulic pressure sensor selects a straight-through oilway installation site, and uses an impact isolation structure with a damping hole in the pipeline after the main control valve to prevent damage to the sensor probe. The cable laying strictly distinguishes the power line and signal line channels, and uses shielded twisted pair to implement 360 degree full wrapping grounding treatment. The hardware system can continuously operate for not less than 8000 hours under the working condition of vibration intensity 5.5mm / s.

[0026] The software architecture of the edge node adopts a microkernel real-time operating system, and the time synchronization service runs in a priority 0 thread in the kernel space, directly driving the hardware clock counter to perform cross-node clock correction. Data processing tasks are divided into three independent processes: the positioning data optimization process occupies 20% of the CPU time slice, the vibration signal processing occupies 35% of the time slice, the pressure compensation and verification task occupies 25% of the time slice, and the remaining resources are used for system monitoring. The memory configuration fixes the sensor data buffer in a physically continuous area, and disables the page exchange function through the MMU to ensure real-time performance. The watchdog circuit is designed as a two-level monitoring architecture, with the hardware watchdog checking the software heartbeat packet every second, and the application layer daemon process polling the task state identifier every 200 milliseconds. All configuration parameters are stored in FRAM memory, which can maintain 5 years of configuration data integrity in the case of accidental power failure. When the system enters maintenance mode, the USB debugging interface can output the raw data mirror file for fault backtracking analysis.

[0027] Example 2: see Figure 3The embodiments relate to the technical implementation of the working condition feature dynamic extraction module and the abnormal state identification module. The two modules work together to convert the pre-processed standardized data stream into a feature vector with clear engineering significance, and identify abnormal states through comparison with the benchmark model.

[0028] The working condition feature dynamic extraction module is deployed on the data processing server of the edge computing node. The server is configured with a multi-core Xeon processor and 128 GB DDR4 memory, and is equipped with a dedicated tensor calculation acceleration card. The module receives the standardized data stream from the pre-processing data output unit, extracts three types of features through four parallel processing channels. The time domain stability feature calculation channel opens an independent memory pool to cache the mechanical position coordinate sequence, and uses a sliding window mechanism to process the data stream. The window width is dynamically adjusted according to the type of the machine, and is set to 5 seconds for mobile machines and 30 seconds for fixed machines. The position drift accumulation is calculated within each window, and the trajectory stability index is obtained by integrating the Euclidean distance between consecutive coordinate points. The speed mutation frequency is calculated by the difference method between adjacent time points, and when the change rate exceeds the set threshold, it is recorded as a mutation event. The time domain feature vector containing the drift amount and mutation frequency is finally output.

[0029] The frequency domain energy distribution feature extraction channel is designed to process vibration spectrum data. The channel is configured with a real-time frequency band decomposition engine to divide the input 4096-point FFT spectrum data into multiple sensitive frequency bands. The frequency band division scheme is pre-set based on the inherent characteristics of the mechanical transmission system, usually including the rotor passing frequency band, gear meshing frequency band, and bearing characteristic frequency band. The energy density ratio of each frequency band is calculated, which is the ratio of the sum of the square of the spectrum amplitude in the frequency band to the total frequency band energy. The calculation process uses SIMD instruction set to process multiple frequency bands in parallel, and the final frequency domain feature vector contains the energy distribution quantization value of each sensitive frequency band.

[0030] The load change gradient evaluation channel is designed for hydraulic pressure fluctuation data. The channel first detects the load amplitude mutation point by finding the extreme point of the first derivative of the pressure time series. The slope distribution is calculated for each mutation point, and the slope value is determined by linear fitting of the left and right 5 sampling points. The interval dispersion degree is calculated by calculating the standard deviation and skewness of the interval time. The final generated load change gradient vector contains the slope distribution statistical characteristics and interval dispersion index.

[0031] The feature fusion processing unit receives the above three feature vectors and performs a tensor splicing operation to splice the time domain stability feature vector, the frequency domain energy distribution feature vector and the load change gradient vector in the feature dimension to form a high-order tensor structure. In the normalization and weighted fusion process, the weight coefficients of each feature dimension are dynamically adjusted based on the feature importance. The weight adjustment algorithm adopts an optimization method based on gradient descent, and dynamically updates the weight coefficients according to the feedback of the subsequent abnormality identification module. The finally output working condition feature vector is a floating point array of fixed dimension, which is convenient for subsequent module processing.

[0032] The abnormal state identification module runs on a distributed computing cluster, which includes multiple computing nodes, each equipped with a dual-GPU computing card. The benchmark model construction unit trains the benchmark working condition model based on historical normal operation data. The training data comes from the working condition feature vector samples collected under the normal operation state of the machine, and a multi-dimensional benchmark range is established through an unsupervised learning algorithm. The position stability threshold is determined by the distribution characteristics of the accumulated amount of position drift in the historical data, and the normal range boundary is set by using the percentile method. The vibration energy benchmark spectrum is analyzed by clustering algorithm to establish the standard distribution mode of energy density in each frequency band. The load pressure standard range is modeled based on Gaussian mixture model to determine the reasonable fluctuation range of each gradient index.

[0033] The multi-dimensional comparison execution unit adopts a streaming processing architecture and receives the working condition feature vector stream in real time. The comparison process is performed according to the alignment of the feature dimensions: the time domain stability feature dimension calculates the deviation of the current feature vector from the position stability threshold in the benchmark model, and the Mahalanobis distance is used to measure the abnormality degree between multiple variables; the frequency domain energy distribution feature dimension calculates the correlation coefficient of the energy density of each frequency band with the benchmark spectrum, and quantifies the distribution form difference through the spectral entropy value; the load change gradient dimension evaluates the abnormality by calculating the probability distance between the current gradient vector and the center of the Gaussian mixture model. The comparison results of all dimensions are combined to form an initial abnormality index set.

[0034] The abnormal state indication matrix generation unit performs standardization processing on the initial abnormality index set, and the importance weighting coefficients are dynamically configured according to the type of machine and working condition environment. For precision working machines, the position stability weight is increased, and for heavy load machines, the load change gradient index is emphasized. Dimensionless elimination adopts Z-score standardization method to convert each abnormality index into a distribution with mean value of 0 and standard deviation of 1. The finally generated abnormal state indication matrix is a three-dimensional array structure, the first dimension index is the machine number, the second dimension records the timestamp sequence, and the third dimension identifies the abnormal type code. The matrix element value is the standardized abnormality index value, and the value size directly reflects the severity of the abnormality. The matrix is stored through a distributed memory database, supporting real-time query and historical backtracking analysis.

[0035] The data flow of the entire module adopts a pipeline architecture design, and a bidirectional feedback mechanism is established between the feature extraction and abnormality identification stages. When the abnormality identification module detects a specific type of abnormal pattern, the sensitive frequency band settings or feature weight distribution of the feature extraction module can be adjusted in reverse. All algorithm implementations use C++ to write the core computing part, Python to encapsulate business logic, and Apache Kafka to realize data flow between modules. The computing nodes are interconnected through the RDMA high-speed network to ensure low-latency performance for large-scale data exchange. The module deployment is containerized and encapsulated, supporting elastic scaling to meet the monitoring needs of mechanical clusters of different sizes.

[0036] Embodiment 3: see Figure 4 The embodiments relate to the complete technical implementation of a control strategy adaptive generation module that dynamically configures remote control parameters, including the activation of high-precision positioning tracking mode and the execution of dynamic load adjustment tests, based on the analysis results of the abnormal state indication matrix. The system is deployed on an industrial control server cluster with a high-availability architecture, using a master-slave node hot switchover mechanism to ensure service continuity.

[0037] The configuration logic of the high-precision positioning tracking mode is based on real-time monitoring of the position drift cumulative amount deviation in the abnormal state indication matrix. When the system detects that the position drift cumulative amount deviation value of a certain mechanical node exceeds the preset deviation threshold, the mode switching process is triggered. The deviation threshold is determined through historical data analysis and uses a dynamic adjustment mechanism, with the formula being:

[0038] wherein, represents the dynamic deviation threshold, represents the mean of the historical position drift amount, represents the standard deviation of the historical position drift amount, is the environmental adaptation coefficient (value range 0.1-0.3), represents the current continuous running time, represents the maximum allowed continuous running time. This formula realizes the mechanism of adaptively increasing the threshold as the running time increases.

[0039] When the trigger condition is met, the system sends a mode switching instruction to the mechanical control terminal, increasing the positioning sampling frequency to 3-5 times the original frequency. For mechanical devices using Beidou / GPS dual-mode positioning, the system simultaneously activates the carrier phase differential positioning technology, improving the positioning accuracy from meters to centimeters through base station correction signals. The position data output format is converted to a complex data structure containing latitude, longitude, elevation, attitude angle, and precision factor, and the sampling interval is shortened from 1 second in the regular mode to 200-300 milliseconds.

[0040] The execution of the dynamic load regulation test targets the mechanical nodes with the largest slope distribution difference of the load change gradient in the abnormal state indication matrix. The system first generates a multi-stage load pressure disturbance sequence through a load disturbance injection unit composed of an electro-hydraulic proportional valve group and a pressure servo controller. The stepwise pressure increment test starts from 50% of the rated pressure and gradually increases to 110% of the rated pressure with a 5% increment at each stage, and each stage pressure maintains for 60 seconds. The pulse pressure fluctuation test generates a square wave signal with an amplitude of ±15% of the rated pressure, a pulse width of 100 milliseconds, a repetition frequency of 2Hz, and a duration of 30 seconds.

[0041] The response data capture unit synchronously collects mechanical vibration response spectra and position offsets during load disturbance injection. The vibration monitoring uses ICP type acceleration sensors installed at key positions of the actuator, with a sampling frequency of 51.2kHz, and obtains vibration data in the 0-20kHz frequency band after anti-aliasing filtering. The position monitoring uses a combination of laser range finders and inclination sensors to measure micron-level displacement changes and angular deflection of the actuator. All response data are attached with high-precision time stamps (μs level precision) to establish a strict time correspondence with the disturbance injection signal.

[0042] The abnormal correlation determination unit uses spectral correlation analysis to process the collected response data, compares the measured vibration response spectrum with the theoretical response spectrum, and generates a load abnormal correlation factor matrix based on the finite element model and multi-body dynamics simulation of the mechanical system. By calculating the coherence coefficient and phase difference of each frequency band, a load abnormal correlation factor matrix is generated:

[0043] wherein, represents the load abnormal correlation factor, is the number of analysis frequency bands, is the weight coefficient of the i-th frequency band, represents the amplitude of the measured vibration spectrum in the i-th frequency band, represents the amplitude of the theoretical vibration spectrum in the i-th frequency band, represents the phase difference of the i-th frequency band. This factor comprehensively reflects the amplitude and phase deviation degree of the measured response and the theoretical expectation.

[0044] The control parameter output unit encapsulates the configuration results into a standardized control instruction set. The instruction set uses ASN.1 encoding format, including mode switching instructions, load test parameters, safety boundary conditions, and other structured data. Each instruction package is attached with a digital signature and a security check code, and is transmitted to the mechanical control terminal through a 4G / 5G industrial router. The transmission protocol uses the Time-Sensitive Networking (TSN) standard to ensure that the transmission delay of critical instructions is less than 50 milliseconds. The system maintains instruction sending logs, recording the sending time, execution status, and mechanical response summary of each instruction.

[0045] The entire module is deployed in a distributed architecture, with the main control node responsible for decision-making and the backup node synchronizing state data in real time. When the main node fails, the backup node takes over the control task within 100 milliseconds. All configuration parameters and running states are stored persistently in a distributed database, supporting historical operation backtracking and analysis. The module maintains a heartbeat detection mechanism with the mechanical terminal, exchanging state information every second to ensure the reliability of the control link.

[0046] During implementation, the system adopts differentiated control strategies for different types of infrastructure machinery. For hoisting machinery, it focuses on position accuracy control and load stability testing; for excavating machinery, it focuses on vibration characteristic analysis and dynamic response testing. The adjustment of all control parameters follows the gradual principle to avoid sudden state changes that may impact the mechanical system. The execution duration and intensity of the test sequence are dynamically adjusted according to the real-time working state of the machinery, minimizing interference with normal operations while ensuring test effectiveness.

[0047] Example 4: Refer to Figure 5 This example relates to the specific implementation of the supervision report comprehensive generation module, which is responsible for integrating abnormal state data and control execution records, generating structured supervision reports, and implementing risk level assessment. The following through a specific application example illustrates the operation process of this module: in a large infrastructure project, a cluster of hydraulic excavators (including 20 same type devices) in continuous operation process, the system detects that multiple devices have abnormal conditions.

[0048] The data integration unit first extracts the abnormal state indication matrix data for a specified time period (such as August 15, 2023, 8:00-12:00) from the distributed database. There are 127 abnormal events recorded in this time period, covering three types of abnormalities: position drift anomaly (type code POS01), vibration energy anomaly (type code VIB02), and load gradient anomaly (type code LOAD03). At the same time, the unit extracts the control operations performed in the corresponding period from the control log, including 3 high-precision positioning mode enable records and 8 dynamic load test records. Through a sliding window matching algorithm based on nanosecond-level timestamps, the abnormal events and control operations are aligned on the time axis to establish an event-operation association mapping table. For abnormal events with longer duration, the overlapping window segmentation technique is used to divide them into multiple time segments, which are associated with the control operations in the same period.

[0049] The structured template mapping unit maps the integrated data to a preset XML schema report template. The template defines a three-layer structure: the header contains report metadata (project number, machine cluster ID, time range); the main body contains three main field groups: the anomaly type distribution field group records the number, duration, and spatio-temporal distribution characteristics of each type of anomaly; the control execution frequency field group statistics the trigger number, execution duration, and machine coverage rate of each type of control operation; the test result field group stores the response data summary and anomaly correlation factor of the load test. During the mapping process, the system converts the latitude and longitude coordinates of the position drift anomaly event into plane coordinates in the engineering coordinate system, and compresses the vibration anomaly data into a frequency band energy distribution histogram, and summarizes the load test results into the correspondence between the test sequence ID and the correlation factor.

[0050] The regulatory report generation unit generates a structured report based on the mapping results. The report adopts a paging design, with the first page displaying an anomaly event overview chart, the second page presenting a detailed data table, and the third page containing a control operation execution situation summary. The report supports multiple format output, with the PDF version using vector graphics to display the anomaly event spatio-temporal distribution heat map, and the HTML version providing interactive data filtering function. The anomaly type distribution statistics contained in the report are shown in Table 1.

[0051] Table 1: Anomaly type distribution statistics table.

[0052]

[0053] The risk level labeling unit calculates the comprehensive risk score for each machine. The scoring model uses a weighted scoring method: position anomaly weight 0.3, vibration anomaly weight 0.4, load anomaly weight 0.3. The scoring benchmark is the product of the anomaly duration and the maximum intensity, and the normalized risk index is 0-1. According to the project safety specification, the risk level is divided into three levels: 0-0.3 for low risk (green), 0.3-0.7 for medium risk (yellow), and 0.7-1 for high risk (red). In the example, machine EQP-005 has a vibration anomaly intensity of 0.92 and has not been timely mitigated, with a risk score of 0.84, marked as high risk level.

[0054] The report optimization unit hierarchically sorts the report content according to the risk score. The data of high-risk machines is displayed in a prominent position on the first page of the report, and the data of low-risk machines is arranged in time sequence. The report content uses a color coding system, with high-risk items marked with a red border and medium-risk items marked with a yellow border. For devices with continuous anomalies, the report automatically adds a historical anomaly frequency comparison chart to show the anomaly trend of the device in the last 24 hours. After the report is generated, the integrity is guaranteed through digital signature technology, and is uploaded to the block chain storage system for tamper-proof storage.

[0055] The entire module is implemented using a microservices architecture, with the data integration unit running on a Spark computing engine to process large-scale historical data, the template mapping unit implemented in Node.js for real-time data conversion, and the report generation unit using Python's ReportLab library to generate PDF documents. The system supports multi-tenant isolation, with different project data processing completely independent, and report templates can be customized according to project needs. Output reports are distributed asynchronously to the project management system, equipment maintenance system, and security supervision platform through a message queue, and each system can extract specific data fields from the report according to its own needs.

[0056] Embodiment 5: The implementation involves a complete technical implementation of a remote execution feedback module that establishes a closed-loop system from control instruction generation to execution and feedback, enabling dynamic optimization and adjustment of control strategies. The module is deployed in a distributed cloud-edge collaborative architecture, consisting of a cloud control center and edge data acquisition nodes.

[0057] The control instruction transmission unit is designed using a multi-protocol adaptation architecture. After receiving the instruction set from the control strategy adaptive generation module, the unit first performs instruction validity verification, including parameter range checking, mechanical state compatibility verification, and safety constraint condition review. The verified instructions are converted into a standard industrial protocol format recognizable by the equipment, such as Modbus TCP, CANopen, or EtherCAT, with the specific protocol automatically selected based on the target machine's control system type. Instruction sequence numbers, timestamps, and CRC check codes are added during the instruction encapsulation process to form complete data frames. The transmission channel dynamically selects between direct transmission mode when the 4G / 5 network is smooth and message queue caching mechanism when network latency is large, ensuring reliable delivery of instructions. The transmission process uses end-to-end encryption, with the instruction content encrypted using the AES-256 algorithm and the recipient's identity verified through digital certificates.

[0058] The execution state monitoring unit is deployed in the edge controller of the machine body, consisting of a multi-channel data acquisition system and a real-time processing kernel. When the control instruction reaches the machine control terminal, the unit immediately starts the execution state tracking program. By reading the feedback sensor data of the actuator, the machine's response characteristics are captured in real time: for positioning control instructions, the positioning accuracy factor of the GNSS receiver and the satellite lock state are monitored; for load regulation instructions, the data stream of the hydraulic system pressure sensor and flow meter is collected. Monitoring data is sampled at millisecond-level frequency and attached with timestamps accurate to microseconds. The unit's embedded state machine model sets the expected response time window based on the instruction type, and if the expected mechanical state change is not detected within the specified time, an execution timeout alarm is triggered. After preliminary filtering, all monitoring data is tagged with the corresponding instruction sequence number and time interval, ready to be uploaded to the feedback system.

[0059] The feedback data backflow unit establishes a bidirectional data channel management mechanism. The unit packages the mechanical response data collected by the execution state monitoring unit into standardized data packets, and the data packet structure includes two parts: a metadata header and a payload. The metadata header records the machine number, instruction sequence number, data collection time range, and data verification information; the payload includes compressed raw sensor data and an execution state summary. The data transmission uses an adaptive compression algorithm that dynamically adjusts the data compression rate and transmission frequency based on network bandwidth conditions. When the bandwidth is sufficient, the complete data details are transmitted, and when the bandwidth is limited, only the abstract information extracted by feature extraction is transmitted. The data backflow path is distributed through a load balancer to ensure that a large number of machines simultaneously feedback will not cause network congestion. After the backflow data reaches the cloud, it first enters the data buffer pool for time series alignment and integrity check, and then is forwarded to the data preprocessing interface of the working condition feature dynamic extraction module.

[0060] The response data analysis unit runs on the cloud-based stream processing engine and receives the mechanical response data forwarded by the feedback data backflow unit. The analysis process is divided into three levels: the raw data layer performs signal denoising and outlier removal, uses a wavelet transform-based filtering algorithm to process high-frequency vibration data, and uses a sliding window average method to smooth pressure fluctuation data; the feature extraction layer calculates key indicators from the preprocessed data, including positioning accuracy stability coefficient, load response delay time, vibration frequency spectrum feature value, etc.; the pattern recognition layer compares the extracted features with the expected response model, which is based on the mechanical model and historical normal data. The analysis results are organized into response feature vectors, with vector dimensions including numerical features and classification features. Numerical features are normalized, and classification features are represented using one-hot encoding.

[0061] The policy adjustment triggering unit continuously monitors the deviation of the response feature vector from the expected model. The deviation calculation uses a multi-dimensional weighted distance measurement method, considering both the Euclidean distance of numerical features and the similarity of classification features. When the deviation value exceeds the preset dynamic adjustment threshold, the unit generates a policy adjustment suggestion message. The adjustment threshold is adaptively calculated based on the mechanical operation history and current working condition environment, with appropriate threshold tolerance increased in harsh working conditions and reduced in precision operation stages. After triggering the adjustment, the unit sends a reconfiguration request to the control strategy adaptive generation module, which includes the specific deviation analysis report and the adjustment direction suggestion. At the same time, the unit records relevant data for each triggering event, including triggering time, deviation value, adjustment suggestion content, and subsequent improvement effect. These data are used to optimize the calculation model of the adjustment threshold.

[0062] The whole module implements full-link state tracking function, and each control instruction forms a complete trace chain from sending to feedback. The module maintains an instruction life cycle database, which records the creation time, sending time, execution start time, execution end time, feedback receiving time and processing result of each instruction. Based on time series analysis of instruction execution efficiency, network delay, mechanical response delay and other bottleneck links are identified. The system supports manual intervention mode, and operation and maintenance personnel can check the instruction execution status at any time, manually trigger data backflow or strategy adjustment. All functions are provided through RESTful API, supporting integration with other operation and maintenance management systems. The module adopts a high-availability deployment architecture, and key components have redundant backups to ensure that basic functions can still run when some nodes fail.

[0063] It should be noted that, in this text, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between or among the entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0064] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A capital construction machinery remote control and supervision system based on operating condition analysis, characterized by, The application relates to a real-time monitoring and control system for construction machinery, comprising: a running data real-time acquisition module, which acquires running condition data in real time based on a heterogeneous sensor network deployed on construction machinery, including mechanical position coordinate sequences, vibration frequency spectrum distribution and load pressure fluctuation; a working condition feature dynamic extraction module, which inputs the running condition data into a multi-scale fusion processing unit and outputs a working condition feature vector containing time domain stability features, frequency domain energy distribution features and load change gradients; an abnormal state recognition module, which compares the working condition feature vector with a preset reference working condition model in multiple dimensions to generate an abnormal state indication matrix; a control strategy adaptive generation module, which configures remote control parameters, including enabling a high-precision positioning tracking mode and applying dynamic load adjustment tests, based on the abnormal state indication matrix; a supervision report comprehensive generation module, which integrates the abnormal state indication matrix and control parameter execution logs to generate a structured supervision report.

2. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 1, characterized in that, The running data real-time acquisition module specifically comprises: a sensor data synchronization unit which coordinates the sampling frequency of the heterogeneous sensor network through an edge computing node to ensure that the timestamps of the mechanical position coordinate sequences, vibration frequency spectrum distribution and load pressure fluctuation are aligned; a data quality enhancement unit which performs environmental noise suppression processing on the real-time acquired running condition data to eliminate measurement deviations caused by temperature and humidity interference; a preprocessed data output unit which converts the processed running condition data into a standardized data stream and inputs it into the working condition feature dynamic extraction module.

3. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 2, characterized in that, The working condition feature dynamic extraction module specifically comprises: time domain stability feature calculation which performs window sliding analysis on the mechanical position coordinate sequences in the standardized data stream, calculates position drift accumulation and speed mutation frequency, and generates a time domain stability feature vector; frequency domain energy distribution feature extraction which performs frequency band decomposition algorithm on the vibration frequency spectrum distribution in the standardized data stream, quantifies the energy density ratio in the preset sensitive frequency band, and constructs a frequency domain energy distribution feature vector; load change gradient evaluation which calculates the slope distribution and interval dispersion of the load amplitude mutation point based on the load pressure fluctuation in the standardized data stream, and generates a load change gradient vector; feature fusion processing which tensor splices the time domain stability feature vector, frequency domain energy distribution feature vector and load change gradient vector, and outputs the working condition feature vector through normalized weighted fusion.

4. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 3, characterized in that, The abnormal state recognition module specifically comprises: a reference model construction unit which trains a reference working condition model based on historical normal running condition data, and the reference working condition model contains a position stability threshold, a vibration energy reference spectrum and a load pressure standard range; a multi-dimensional comparison execution unit which aligns the working condition feature vector with the reference working condition model by dimension, calculates the position drift accumulation deviation of the time domain stability feature, the energy density deviation of the frequency domain energy distribution feature and the slope distribution difference of the load change gradient, and generates an initial abnormal indicator set; an abnormal state indication matrix generation unit which performs importance weighting processing on the initial abnormal indicator set to eliminate dimensional influence and outputs a three-order abnormal state indication matrix with dimensions of [machine number x timestamp x abnormal type].

5. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 4, characterized in that, The control strategy adaptive generation module specifically comprises: A high-precision positioning tracking mode configuration, when the position drift accumulation deviation in the abnormal state indication matrix exceeds a preset deviation threshold, a high-precision positioning tracking mode is enabled, and the position sampling frequency is increased to 3-5 times of the original frequency; A dynamic load adjustment test execution, a multi-stage load pressure disturbance sequence is applied to the mechanical node with the largest slope distribution difference of the load change gradient in the abnormal state indication matrix; A control parameter output unit encapsulates the configured remote control parameters into a control instruction set and inputs the control instruction set into a remote execution interface.

6. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 5, characterized in that, The dynamic load adjustment test execution specifically comprises: A load disturbance injection unit generates a multi-stage load pressure disturbance sequence through a controllable actuator, including a step pressure increment and a pulse pressure fluctuation; A response data capture unit records the mechanical vibration response spectrum and the position offset after the load pressure disturbance sequence injection; An abnormal association determination unit compares the offset degree of the measured vibration response spectrum and the theoretical response spectrum to calculate a load abnormal association factor.

7. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 4, characterized in that, The supervision report comprehensive generation module specifically comprises: A data integration unit extracts abnormal type distribution data in the abnormal state indication matrix and aligns the data with the mode switching record and the test result in the control parameter execution log on a time axis; A structured template mapping unit maps the integrated data to a preset report template field, including an abnormal type distribution field, a control execution frequency field and a test result field; A supervision report generation unit generates the structured supervision report containing a multi-dimensional data table based on the mapping result.

8. The capital construction machinery remote control and supervision system based on operating condition analysis according to claim 7, characterized in that, The supervision report comprehensive generation module further comprises: A risk level labeling unit calculates the comprehensive risk score of each mechanical node according to the abnormal type distribution data in the abnormal state indication matrix and divides the risk level interval; A report optimization unit performs information level sorting processing on the structured supervision report based on the comprehensive risk score.

9. The system for remote control and supervision of capital equipment based on operating condition analysis according to claim 1, characterized in that, The system further comprises a remote execution feedback module: A control instruction transmission unit sends the control instruction set output by the control strategy adaptive generation module to a capital construction machinery control terminal; An execution state monitoring unit captures mechanical response data in a control instruction execution process in real time; A feedback data backflow unit inputs the mechanical response data into the working condition characteristic dynamic extraction module.

10. The capital machinery remote control and monitoring system based on operating condition analysis of claim 9, wherein, The remote execution feedback module specifically comprises: A response data analysis unit performs feature extraction processing on the captured mechanical response data to generate a response feature vector; A strategy adjustment triggering unit triggers the reconfiguration of the control strategy adaptive generation module when the response feature vector deviates from the expected response model by more than a preset adjustment threshold.

Images