Intelligent monitoring method and system for high-precision underwater special equipment operation

Abstract

The invention relates to a high-precision underwater special equipment operation intelligent monitoring method, and belongs to the field of underwater robot control. According to the method, body attitude, tail end pose, environment point cloud and channel state data are collected through a multi-mode heterogeneous sensing array, and a heterogeneous data set is generated through space-time registration; a guided acousto-optic coupling propagation model is adopted to estimate a suspended matter concentration field and a turbulence velocity field so as to suppress interference; constructing a digital twinborn evolution network embedded in the sealed cavity multi-field coupling constraint, and outputting a state evolution trend and residual operation capability evaluation; aligning the process knowledge graph through a causal inference engine, interpretively evaluating the trajectory deviation and generating a compensation instruction; and introducing a time sequential dependency constraint into the attention mechanism to realize proshot early warning and active strategy reconstruction. According to the method, the crossing from passive response to active prediction, from local observation to holographic deduction and from black box diagnosis to interpretable decision is realized, and the precision, safety and autonomy of underwater operation are improved.

Description

Technical Field

[0001] This invention relates to the technical field of underwater robot control, and in particular to an intelligent monitoring method and system for high-precision underwater special equipment operations. Background Technology

[0002] With the technological advancements in marine resource development, deep-sea engineering construction, and underwater emergency rescue, underwater special equipment (such as underwater welding robots, submarine cable laying devices, and deep-sea sampling robotic arms) needs to perform high-precision operations in complex aquatic environments with strong interference and high risks.

[0003] Currently, underwater monitoring technology mainly relies on a single-modal sensing approach (such as optical imaging or sonar ranging) combined with post-event data analysis. However, in turbid waters, the scattering effect of suspended particles on light and the multipath interference of sound waves cause severe degradation of sensor data. Existing fixed-parameter filtering methods cannot adapt to dynamic changes in water quality online, resulting in target feature obscuration and measurement distortion, making it difficult to guarantee operational accuracy. This reduces the precision of underwater operations and therefore requires improvement. Summary of the Invention

[0004] To improve the accuracy of underwater operations, this application provides an intelligent monitoring method and system for high-precision underwater special equipment operations.

[0005] Firstly, the above-mentioned inventive objective of this application is achieved through the following technical solution: A smart monitoring method for high-precision underwater special equipment operations, the method comprising the following steps: Multi-scale data of underwater special equipment is collected by multi-modal heterogeneous sensor arrays. After spatiotemporal registration and fusion of multi-scale data, a heterogeneous dataset with standardized monitoring semantics is generated. The heterogeneous dataset is input into the dynamic water area interference suppression layer, which estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The enhanced sensing data stream is sent into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment state evolution trend characteristics and remaining working capacity assessment value. The state evolution features are aligned with the knowledge graph of the operation process procedure. An interpretable causal reasoning engine is used to evaluate the deviation of the current operation trajectory and the compliance with the process in real time, and outputs the accuracy compensation amount and the process adaptive adjustment instruction. The accuracy compensation command and the remaining work capacity assessment value are input into the fault proactive early warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical early warning information, and triggers proactive work strategy reconstruction commands.

[0006] By adopting the above technical solutions, this application deeply embeds physical laws into the perception, modeling, reasoning, and verification processes, forming a new paradigm of intelligent monitoring with adaptive perception, proactive early warning, interpretable decision-making, and twin verification. This enables underwater special equipment operations to have strong interference resistance, transparent status, interpretable decision-making, and high operational reliability.

[0007] In a preferred example, this application can be further configured as follows: the step of generating a heterogeneous dataset with standardized monitoring semantics after collecting multi-scale data of underwater special equipment through a multi-modal heterogeneous sensor array, performing spatiotemporal registration and multi-scale data fusion includes the following steps: A nine-axis inertial measurement unit, a fiber optic strain sensor, and an ultrasonic ranging probe are deployed on the body of the special equipment, and a microelectromechanical system attitude sensor and a vision camera are integrated in the end effector. A submodule for underwater acoustic communication channel detection is constructed. By sending orthogonal frequency division multiplexing pilot signals and receiving echoes, the multipath delay spread and Doppler frequency shift parameters of the channel are estimated. The collected body posture data, end pose data, environmental point cloud data and channel status data are timestamped and aligned. A dynamic time warping algorithm is used to eliminate timing mismatch caused by long delays in underwater acoustic communication. The aligned heterogeneous data are fused into multi-scale feature volumes under a unified monitoring semantic framework to generate a standardized heterogeneous dataset containing spatial geometric information, mechanical state information, and channel quality information.

[0008] In a preferred example, this application can be further configured as follows: The step of inputting the heterogeneous dataset into a dynamic water area interference suppression layer, wherein the dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the operating area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression, includes the following steps: A differentiable underwater acoustic-optical coupling propagation sub-model was constructed, using point cloud data of the operating area as spatial constraints, to inversely deduce the forward and backward scattering intensity distributions of sound and light waves in the suspended particle swarm. The suspended matter concentration field and turbulent velocity field are estimated based on the scattering intensity distribution. The concentration field is modeled as a learnable three-dimensional implicit representation, and a smoothing constraint is applied through the fluid continuity equation. Based on the estimated concentration field and velocity field, the depth measurement deviation of the point cloud data and the contrast attenuation of the visual image are dynamically compensated to generate an enhanced perception data stream after interference suppression. A rendering consistency self-supervised mechanism is introduced, in which the compensated data stream is re-input into the propagation model to generate simulated sensor readings, and the residual loss is calculated with the original readings to update the model parameters, thereby achieving unsupervised online learning.

[0009] In a preferred embodiment, this application can be further configured as follows: In the step of sending the enhanced sensing data stream into a special equipment digital twin evolution network, wherein the special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on a physical information-driven architecture, the construction of the special equipment digital twin evolution network includes the following steps: A rigid-flexible coupling dynamic sub-model of the equipment is constructed. The rigid link of the robotic arm and the flexible sealed cable are modeled as Euler-Bernoulli beam elements and nonlinear spring-damping systems, respectively. The equations of motion are established through the principle of virtual work. The pressure-temperature-deformation multi-field coupling constraint of the sealing cavity is embedded in the dynamic model. The compression of the sealing ring and the leakage risk coefficient are calculated based on the thermoelastic mechanics theory and fed back to the dynamic equation to correct the boundary conditions. Using the enhanced sensing data stream as the observation input, the unscented Kalman filter algorithm is used to recursively estimate the hidden state of the dynamic model, and outputs the evolution trend of the joint torque, end contact force and sealing state of the equipment. Based on the current state estimate, the performance degradation trajectory of the equipment is projected forward over the remaining operating time, and the remaining operating capacity assessment value and health index are output.

[0010] In a preferred embodiment, this application can be further configured as follows: embedding multi-field coupling constraints of pressure-temperature-deformation of the sealing cavity into the dynamic model, calculating the compression of the sealing ring and the leakage risk coefficient based on thermoelasticity theory, and feeding this back to the dynamic equation to correct the boundary conditions. The method for calculating the leakage risk coefficient of the sealing cavity includes: Based on the thermoelastic mechanics theory, the constitutive relation of the sealing ring is established, and the material's compressive modulus and temperature are coupled and expressed as a time-varying function. Calculate the contact stress distribution of the sealing ring under the current working conditions, and perform vector difference with the underwater environmental pressure field to obtain the effective sealing pressure; The effective sealing pressure is input into the fatigue crack propagation sub-model to estimate the microcrack propagation rate of the sealing ring and map it to a leakage probability score. The leakage probability score is fed back to the boundary conditions of the dynamic model. When the score exceeds the preset level, the maximum allowable operating depth and speed of the equipment are automatically reduced to achieve adaptive risk avoidance.

[0011] In a preferred embodiment, this application can be further configured as follows: In the step of aligning the state evolution features with the knowledge graph of the work process procedure, evaluating the current work trajectory deviation and process compliance in real time through an interpretable causal inference engine, and outputting the accuracy compensation amount and process adaptive adjustment instructions, the online work accuracy evaluation and adaptive adjustment process includes the following steps: The equipment state evolution features are mapped into a sequence of operation trajectory points, and subgraph isomorphic matching is performed with the pre-stored process specification knowledge graph to construct a trajectory-process alignment graph structure. Run an interpretable causal inference engine on the alignment graph to identify key state variables that cause trajectory deviations through causal discovery algorithms and quantify the contribution of each variable to the deviation. Based on the contribution analysis results, the accuracy compensation amount of the end effector pose is calculated online, and the compensation amount is decomposed into joint angle fine-tuning component and path replanning component. When causal reasoning discovers that multivariate coupling leads to systematic deviations, an adaptive adjustment instruction is triggered to dynamically modify the process parameters such as operating speed, feed rate, and tool posture, ensuring that the overall process compliance is maintained above the qualified threshold.

[0012] In a preferred embodiment, this application can be further configured such that: in the step of inputting the accuracy compensation command and the remaining operational capacity assessment value into the proactive fault warning module, which introduces time-sequential dependency constraints in the attention mechanism, generates hierarchical warning information, and triggers an active operational strategy reconfiguration command, the proactive fault warning and strategy reconfiguration process includes: The remaining operational capacity assessment value, the historical sequence of accuracy compensation instructions, and the evolution trend of the sealing status are encoded into temporal feature vectors and input into the attention mechanism network; By introducing temporal sequential dependency constraints into the attention mechanism, the attention weights are forced to follow an autoregressive process, which suppresses overfitting to random noise and enhances attention to progressive degradation patterns. The graded warning generator outputs three levels of warning information based on the intensity of the degradation mode of attention focus: normal monitoring level, preventive maintenance level, and emergency shutdown level; When the warning level reaches the preventive maintenance level, the proactive operation strategy reconfiguration command is automatically triggered. The deceleration operation, tool replacement or path detour strategy is rehearsed in the digital twin, and the optimal reconfiguration scheme is selected and executed.

[0013] Secondly, the above-mentioned inventive objective of this application is achieved through the following technical solutions: An intelligent monitoring device for high-precision underwater special equipment operation, the device includes: a heterogeneous dataset construction unit, used to collect multi-scale data of underwater special equipment through a multi-modal heterogeneous sensor array, and generate a heterogeneous dataset with standardized monitoring semantics after spatiotemporal registration and fusion of multi-scale data; An enhanced sensing data stream generation unit is used to input the heterogeneous dataset into a dynamic water area interference suppression layer. The dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The status trend and capability assessment unit is used to send the enhanced perception data stream into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment status evolution trend characteristics and remaining operational capability assessment value. An adaptive adjustment instruction generation unit is used to align the state evolution characteristics with the work process procedure knowledge graph, evaluate the current work trajectory deviation and process compliance in real time through an interpretable causal reasoning engine, and output the accuracy compensation amount and process adaptive adjustment instructions. The graded early warning information generation unit is used to input the accuracy compensation instruction and the remaining operation capacity assessment value into the fault proactive early warning module. This module introduces time-sequential dependency constraints in the attention mechanism to generate graded early warning information and trigger proactive operation strategy reconstruction instructions.

[0014] Thirdly, the above-mentioned objectives of this application are achieved through the following technical solutions: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described intelligent monitoring method for high-precision underwater special equipment operations.

[0015] Fourthly, the above-mentioned objectives of this application are achieved through the following technical solutions: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described intelligent monitoring method for high-precision underwater special equipment operations. Attached Figure Description

[0016] Figure 1 This is a flowchart of an intelligent monitoring method for high-precision underwater special equipment operation according to an embodiment of this application; Figure 2 This is a schematic diagram of a smart monitoring device for high-precision underwater special equipment operation according to one embodiment of this application; Figure 3 This is a schematic diagram of an electronic device according to an embodiment of this application.

[0017] Icon labels: 1. Heterogeneous dataset construction unit; 2. Enhanced perception data stream generation unit; 3. Status trend and capability assessment unit; 4. Adaptive adjustment instruction generation unit; 5. Hierarchical early warning information generation unit. Detailed Implementation

[0018] The present application will be further described in detail below with reference to the accompanying drawings.

[0019] In one embodiment, such as Figure 1 As shown, this application discloses an intelligent monitoring method for high-precision underwater special equipment operations, which specifically includes the following steps: S10: Collect multi-scale data of underwater special equipment through multi-modal heterogeneous sensor array, and generate a heterogeneous dataset with standardized monitoring semantics after spatiotemporal registration and multi-scale data fusion. In this embodiment, the multi-scale data includes body posture data, end effector posture data, working environment point cloud data, and underwater acoustic communication channel status data. Specifically, in an underwater pipeline welding operation scenario, a nine-axis inertial measurement unit and a fiber optic strain sensor are deployed on the welding robot body to collect real-time data on the robot arm joint angles, angular velocities, and link deformation. A microelectromechanical attitude sensor and a vision camera are integrated at the welding torch end to acquire the welding torch spatial posture and weld area point cloud data. Simultaneously, an ultrasonic ranging probe and an underwater acoustic communication detection module are mounted on the robot body to measure the working distance and communication channel delay spread parameters. The system timestamps the data from each sensor, uses a dynamic time warping algorithm to eliminate timing mismatch caused by long underwater acoustic communication delays, and fuses the body kinematics data, end effector posture data, environmental point cloud data, and channel status data under a unified monitoring semantic framework to generate a multi-dimensional heterogeneous dataset containing mechanical state, spatial geometry, and channel quality, providing a standardized input basis for subsequent processing.

[0020] S20: Input the heterogeneous dataset into the dynamic water area interference suppression layer. The dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. Specifically, when welding in turbid waters, the contrast of weld images acquired by the visual camera is severely reduced, and depth measurement errors occur in the point cloud data. A dynamic water area interference suppression layer initiates a differentiable underwater acoustic-optical coupling propagation sub-model. Using point cloud data as spatial constraints, it inversely extrapolates the scattering intensity distribution of acoustic and optical waves in the suspended particle group, estimating the gradient distribution of suspended particle concentration in the working area (lower concentration near the robot, higher concentration at a distance), and calculating the optical path offset caused by the turbulent velocity field. Based on the estimation results, the system dynamically compensates for image contrast attenuation, improving the clarity of weld edges, while simultaneously correcting the offset of point cloud depth measurements, generating an enhanced perception data stream after interference suppression. This enables subsequent feature extraction to accurately identify the weld contour and molten pool morphology. S30: The enhanced sensing data stream is sent to the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment state evolution trend characteristics and remaining working capacity assessment value. Specifically, after the enhanced perception data stream is input into the digital twin evolution network of special equipment, the network constructs a rigid-flexible coupled dynamic model of the equipment: the rigid linkage of the robotic arm is modeled as an Euler-Bernoulli beam element, and the connecting cables are modeled as a nonlinear spring-damped system. The overall motion equation is established through the principle of virtual work. Simultaneously, multi-field coupled constraints of pressure, temperature, and deformation in the sealed cavity are embedded. Based on thermoelasticity theory, the compression of the sealing ring is calculated, and the leakage risk coefficient is assessed. The network uses an unscented Kalman filter algorithm to recursively estimate the hidden states of the dynamic model, outputting the current joint torque, end contact force, and the evolution trend of the sealing state. Based on the current state, the network forward extrapolates the performance degradation trajectory over the next two hours, assesses the remaining operational capability as "able to complete 3 more welds," and outputs a health index of "moderate," indicating the need to monitor the sealing state. S40: Align the state evolution features with the knowledge graph of the operation process procedure, evaluate the deviation of the current operation trajectory and the process compliance in real time through an interpretable causal reasoning engine, and output the accuracy compensation amount and the process adaptive adjustment instruction. Specifically, the state evolution characteristics showed increased fluctuations in the contact force at the welding torch tip, causing the welding trajectory to deviate from the process specifications. The system mapped the state characteristics into a sequence of trajectory points and aligned them with the "argon arc welding process" subgraph in the pre-stored knowledge graph. The interpretable causal reasoning engine identified abnormal joint torque and fluctuations in sealing cavity pressure as strong causal variables causing the trajectory deviation, with joint torque contributing a high percentage. Based on the analysis results, the system calculated the accuracy compensation online: generating a joint angle fine-tuning component to correct the welding torch posture, and simultaneously generating a path replanning component to optimize the welding speed. Due to the systematic deviation caused by multivariate coupling, the system further triggered an adaptive adjustment command, dynamically reducing the welding feed rate and adjusting the argon flow rate to ensure that the welding process compliance remained at a qualified level. After compensation, the trajectory deviation was significantly reduced.

[0021] S50: Input the accuracy compensation command and the remaining work capacity assessment value into the fault proactive early warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical early warning information, and triggers the proactive work strategy reconstruction command. The remaining operational capacity assessment showed a progressive degradation trend in the sealing condition, with health indicators approaching the preventative maintenance threshold. The historical sequence of accuracy compensation commands indicated an increase in recent compensation frequency. After these temporal characteristics were input into the proactive fault warning module, the attention mechanism focused on the progressive degradation pattern rather than random noise. The graded warning generator output a "preventative maintenance level" warning message, indicating a risk of leakage in the sealing condition. The module automatically triggered an active operational strategy reconfiguration command, rehearsing three strategies in the digital twin: deceleration, sealing ring replacement, or early termination and resurfacing. The rehearsal results showed that deceleration could extend the safe operating window by 30 minutes. The system selected this option and issued an execution command; the welding torch operating speed automatically decreased, and after successfully completing the current weld, it safely returned to base.

[0022] This application systematically solves the four core challenges of real-time performance, accuracy, interpretability, and safety in monitoring underwater special equipment in complex aquatic environments by constructing an end-to-end closed-loop architecture of multimodal perception, physical constraint modeling, causal reasoning decision-making, and twin pre-simulation verification. It achieves the following comprehensive beneficial effects that distinguish it from existing technologies: First, the synergistic effect of multimodal heterogeneous sensor fusion and dynamic water disturbance suppression overcomes the bottleneck of information failure in turbid waters caused by single sensing modes. Traditional methods rely on independent vision or sonar perception, which is susceptible to data failure due to scattering by suspended particles and turbulence distortion. This scheme simultaneously acquires the body attitude, end-effector pose, environmental point cloud, and underwater acoustic channel status, and innovatively constructs a differentiable underwater acoustic-optical coupling propagation model, realizing real-time estimation and adaptive compensation of suspended particle concentration field and turbulence velocity field. This mechanism not only recovers the characteristics of the target submerged by disturbance, but also supports online unsupervised learning through the differentiability of the propagation model, enabling the system to continuously optimize the suppression effect as water quality dynamically changes, significantly improving the perception robustness and measurement accuracy in complex water environments.

[0023] Secondly, the physical information-driven digital twin evolutionary network enables holographic projection of equipment status and quantification of remaining capacity. Existing monitoring technologies are mostly limited to the surface reflection of equipment kinematics, failing to characterize the multi-field coupling effects of pressure-temperature-deformation in sealed cavities, and even more difficult to assess remaining operational capacity. This solution innovatively embeds rigid-flexible coupling dynamic equations and thermoelastic mechanical constraints of sealing rings into the twin network, enabling the model to not only estimate latent states such as joint torques and contact forces in real time, but also to project performance degradation trajectories forward and output remaining operational capacity assessment values. This capability elevates the system from passive monitoring to proactive prediction, providing a quantitative basis for preventative maintenance and work planning, and effectively avoiding the risk of work failure or equipment damage due to unknown status.

[0024] Furthermore, the alignment mechanism between the interpretable causal reasoning engine and the process knowledge graph solves the problem of opaque diagnosis inherent in traditional black-box models. Existing deviation analysis cannot pinpoint specific causal components, leading to blind and inefficient maintenance. This solution, by constructing a trajectory-process alignment graph structure and running causal reasoning, accurately identifies key variables causing operational deviations and quantifies their contribution, mapping abstract errors to physical problems of specific components. This interpretability not only supports online generation of accuracy compensation and process adjustment instructions but also provides highly visualized diagnostic evidence, enabling operators to quickly locate fault sources, significantly shortening underwater maintenance response time, and improving the targeting and flexibility of operational quality control.

[0025] The dual closed-loop mechanism of proactive fault warning and digital twin pre-simulation constructs a dual safety guarantee of in-process risk control and pre-event verification. Traditional systems rely on post-event alarms, making it difficult to avoid unplanned downtime and underwater rescue costs. This solution, through the temporal sequence constraint in the attention mechanism, achieves early identification and graded warning of progressive degradation, enabling strategy reconfiguration to shift from passive response to proactive prevention. The multiphysics field pre-simulation capability of the digital twin provides a virtual test field before strategy execution, evaluating robustness to extreme operating conditions through random perturbation injection, ensuring the safety and economy of the instruction sequence. The closed-loop feedback of these two mechanisms realizes a complete chain of monitoring-early warning-decision-verification, significantly reducing the probability of decision-making errors in high-risk operations.

[0026] In summary, this application deeply integrates physical laws into the perception, modeling, reasoning, and verification processes, forming a new paradigm of intelligent monitoring with adaptive perception, proactive early warning, interpretable decision-making, and twin verification. This enables underwater special equipment operations to have strong interference resistance, status transparency, interpretable decision-making, and high operational reliability, overcoming the technical challenges of high-precision intelligent monitoring in complex aquatic environments.

[0027] In step S10: Collecting multi-scale data from underwater special equipment using a multi-modal heterogeneous sensor array, and generating a standardized monitoring semantic heterogeneous dataset after spatiotemporal registration and multi-scale data fusion, the following steps are included: S11: A nine-axis inertial measurement unit, a fiber optic strain sensor, and an ultrasonic ranging probe are deployed on the body of the special equipment, and a microelectromechanical system attitude sensor and a vision camera are integrated in the end effector. S12: Construct an underwater acoustic communication channel detection submodule, and estimate the channel multipath delay spread and Doppler frequency shift parameters by sending orthogonal frequency division multiplexing pilot signals and receiving echoes. S13: Perform timestamp alignment on the collected body posture data, end pose data, environmental point cloud data and channel status data, and use dynamic time warping algorithm to eliminate timing mismatch caused by long delay in underwater acoustic communication; S14: The aligned heterogeneous data is fused with multi-scale feature volumes under a unified monitoring semantic framework to generate a standardized heterogeneous dataset containing spatial geometric information, mechanical state information and channel quality information.

[0028] In this embodiment of the application, steps S11-S14 continue to be taken as an example of intelligent underwater pipeline welding operations. A multimodal sensor array is deployed inside and outside the welding robot's main body section: a nine-axis inertial measurement unit is installed at the robot arm base and three rotary joints to sense the arm's roll, pitch, yaw angles, and three-axis acceleration in real time; a fiber optic strain sensor array is attached along the robot arm's connecting rod axis to monitor the micro-strain distribution of the connecting rod material under underwater high pressure; an ultrasonic ranging probe is configured at the robot's front end to measure the distance information to the surface of the pipe to be welded. At the welding torch end effector, a microelectromechanical system (MEMS) attitude sensor is integrated to obtain the spatial orientation of the welding torch tip, and a waterproof vision camera is installed to capture image data of the weld bevel and molten pool morphology. All sensors are connected to a central data acquisition unit via waterproof sealed cables, forming a comprehensive perception system covering mechanical state, spatial pose, and environmental vision.

[0029] During welding operations, the robot needs to communicate with the surface mother ship in real time to receive process parameters. The system incorporates an underwater acoustic communication channel detection submodule, which periodically transmits orthogonal frequency division multiplexing (OFDM) pilot signals during data communication intervals. Once the pilot signal reaches the mother ship via the underwater channel and returns as an echo, the submodule analyzes the echo signal, estimating the current channel's multipath delay spread to be approximately 15 milliseconds and the Doppler shift due to ocean currents to be approximately 3 Hz. These channel parameters reflect the available bandwidth and signal distortion level of the underwater acoustic communication, providing a basis for subsequent adaptive adjustment of the communication coding rate, and are also incorporated into the monitoring dataset as channel quality information.

[0030] Due to the tens of milliseconds of latency in underwater acoustic communication, a significant timing mismatch occurs between the mother ship's commands and the robot's sensor data. Each data stream acquired by the system carries a local timestamp; for example, the robot's inertial data is recorded at 10:00:00.000, the end-effector's visual image at 10:00:00.015, while the mother ship's commands received via the underwater acoustic channel are timestamped at 10:00:00.200. A dynamic time warping algorithm is employed to align these timestamp sequences, using the underwater acoustic communication latency parameter as a constraint, aligning each data stream to a unified time base. This eliminates the timing mismatch, ensures the causal timing correctness in subsequent fusion processing, and avoids misjudging command response delays as equipment failures.

[0031] The aligned heterogeneous data are fused within a unified monitoring semantic framework: the joint angular velocities in the body inertial data are scale-aligned with the spatial orientation data from the end effector attitude sensor, fusing to generate the overall pose state feature volume of the robotic arm; the sparse point cloud data from ultrasonic ranging is spatially registered with the visual camera image, fusing to generate high-resolution 3D geometric features of the weld area; the fiber optic strain data is converted into stress distribution features, which, together with the pose state features, constitute the mechanical state information volume; and the time delay spread and frequency shift parameters obtained from underwater acoustic channel detection are encoded as channel quality features. These multi-scale feature volumes are cascaded along the channel dimension to generate a standardized heterogeneous dataset containing spatial geometry, mechanical state, and channel quality. This dataset, indexed by a unified timestamp, provides structured input for subsequent interference suppression and state estimation.

[0032] For steps S11-S14, the multimodal heterogeneous sensor array construction step achieves full-dimensional information capture of underwater operation status through a comprehensive sensing layout encompassing the sensor body, terminal, environment, and channel, avoiding monitoring blind spots caused by single sensor failure due to water interference. The introduction of the underwater acoustic communication channel detection submodule, for the first time, incorporates communication quality parameters into the monitoring semantic system, providing data support for analyzing the impact of communication status on operations. The dynamic time warping algorithm effectively eliminates the timing mismatch problem caused by long underwater acoustic delays, ensuring the causal timing correctness of monitoring data and avoiding the risk of misjudgment due to time misalignment. The multi-scale feature volume fusion mechanism unifies heterogeneous data into a standardized semantic framework, achieving the organic integration of spatial geometry, mechanical state, and communication quality. This provides high-quality input with a clear structure and complete information for subsequent processing layers, significantly improving the data foundation reliability and information utilization efficiency of the entire monitoring system.

[0033] In step S20: The heterogeneous dataset is input into the dynamic water area interference suppression layer, which estimates the suspended matter concentration field and turbulent velocity field of the operating area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. This step includes the following steps: S21: Construct a differentiable underwater acoustic-optical coupling propagation sub-model, using point cloud data of the work area as spatial constraints, and inversely deduce the forward and backward scattering intensity distribution of sound waves and light waves in the suspended particle group. Specifically, in underwater pipeline welding operations, the weld seam images captured by the vision camera exhibit severe fogging, and the point cloud data shows drastic fluctuations in the measured distance between the welding torch and the pipeline. When constructing a differentiable underwater acoustic-optical coupling propagation sub-model, the system uses the spatial geometry provided by the point cloud data of the work area as a constraint, and inversely extrapolates the forward and backward scattering intensity distribution of light waves in the suspended particle swarm from each pixel. For example, for a spatial point in the weld bevel region, the model calculates that when the incident beam passes through a layer of suspended particles with uneven concentration, part of the light energy is scattered forward to adjacent pixels, and part is scattered backward back to the light source direction, forming a fogging effect. Simultaneously, the acoustic ranging signal generates multipath echoes due to particle scattering. The model extrapolates the acoustic energy attenuation and phase delay of each path, quantitatively estimating the scattering intensity distribution, providing a physically interpretable intermediate characterization for subsequent interference parameter estimation.

[0034] S22: Based on the scattering intensity distribution, the suspended matter concentration field and turbulent velocity field are estimated, the concentration field is modeled as a learnable three-dimensional implicit representation, and a smoothing constraint is applied through the fluid continuity equation; Specifically, based on the scattering intensity distribution estimated by S21, the system further estimates the suspended matter concentration field and the turbulent velocity field. Around the welding area, the scattering intensity exhibits a gradient variation, with the forward scattering intensity weaker near the welding torch and stronger further away. Based on this, the model infers that the suspended matter concentration field is lower near the torch and higher far from it, and models this concentration field as a learnable three-dimensional implicit representation, parameterized through a multilayer perceptron. To ensure the spatial continuity of the concentration field, the model inputs the concentration field and the turbulent velocity field calculated from the point cloud sequence into the fluid continuity equation, applying a smoothing constraint with zero divergence to penalize spatial abrupt changes. For example, if an isolated peak appears in the concentration estimate in a local area, violating the continuity equation, the penalty term drives the model to adjust its parameters, making the concentration distribution conform to fluid physics laws, ultimately generating a smooth and physically reasonable concentration and velocity field estimate.

[0035] S23: Based on the estimated concentration field and velocity field, dynamically compensate for the depth measurement deviation of the point cloud data and the contrast attenuation of the visual image, and generate an enhanced perception data stream after interference suppression. Specifically, after obtaining the concentration and velocity field estimates, the system dynamically compensates for interference components in the sensor data. For visual images, based on the estimated suspended matter concentration field, the system dynamically adjusts the compensation intensity for image contrast attenuation, exponentially stretching pixel values ​​in high-concentration areas to restore the grayscale gradient at the weld bevel edge. For point cloud depth data, based on the refractive index change caused by the turbulent velocity field, the system corrects the round-trip time calculation for acoustic ranging, eliminating path curvature deviations caused by water flow. For example, when the estimated suspended matter concentration in a certain area is high, the system enhances the image contrast compensation factor for that area, making the obscured weld details reappear; simultaneously, it applies a reverse-flow offset correction to the point cloud depth measurement value for that area, restoring the welding torch-to-pipe distance reading to a stable and accurate state, ultimately generating an enhanced sensing data stream with interference suppression.

[0036] S24: Introducing a rendering consistency self-supervised mechanism, the compensated data stream is re-inputted into the propagation model to generate simulated sensor readings, and the residual loss is calculated with the original readings to update the model parameters, thus achieving unsupervised online learning; Specifically, to achieve online adaptive optimization, the system introduces a rendering consistency self-supervised mechanism. The enhanced sensing data stream compensated for S23 is re-input into the acoustic-optical coupling propagation sub-model. Based on the current concentration and velocity field estimates, the model forward renders simulated sensor readings, i.e., simulates the image and point cloud data that should be captured under the estimated interference environment. The pixel-level and point-level residual losses are calculated between the simulated readings and the original real readings. If the residuals are large, it indicates that the current interference estimation is inaccurate, and the model parameters of the concentration and velocity fields are updated through the backpropagation algorithm. For example, if the simulated rendering result of the compensated image still differs significantly from the original fogged image, the loss function drives the model to adjust the concentration field parameters to more accurately reflect the real water quality, achieving unsupervised online learning without manual annotation and continuously optimizing the interference suppression effect.

[0037] In summary, the dynamic water area interference suppression layer, by constructing a differentiable acoustic-optical coupling propagation model, achieves for the first time physical modeling and parameterized estimation of underwater operational interference, overcoming the limitations of traditional fixed filtering methods that cannot adapt to dynamic changes in water quality. The inverse deduction scattering intensity distribution mechanism provides interpretable source analysis of acoustic-optical interference, shifting interference suppression from empirical processing to being driven by physical laws. The combination of three-dimensional implicit representation and fluid continuity constraints ensures the spatial continuity and physical rationality of concentration and velocity field estimations, avoiding misjudgments of isolated noise points. The dynamic compensation mechanism can specifically eliminate the influence of interference of different regions and intensities, significantly improving the accuracy of point cloud depth measurement and image visual clarity. The rendering consistency self-supervised mechanism constructs a closed-loop optimization framework of data-model-simulation, realizing online adaptive learning without human intervention, enabling the system to continuously self-optimize as the operational water environment changes. Overall, this layer enhances the perception robustness and measurement accuracy of the monitoring system in turbid, turbulent, and other harsh water conditions, providing a reliable enhanced perception data foundation for subsequent high-precision state estimation and decision-making.

[0038] In step S30: sending the enhanced sensing data stream into the special equipment digital twin evolution network, where the special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on a physical information-driven architecture, the construction of the special equipment digital twin evolution network includes the following steps: S31: Construct a rigid-flexible coupling dynamic sub-model of the equipment, model the rigid link of the robotic arm and the flexible sealed cable as Euler-Bernoulli beam elements and nonlinear spring-damping systems respectively, and establish the equation of motion through the principle of virtual work. Specifically, in underwater pipeline welding operations, the special equipment robotic arm consists of three rigid connecting rods and two flexible sealing cables. When constructing the rigid-flexible coupled dynamic sub-model of the equipment, the three rigid connecting rods are modeled as Euler-Bernoulli beam elements, considering their bending stiffness and axial deformation; the flexible sealing cables connecting the joints are modeled as a nonlinear spring-damped system, simulating their elastic hysteresis and energy dissipation characteristics under tension and torsion. When establishing the overall motion equations using the principle of virtual work, the virtual work of inertial force of the rigid elements and the virtual work of elastic force of the flexible elements participate in the balance, accurately characterizing the end-effector posture hysteresis caused by the elastic deformation of the flexible cables when the robotic arm moves under seawater resistance, making the dynamic model closer to actual physical behavior.

[0039] S32: Embed the multi-field coupling constraint of pressure-temperature-deformation of the sealing cavity into the dynamic model, calculate the compression of the sealing ring and the leakage risk coefficient based on the thermoelastic mechanics theory, and feed it back to the dynamic equation to correct the boundary conditions; Specifically, as the welding depth increases, the external pressure of the sealed cavity increases, and the heating of internal electronic components leads to a significant temperature gradient. A multi-field coupling constraint of pressure, temperature, and deformation within the sealed cavity is embedded in the kinetic model. Based on thermoelasticity theory, the change in the compressive modulus of the fluororubber seal is calculated, and the leakage risk coefficient is assessed. For example, when the external pressure increases with depth, the model calculates an increase in the seal compression. If the temperature also rises, causing the material to soften, the leakage risk coefficient is calculated to increase. This coefficient is fed back as a boundary condition to the kinetic equations, automatically reducing the maximum allowable speed of the equipment to avoid exacerbating the risk of seal failure due to severe vibration.

[0040] S32: Using the enhanced sensing data stream as the observation input, the unscented Kalman filter algorithm is used to recursively estimate the hidden state of the dynamic model, and outputs the device joint torque, end contact force and sealing state evolution trend. In step S32, the enhanced sensing data stream output from S20 (including the compensated welding torch pose point cloud, joint angle encoder values, and strain sensor data) is used as the observation input. An unscented Kalman filter algorithm is then used to recursively estimate the hidden states of the dynamic model. Based on the difference between the observed data and the model predictions, the filter estimates the deviation of the actual joint driving torque from the theoretical value due to seawater viscous resistance. It also estimates the slight offset of the contact force between the welding torch tip and the pipe due to flexible deformation, as well as the degradation trend of the compression of the sealing ring due to fatigue accumulation. These accurate estimates of the hidden states provide reliable internal state information for subsequent health assessments.

[0041] S33: Based on the current state estimate, the performance degradation trajectory of the equipment in the remaining working time is forward-engineered, and the remaining working capacity assessment value and health index are output; Based on the current estimated trend of seal compression degradation and the wear status of the spherical bearing, the digital twin evolutionary network forward extrapolates the performance degradation trajectory over future operating periods. For example, if the current seal compression has dropped to 70% of its initial value and the degradation rate is accelerating, the model predicts that under continued current intensity operation, the compression will drop below the critical threshold after completing five weld seams. Accordingly, the remaining operational capacity assessment value is output as "five more weld seams can be completed," the health index is "below average," and a warning is given regarding the sealing status. This provides a quantitative basis for S50's early warning and strategy reconfiguration. It should be noted that in other embodiments, the naming of the health index can be flexibly defined according to different applications.

[0042] The digital twin evolution network for special equipment, through rigid-flexible coupling dynamic modeling, overcomes the limitations of traditional rigid models in depicting the elastic deformation of flexible components, significantly improving the accuracy and realism of underwater equipment dynamic simulation. The embedding of multi-field coupling constraints, for the first time, incorporates the pressure-temperature-deformation effects of sealed cavities into the monitoring scope, enabling holographic extrapolation of the health status of core sealing components and quantification of leakage risks, filling the technical gap of unobservable sealing status in underwater equipment. The unscented Kalman filter algorithm effectively integrates multi-source heterogeneous sensing data, accurately estimating the hidden states (torque, contact force, sealing compression) within the equipment, solving the problem of large state estimation errors in highly disturbed underwater environments. Forward extrapolation capabilities enable the system to perform predictive maintenance and assess remaining operational capacity, achieving a fundamental leap from passive monitoring to proactive prediction. Overall, this network improves the transparency, accuracy, and quantification of the monitoring system's status, providing a solid model foundation for the long-term reliable operation of underwater special equipment.

[0043] In step S32: embedding multi-field coupled constraints of pressure-temperature-deformation of the sealed cavity into the dynamic model, calculating the compression of the sealing ring and the leakage risk coefficient based on thermoelasticity theory, and feeding this back to the dynamic equation to correct the boundary conditions, the calculation method for the leakage risk coefficient of the sealed cavity includes: S321: Based on the thermoelasticity theory, the constitutive relationship of the sealing ring is established, and the material's compressive modulus and temperature are coupled and expressed as a time-varying function; Specifically, in underwater pipeline welding operations, fluororubber O-rings are used as static sealing elements in the sealing cavity. When establishing the constitutive relationship of the sealing ring based on thermoelastic mechanics theory, the system collects compression test data of the sealing ring material at different temperatures, revealing that its compressive modulus decreases non-linearly with increasing temperature. This relationship is coupled and expressed as a time-varying function: when the welding operation descends to deep water, and the electronic components inside the sealing cavity generate heat, causing a local temperature rise, the function automatically calculates the equivalent compressive modulus value at the current temperature, reflecting the material's softening mechanical properties in real time, and providing an accurate material parameter basis for subsequent contact stress calculations.

[0044] S322: Calculate the contact stress distribution of the sealing ring under the current working conditions, and perform vector difference with the underwater environmental pressure field to obtain the effective sealing pressure; Specifically, as the diving depth of the welding robot increases from fifty meters to one hundred meters, the underwater environmental pressure field increases linearly. When calculating the contact stress distribution of the sealing ring under the current operating conditions, the system considers the pre-compression of the O-ring in the groove and the stress concentration effect under fluid pressure to obtain the contact stress vector at each point on the sealing ring surface. This stress distribution is then subjected to vector difference calculation with the external water pressure field. For example, in the top region of the sealing ring, if the contact stress is lower than the environmental pressure, the difference is negative, indicating insufficient sealing pressure at that point; in the sidewall region, the difference is positive and relatively large, indicating sufficient effective sealing pressure. The effective sealing pressure distribution map is generated by combining the differences at each point, identifying weak points in the seal.

[0045] S323: Input the effective sealing pressure into the fatigue crack propagation sub-model, estimate the microcrack propagation rate of the sealing ring, and map it to a leakage probability score; Specifically, under long-term cyclic pressure, microcracks have already formed inside the sealing ring. When the effective sealing pressure distribution calculated by S322 is input into the fatigue crack propagation sub-model, the model, based on Paris's crack propagation law, estimates the stress intensity factor amplitude at the microcrack tip under alternating pressure difference, and then calculates the crack propagation rate. For example, when the robot frequently changes its working depth, causing increased fluctuations in sealing pressure, the model predicts a significantly faster crack propagation rate. This rate is mapped to a leakage probability score; the faster the crack propagation, the higher the score. When the score exceeds the preset "medium risk" level, it indicates a significant increase in the probability of seal failure, requiring immediate risk mitigation actions.

[0046] S324: Feed the leakage probability score back to the boundary conditions of the dynamic model. When the score exceeds the preset level, the maximum allowable operating depth and speed of the equipment are automatically reduced to achieve adaptive risk avoidance. Specifically, when the leakage probability score exceeds a preset level, the system feeds this score back into the boundary conditions of the equipment dynamics model, automatically correcting the maximum allowable load. For example, if the score exceeds the "medium risk" threshold, the boundary conditions in the dynamics model automatically reduce the maximum allowable operating depth of the robotic arm from 100 meters to 80 meters, while simultaneously reducing the end effector's movement speed from 5 degrees per second to 3 degrees per second, to mitigate the frequency and amplitude of alternating pressure on the sealing ring. This adaptive risk avoidance mechanism requires no manual intervention, achieving dynamic matching between equipment operating parameters and the seal's health status, effectively extending the seal's remaining lifespan and ensuring the safety of underwater operations.

[0047] In summary, the method for calculating the leakage risk coefficient of the sealed cavity, through thermoelastic constitutive modeling and multiphysics coupled analysis, achieves for the first time a quantifiable assessment and dynamic prediction of the underwater sealing state, overcoming the limitation of traditional periodic maintenance methods that cannot monitor sealing health in real time. Temperature-coupled time-varying material parameters enable the model to reflect material performance degradation under actual operating conditions, improving the realism of the assessment. The contact stress and ring pressure differential mechanism accurately locates weak points in the seal, providing a physical basis for risk tracing. The fatigue crack propagation sub-model correlates microscopic damage evolution with macroscopic leakage probability, achieving a leap from condition monitoring to lifespan prediction. The risk adaptive avoidance mechanism dynamically adjusts equipment operating parameters through a feedback closed loop, proactively reducing load during seal degradation, effectively preventing catastrophic underwater leakage accidents and significantly improving the inherent safety of underwater special equipment operations and the reliability of the equipment throughout its entire lifecycle.

[0048] In step S40: Aligning the state evolution features with the work process specification knowledge graph, and using an interpretable causal reasoning engine to evaluate the current work trajectory deviation and process compliance in real time, and outputting the accuracy compensation amount and process adaptive adjustment instructions, the online work accuracy evaluation and adaptive adjustment process includes the following steps: S41: Map the equipment state evolution features into a sequence of work trajectory points, perform subgraph isomorphic matching with the pre-stored process specification knowledge graph, and construct a trajectory-process alignment graph structure; S42: Run an interpretable causal inference engine on the alignment graph to identify key state variables that cause trajectory deviations through causal discovery algorithms and quantify the contribution of each variable to the deviation. S43: Based on the contribution analysis results, calculate the accuracy compensation amount of the end effector pose online, and decompose the compensation amount into joint angle fine-tuning components and path replanning components. S44: When causal reasoning discovers that multivariate coupling leads to systematic deviation, an adaptive adjustment instruction is triggered to dynamically modify the operation speed, feed rate, and tool posture process parameters to ensure that the overall process compliance is maintained above the qualified threshold.

[0049] For steps S41-S44: During underwater pipeline welding operations, the equipment state evolution characteristics show an increasing joint torque, fluctuating end contact force, and a decreasing sealing pressure trend. The system maps these characteristics to a sequence of actual working trajectory points at the welding torch tip, finding that the trajectory deviates from the standard weld path by approximately 2 mm at the pipe joint. This trajectory sequence is then matched with a pre-stored "argon arc welding process specification" knowledge graph to construct a trajectory-process alignment graph structure. The graph includes key trajectory points such as "weld start point," "joint point," and "arc end point," as well as process parameter nodes such as "welding torch attitude angle," "travel speed," and "argon flow rate." Edge relationships reveal the geometric tolerance range of the actual trajectory deviating from the standard process requirements at the "joint point" node, clearly identifying the location and type of deviation.

[0050] When running the interpretable causal inference engine on the trajectory-process alignment graph, the causal discovery algorithm identifies three key causal variables by calculating the conditional mutual information of each state variable with respect to trajectory deviation: abnormal joint torque (contribution 45%), seal pressure fluctuation (contribution 30%), and end contact force change (contribution 25%). The causal intensity heatmap clearly shows that abnormal joint torque is the primary cause, indirectly triggered by changes in connecting rod stiffness due to a decrease in seal pressure. This quantitative contribution analysis transforms the abstract phenomenon of "trajectory deviation" into a concrete "seal-torque-contact" causal chain, enabling operators to clearly understand that sealing issues should be addressed first rather than blindly adjusting the welding torch posture, achieving transparent diagnosis of the root cause.

[0051] Based on the contribution analysis results of S42, the system calculates the accuracy compensation amount of the end effector pose online. For a 2 mm trajectory deviation, the compensation amount is intelligently decomposed into two parts: a joint angle fine-tuning component (adjusting the shoulder joint by 0.5 degrees and the elbow joint by 0.3 degrees to achieve a 0.8 mm correction) and a path replanning component (adjusting the welding torch travel path to the standard trajectory in the deviation segment to achieve a 1.2 mm correction). The compensation command simultaneously includes the joint angle closed-loop control target value and the end effector spatial path replanning curve. The two work together to ensure that the welding torch smoothly returns to the standard weld trajectory, avoiding overshoot or oscillation caused by a single compensation method.

[0052] When causal reasoning detects a strong coupling between two variables—abnormal joint torque and fluctuating sealing pressure—that jointly cause a systematic deviation, the system determines that end-point compensation alone is insufficient to address the problem. At this point, an adaptive adjustment command is triggered, dynamically modifying core process parameters: reducing the welding speed from 5 mm / s to 3 mm / s to minimize dynamic force impact, fine-tuning the argon flow rate to stabilize the arc, and adjusting the welding torch angle from vertical to a 5-degree forward tilt to accommodate changes in vibration patterns caused by decreased sealing stiffness. After adjustment, the trajectory deviation stabilizes within 0.5 mm, and the process compliance recovers to above the acceptable threshold, ensuring welding quality meets standards.

[0053] The online assessment and adaptive adjustment process for operational accuracy, aligned with a trajectory-process knowledge graph, achieves, for the first time, semantic-level correlation analysis between underwater operational deviations and process specifications. This upgrades monitoring from blind error correction to intelligent adjustment that aligns with process semantics. The explainable causal reasoning engine provides transparent root cause diagnosis capabilities, decomposing complex deviation phenomena into quantifiable causal links. This avoids the shortcomings of traditional black-box models that "know only what happens, but not why," significantly improving fault location efficiency and targeted maintenance. The hierarchical compensation strategy (joint fine-tuning + path replanning) balances adjustment accuracy with system stability, avoiding secondary disturbances caused by a single compensation method. The multivariate coupling detection and process adaptive adjustment mechanism demonstrates the system's global optimization capability for systemic problems. By dynamically modifying process parameters rather than simply correcting the trajectory, it fundamentally improves process robustness and operational success rate. Overall, this process enables underwater special equipment to maintain high-precision process compliance even in harsh environments with strong interference and multiple fault coupling, significantly reducing rework rates and underwater rescue risks.

[0054] In step S50: The accuracy compensation command and the remaining operational capacity assessment value are input into the proactive fault warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical warning information, and triggers the proactive operational strategy reconfiguration command. The proactive fault warning and strategy reconfiguration process includes: S51: Encode the remaining operational capacity assessment value, the historical sequence of accuracy compensation instructions, and the evolution trend of the sealing status into a temporal feature vector, and input it into the attention mechanism network; S52: Introduce temporal sequential dependency constraints into the attention mechanism. By forcing the attention weights to follow an autoregressive process, overfitting to random noise is suppressed and attention to progressive degradation patterns is enhanced. S53: The graded warning generator outputs three levels of warning information based on the intensity of the degradation mode of attention focus: normal monitoring level, preventive maintenance level, and emergency shutdown level; S54: When the warning level reaches the preventive maintenance level, the proactive operation strategy reconfiguration command is automatically triggered. The deceleration operation, tool replacement or path detour strategy is rehearsed in the digital twin, and the optimal reconfiguration scheme is selected and executed.

[0055] Specifically, in the embodiments of this application: The proactive fault warning module continuously collects three types of time-series information: the remaining operational capability assessed by the digital twin evolutionary network shows a slow downward trend (from "can complete five welds" to "can complete three welds"); the historical sequence of accuracy compensation commands shows that the frequency of joint angle compensation increases with each successive weld (the number of compensations required to complete each weld increases); and the evolution trend of the sealing status indicates that the leakage risk score gradually increases. The system encodes the above time-series information into high-dimensional feature vectors and inputs them into an attention mechanism network, enabling the network to simultaneously perceive the multi-dimensional evolutionary patterns of equipment performance degradation, control deviation accumulation, and seal health decay, providing a comprehensive time-series representation for subsequent degradation pattern identification.

[0056] When processing the aforementioned temporal features, the attention mechanism network introduces a temporal sequential dependency constraint, forcing the calculation of attention weights to follow an autoregressive process. For example, when the sealing state exhibits a gradual increase in leakage risk, this constraint ensures that the attention weights change smoothly over consecutive time steps, focusing on the long-term trend of continuously rising risk, rather than instantaneous score fluctuations caused by accidental water flow impacts. This design effectively suppresses overfitting to random noise, enabling the network to distinguish between real degradation patterns and transient disturbances, significantly enhancing the sensitivity and stability of recognizing slowly accumulating gradual degradation patterns.

[0057] The graded early warning generator outputs three levels of early warning information based on the intensity of the degradation pattern focused by the attention mechanism: When the attention weight distribution is stable and the risk score shows no obvious upward trend, it outputs "normal monitoring level," and the system only maintains the normal monitoring frequency; when the attention detects that the seal risk score has been monotonically increasing for three consecutive hours and has exceeded the initial threshold, it outputs "preventive maintenance level," indicating that the seal status has entered an accelerated degradation period; when it detects a sharp increase in the risk score and a sudden drop in the remaining operational capacity assessment value within one hour, it outputs "emergency shutdown level," warning of an impending catastrophic seal failure. This three-level early warning system enables operators to clearly understand the equipment's health status and plan intervention measures in advance.

[0058] When the warning level reaches the "preventive maintenance level," the proactive operation strategy reconfiguration command is automatically triggered. In an underwater pipeline welding scenario, the system performs parallel simulations of three strategies in the digital twin: Strategy 1, "deceleration," reduces the welding speed by 30%, which the simulation shows extends the seal life but increases the total operation time; Strategy 2, "early termination and ascent," which the simulation shows immediately stops losses but leaves the current weld incomplete; Strategy 3, "path detour," moves the welding torch to a calmer water area to continue operation, which the simulation shows reduces the seal load and maintains operational continuity. The simulation evaluation indicates that Strategy 3 is the optimal strategy in terms of safety, economy, and operational integrity. The system selects this strategy and issues the execution command. The robot automatically adjusts its welding path to a calmer water area, successfully completes the weld task, and returns safely.

[0059] The proactive fault warning and strategy reconfiguration process, through temporal sequence constraints using temporal feature encoding and attention mechanisms, achieves accurate identification and early warning of progressive degradation patterns, solving the technical challenge of traditional post-event alarm mechanisms failing to capture slowly accumulating faults. The three-tiered warning system provides a clear risk classification, avoiding information overload and false alarm interference, enabling operators to take intervention measures as needed. The proactive strategy simulation and reconfiguration mechanism upgrades decision-making from passive response to proactive planning. By virtually testing multiple response schemes in a digital twin, the effectiveness of strategies can be evaluated and the optimal solution selected before actual execution, significantly reducing the trial-and-error costs of high-risk decisions. This process as a whole realizes a paradigm shift from "fault occurrence - shutdown and rescue" to "degradation warning - proactive avoidance," significantly improving the safety, continuity, and autonomy of underwater special equipment operations, and effectively avoiding underwater rescue costs and operational failure risks caused by sudden faults.

[0060] This application embodiment also includes an edge-cloud collaborative deployment architecture, the steps of which are as follows: S61: Deploy a lightweight water interference suppression layer and a causal inference engine on the end side of special equipment, and use a field-programmable gate array to realize the forward calculation of the acoustic-optical propagation model to meet the millisecond-level real-time requirements; S62: Runs a digital twin evolutionary network and attention warning module on the edge computing node of the unmanned vehicle mothership, and deploys the dynamic model on the graphics processor cluster through a model parallel partitioning strategy; S63: Maintain the global library of process procedure knowledge graph on the cloud platform of the remote monitoring center, and regularly collect rare working condition data from the edge side to perform incremental training. The updated model parameters are sent to the edge end through an encrypted channel. S64: Establish a three-level data flow control mechanism of cloud-edge-device. The device side only uploads a summary of abnormal working conditions, the edge device uploads a daily health report, and the cloud pulls all historical data on demand, reducing communication bandwidth consumption.

[0061] Example of an edge-cloud collaborative deployment architecture: A lightweight water interference suppression layer and a causal inference engine are deployed on the end face of the underwater pipeline welding robot. Specifically, a low-power field-programmable gate array (FPGA) chip is integrated into the robot's sealed control cabin to solidify the forward computation logic of the acoustic-optical coupling propagation model into a hardware parallel circuit. When the vision camera captures weld seam images in real time, the FPGA synchronously completes the calculation of scattering intensity distribution and image contrast compensation within milliseconds, outputting enhanced and clear weld seam features, meeting the stringent real-time requirements of the welding process for visual feedback. Simultaneously, the lightweight causal inference engine retains only the core causal link calculation unit, directly analyzing the local causal relationship between joint torque and trajectory deviation on the end face, achieving microsecond-level compensation command generation, avoiding communication delays caused by uploading all raw data to a remote location.

[0062] On the edge computing nodes of the unmanned aerial vehicle (UAV) mothership, a digital twin evolutionary network and an attention-based early warning module are deployed. The mothership's engine room is equipped with a graphics processing unit (GPU) cluster. A model-parallel partitioning strategy is used to decompose the rigid-flexible coupled dynamics model of the robotic arm into multiple sub-models based on joint degrees of freedom. Each GPU is responsible for the state estimation calculation of one sub-model. When compressed sensor data packets are uploaded from the edge, the cluster performs unscented Kalman filtering recursion in parallel to quickly reconstruct the complete state of the equipment and runs an attention mechanism network to identify sealing degradation patterns. For example, the GPU cluster can simultaneously process the dynamic equations of the six joints of a welding robot and the multi-field coupling constraints of the sealed cavity, completing state evolution estimation and remaining capacity assessment within seconds, generating early warning information and caching it for later upload, fully utilizing edge computing power to achieve real-time inference of complex models.

[0063] On a remote shore-based monitoring center cloud platform, a global database of process specification knowledge graphs containing hundreds of welding process parameters, material properties, and acceptance criteria is maintained. The cloud platform proactively collects rare operational data from different sea areas monthly, such as case data of welding failures in turbulent and murky waters, and updates the implicit parameters of the acoustic-optical propagation model through incremental training. The updated model parameters are packaged and sent to the edge terminal of the mother ship via an encrypted channel. After verifying the integrity of the parameters, the edge terminal performs a hot-load update without stopping the ongoing welding operation. This mechanism of centralized cloud maintenance and on-demand edge updates ensures the continuous evolution of the knowledge graph and the continuous improvement of the model's generalization ability.

[0064] A three-tiered data flow control mechanism (cloud-edge-device) was established to optimize communication. The welding robot at the edge only uploads a summary of abnormal operating conditions, including timestamps, risk levels, and location coordinates, to the mother ship via underwater acoustic communication when the sealing risk score exceeds the preventative maintenance threshold; the data volume is controlled to the hundreds of bytes level. The mother ship's edge terminal periodically summarizes equipment health, work completion rate, and warning frequency daily, generating a kilobyte-level daily health report which is then uploaded to the cloud. The cloud only issues commands to retrieve the full historical data stored on the edge side when it needs to reproduce rare faults or conduct in-depth audits. This mechanism reduces daily communication data volume to less than 20% of the traditional continuous upload mode, greatly alleviating the bottleneck problems of limited bandwidth and high cost in underwater acoustic communication.

[0065] The edge-cloud collaborative deployment architecture achieves real-time response capabilities for key algorithms through lightweight edge-side architecture. Hardware acceleration via field-programmable gate arrays ensures millisecond-level interference suppression and causal inference, meeting the timeliness requirements of welding closed-loop control. The parallel computing capabilities of the edge-side graphics processing unit cluster support the real-time operation of complex models such as digital twin evolutionary networks. A model partitioning strategy balances the computational load, avoiding monitoring lag caused by insufficient edge-side computing power. Centralized maintenance of the knowledge graph and execution of incremental training in the cloud enable continuous accumulation of global process experience and iterative improvement of model generalization capabilities, ensuring the long-term evolution of the monitoring system. A three-level data flow control mechanism specifically optimizes communication efficiency, prioritizing limited bandwidth resources for anomaly warnings and critical command transmission, significantly reducing operating costs. The overall architecture achieves efficient collaboration between "real-time edge processing, edge collaborative analysis, and cloud-based global optimization," significantly improving the system's deployment flexibility, operational economy, and real-time monitoring performance in underwater environments with limited communication and heterogeneous computing resources.

[0066] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0067] In one embodiment, an intelligent monitoring device for high-precision underwater special equipment operation is provided, which corresponds one-to-one with the intelligent monitoring method for high-precision underwater special equipment operation described in the above embodiments. For example... Figure 2 As shown, the intelligent monitoring device for high-precision underwater special equipment operations includes: Heterogeneous dataset construction unit 1 is used to collect multi-scale data of underwater special equipment through multimodal heterogeneous sensor arrays, and generate a heterogeneous dataset with standardized monitoring semantics after spatiotemporal registration and multi-scale data fusion. The enhanced sensing data stream generation unit 2 is used to input the heterogeneous dataset into the dynamic water area interference suppression layer. The dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The status trend and capability assessment unit 3 is used to send the enhanced perception data stream into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment status evolution trend characteristics and remaining operational capability assessment value. The adaptive adjustment instruction generation unit 4 is used to align the state evolution characteristics with the work process procedure knowledge graph, evaluate the current work trajectory deviation and process compliance in real time through an interpretable causal reasoning engine, and output the accuracy compensation amount and process adaptive adjustment instruction. The graded early warning information generation unit 5 is used to input the accuracy compensation instruction and the remaining operation capacity assessment value into the fault proactive early warning module. This module introduces time-sequential dependency constraints in the attention mechanism, generates graded early warning information, and triggers the proactive operation strategy reconstruction instruction.

[0068] Specific limitations regarding the intelligent monitoring device for high-precision underwater special equipment operations can be found in the limitations of the intelligent monitoring method for high-precision underwater special equipment operations described above, and will not be repeated here. Each module in the aforementioned intelligent monitoring device for high-precision underwater special equipment operations can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device, or stored in the memory of the electronic device as software, so that the processor can call and execute the corresponding operations of each module.

[0069] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the database. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent monitoring method for high-precision underwater special equipment operations.

[0070] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Multi-scale data of underwater special equipment is collected by multi-modal heterogeneous sensor arrays. After spatiotemporal registration and fusion of multi-scale data, a heterogeneous dataset with standardized monitoring semantics is generated. The heterogeneous dataset is input into the dynamic water area interference suppression layer, which estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The enhanced sensing data stream is sent into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment state evolution trend characteristics and remaining working capacity assessment value. The state evolution features are aligned with the knowledge graph of the operation process procedure. An interpretable causal reasoning engine is used to evaluate the deviation of the current operation trajectory and the compliance with the process in real time, and outputs the accuracy compensation amount and the process adaptive adjustment instruction. The accuracy compensation command and the remaining work capacity assessment value are input into the fault proactive early warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical early warning information, and triggers proactive work strategy reconstruction commands.

[0071] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Multi-scale data of underwater special equipment is collected by multi-modal heterogeneous sensor arrays. After spatiotemporal registration and fusion of multi-scale data, a heterogeneous dataset with standardized monitoring semantics is generated. The heterogeneous dataset is input into the dynamic water area interference suppression layer, which estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The enhanced sensing data stream is sent into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment state evolution trend characteristics and remaining working capacity assessment value. The state evolution features are aligned with the knowledge graph of the operation process procedure. An interpretable causal reasoning engine is used to evaluate the deviation of the current operation trajectory and the compliance with the process in real time, and outputs the accuracy compensation amount and the process adaptive adjustment instruction. The accuracy compensation command and the remaining work capacity assessment value are input into the fault proactive early warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical early warning information, and triggers proactive work strategy reconstruction commands.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method and system for intelligent monitoring of high-precision underwater special equipment operations, characterized in that, The method includes the following steps: collecting multi-scale data of underwater special equipment through a multi-modal heterogeneous sensor array, and generating a heterogeneous dataset with standardized monitoring semantics after spatiotemporal registration and fusion of multi-scale data; The heterogeneous dataset is input into the dynamic water area interference suppression layer, which estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The enhanced sensing data stream is sent into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture, and embeds multi-field coupling constraints of pressure-temperature-deformation in the sealed cavity, and outputs the equipment state evolution trend characteristics and remaining working capacity assessment value. The state evolution features are aligned with the knowledge graph of the operation process procedure. An interpretable causal reasoning engine is used to evaluate the deviation of the current operation trajectory and the compliance with the process in real time, and outputs the accuracy compensation amount and the process adaptive adjustment instruction. The accuracy compensation command and the remaining work capacity assessment value are input into the fault proactive early warning module. This module introduces time-sequential dependency constraints into the attention mechanism, generates hierarchical early warning information, and triggers proactive work strategy reconstruction commands.

2. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 1, characterized in that, The step of generating a heterogeneous dataset with standardized monitoring semantics after collecting multi-scale data from underwater special equipment through a multimodal heterogeneous sensor array, spatiotemporal registration, and multi-scale data fusion includes the following steps: A nine-axis inertial measurement unit, a fiber optic strain sensor, and an ultrasonic ranging probe are deployed on the body of the special equipment, and a microelectromechanical system attitude sensor and a vision camera are integrated in the end effector. A submodule for underwater acoustic communication channel detection is constructed. By sending orthogonal frequency division multiplexing pilot signals and receiving echoes, the multipath delay spread and Doppler frequency shift parameters of the channel are estimated. The collected body posture data, end pose data, environmental point cloud data and channel status data are timestamped and aligned. A dynamic time warping algorithm is used to eliminate timing mismatch caused by long delays in underwater acoustic communication. The aligned heterogeneous data are fused into multi-scale feature volumes under a unified monitoring semantic framework to generate a standardized heterogeneous dataset containing spatial geometric information, mechanical state information, and channel quality information.

3. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 1, characterized in that, The step of inputting the heterogeneous dataset into the dynamic water area interference suppression layer, where the dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the operating area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression, includes the following steps: A differentiable underwater acoustic-optical coupling propagation sub-model was constructed, using point cloud data of the operating area as spatial constraints, to inversely deduce the forward and backward scattering intensity distributions of sound and light waves in the suspended particle swarm. The suspended matter concentration field and turbulent velocity field are estimated based on the scattering intensity distribution. The concentration field is modeled as a learnable three-dimensional implicit representation, and a smoothing constraint is applied through the fluid continuity equation. Based on the estimated concentration field and velocity field, the depth measurement deviation of the point cloud data and the contrast attenuation of the visual image are dynamically compensated to generate an enhanced perception data stream after interference suppression. A rendering consistency self-supervised mechanism is introduced, in which the compensated data stream is re-input into the propagation model to generate simulated sensor readings, and the residual loss is calculated with the original readings to update the model parameters, thereby achieving unsupervised online learning.

4. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 1, characterized in that, In the step of sending the enhanced sensing data stream into the special equipment digital twin evolution network, and in which the special equipment digital twin evolution network constructs a rigid-flexible coupled dynamic model of the equipment based on a physical information-driven architecture, the construction of the special equipment digital twin evolution network includes the following steps: A rigid-flexible coupling dynamic sub-model of the equipment is constructed. The rigid link of the robotic arm and the flexible sealed cable are modeled as Euler-Bernoulli beam elements and nonlinear spring-damping systems, respectively. The equations of motion are established through the principle of virtual work. The pressure-temperature-deformation multi-field coupling constraint of the sealing cavity is embedded in the dynamic model. The compression of the sealing ring and the leakage risk coefficient are calculated based on the thermoelastic mechanics theory and fed back to the dynamic equation to correct the boundary conditions. Using the enhanced sensing data stream as the observation input, the unscented Kalman filter algorithm is used to recursively estimate the hidden state of the dynamic model, and outputs the evolution trend of the joint torque, end contact force and sealing state of the equipment. Based on the current state estimate, the performance degradation trajectory of the equipment is projected forward over the remaining operating time, and the remaining operating capacity assessment value and health index are output.

5. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 4, characterized in that, In the step of embedding multi-field coupled constraints of pressure, temperature, and deformation of the sealed cavity into the dynamic model, calculating the compression of the sealing ring and the leakage risk coefficient based on thermoelasticity theory, and feeding this back to the dynamic equations to correct the boundary conditions, the calculation method for the leakage risk coefficient of the sealed cavity includes: Based on the thermoelastic mechanics theory, the constitutive relation of the sealing ring is established, and the material's compressive modulus and temperature are coupled and expressed as a time-varying function. Calculate the contact stress distribution of the sealing ring under the current working conditions, and perform vector difference with the underwater environmental pressure field to obtain the effective sealing pressure; The effective sealing pressure is input into the fatigue crack propagation sub-model to estimate the microcrack propagation rate of the sealing ring and map it to a leakage probability score. The leakage probability score is fed back to the boundary conditions of the dynamic model. When the score exceeds the preset level, the maximum allowable operating depth and speed of the equipment are automatically reduced to achieve adaptive risk avoidance.

6. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 1, characterized in that, In the step of aligning the state evolution features with the knowledge graph of the work process specifications, and using an interpretable causal reasoning engine to evaluate the current work trajectory deviation and process compliance in real time, and outputting accuracy compensation and adaptive adjustment instructions for the process, the online work accuracy evaluation and adaptive adjustment process includes the following steps: The equipment state evolution features are mapped into a sequence of operation trajectory points, and subgraph isomorphic matching is performed with the pre-stored process specification knowledge graph to construct a trajectory-process alignment graph structure. Run an interpretable causal inference engine on the alignment graph to identify key state variables that cause trajectory deviations through causal discovery algorithms and quantify the contribution of each variable to the deviation. Based on the contribution analysis results, the accuracy compensation amount of the end effector pose is calculated online, and the compensation amount is decomposed into joint angle fine-tuning component and path replanning component. When causal reasoning discovers that multivariate coupling leads to systematic deviations, an adaptive adjustment instruction is triggered to dynamically modify the process parameters such as operating speed, feed rate, and tool posture, ensuring that the overall process compliance is maintained above the qualified threshold.

7. The intelligent monitoring method and system for high-precision underwater special equipment operation according to claim 1, characterized in that, In the step of inputting the accuracy compensation command and the remaining operational capacity assessment value into the proactive fault warning module, which introduces time-sequential dependency constraints into the attention mechanism to generate hierarchical warning information and trigger proactive operational strategy reconfiguration commands, the proactive fault warning and strategy reconfiguration process includes: The remaining operational capacity assessment value, the historical sequence of accuracy compensation instructions, and the evolution trend of the sealing status are encoded into temporal feature vectors and input into the attention mechanism network; By introducing temporal sequential dependency constraints into the attention mechanism, the attention weights are forced to follow an autoregressive process, which suppresses overfitting to random noise and enhances attention to progressive degradation patterns. The graded warning generator outputs three levels of warning information based on the intensity of the degradation mode of attention focus: normal monitoring level, preventive maintenance level, and emergency shutdown level; When the warning level reaches the preventive maintenance level, the proactive operation strategy reconfiguration command is automatically triggered. The deceleration operation, tool replacement or path detour strategy is rehearsed in the digital twin, and the optimal reconfiguration scheme is selected and executed.

8. An intelligent monitoring device for high-precision underwater special equipment operation, applied to the intelligent monitoring method for high-precision underwater special equipment operation as described in any one of claims 1 to 7, characterized in that, The device includes: Heterogeneous dataset construction unit (1) is used to collect multi-scale data of underwater special equipment through multi-modal heterogeneous sensor array, and generate a heterogeneous dataset with standardized monitoring semantics after spatiotemporal registration and multi-scale data fusion. The enhanced sensing data stream generation unit (2) is used to input the heterogeneous dataset into the dynamic water area interference suppression layer. The dynamic water area interference suppression layer estimates the suspended matter concentration field and turbulent velocity field of the working area in real time through a differentiable underwater acoustic-optical coupling propagation model, and generates an enhanced sensing data stream after interference suppression. The status trend and capability assessment unit (3) is used to send the enhanced perception data stream into the special equipment digital twin evolution network. The special equipment digital twin evolution network constructs a rigid-flexible coupling dynamic model of the equipment based on the physical information driven architecture and embeds multi-field coupling constraints of pressure-temperature-deformation of the sealed cavity, and outputs the status evolution trend characteristics and remaining operational capability assessment value of the equipment. The adaptive adjustment instruction generation unit (4) is used to align the state evolution characteristics with the knowledge graph of the operation process procedure, evaluate the deviation of the current operation trajectory and the process compliance in real time through the interpretable causal reasoning engine, and output the accuracy compensation amount and the adaptive adjustment instruction of the process. The graded early warning information generation unit (5) is used to input the accuracy compensation instruction and the remaining operation capability assessment value into the fault proactive early warning module. This module introduces time-sequential dependency constraints in the attention mechanism, generates graded early warning information, and triggers the proactive operation strategy reconstruction instruction.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent monitoring method for high-precision underwater special equipment operation as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent monitoring method for high-precision underwater special equipment operation as described in any one of claims 1 to 7.

Images