Expansion joint dynamic sensing system based on GIL multi-dimensional sensing

Abstract

The invention discloses an expansion joint dynamic sensing system based on GIL multi-dimensional sensing, and relates to the technical field of expansion joint sensing, the system comprises a data acquisition module, a multi-modal fusion module, a self-adaptive sensing module, a self-adaptive optimization module, a state evaluation module and an application interface module; the data acquisition module is used for acquiring multi-modal data of the expansion joint; the multi-modal fusion module is used for fusing different modal data and outputting integrated features after fusion; the self-adaptive sensing module is used for sensing an environment state and constructing a sensing model; the adaptive optimization module optimizes parameters of the perception model in real time in combination with historical data and real-time input; the state evaluation module is used for evaluating and predicting the state of the expansion joint; and the application interface module provides real-time monitoring, fault diagnosis and prediction maintenance. According to the invention, environment state sensing is carried out, real-time optimization is carried out according to actual conditions, and the capability of judging the state of the expansion joint is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of expansion joint sensing, and specifically to a dynamic sensing system for expansion joints based on multi-dimensional sensing of GIL. Background Art

[0002] As an efficient power transmission device, GIL is widely used in the field of power transmission. The expansion joint, as a key component of GIL, its performance directly affects the stability and reliability of the entire power transmission system. Since GIL usually operates in a complex environment, the expansion joint is prone to failure due to various factors, and these failures may lead to power transmission interruption, causing serious losses to the power system.

[0003] Traditional GIL expansion joint monitoring methods mainly rely on manual inspection and simple sensor monitoring. Manual inspection has problems such as low efficiency, strong subjectivity, and inability to monitor in real time, and it is difficult to meet the requirements of modern power systems for reliability and real-time performance. With the development of sensor technology, although more multi-modal data about the expansion joint can be obtained, how to effectively process and analyze these data has become a new challenge. Traditional data processing methods are difficult to fuse and deeply mine multi-modal data, and cannot make full use of the information in the data, resulting in low accuracy in the evaluation and prediction of the expansion joint state. Summary of the Invention

[0004] The purpose of the present invention is to provide a dynamic sensing system for expansion joints based on multi-dimensional sensing of GIL to solve the problems raised in the prior art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A dynamic sensing system for expansion joints based on multi-dimensional sensing of GIL, the system includes: a data acquisition module, a multi-modal fusion module, an adaptive sensing module, an adaptive optimization module, a state evaluation module, and an application interface module;

[0006] The data acquisition module is used to collect multi-modal data about the expansion joint, and the multi-modal data includes environmental data, vibration data, sound data, and displacement data; and preprocess the multi-modal data;

[0007] The multi-modal fusion module, based on the preprocessed multi-modal data, performs fusion of different modal data and outputs the integrated features after fusion;

[0008] The adaptive sensing module, based on the integrated features after fusion, uses reinforcement learning to perform environmental state sensing and constructs a sensing model;

[0009] The adaptive optimization module, based on the adaptive sensing module, through online learning technology, combines historical data and real-time input to optimize the parameters of the sensing model in real time;

[0010] The state evaluation module performs state evaluation and prediction of the expansion joint based on the optimized perception data.

[0011] The application interface module is used to provide real-time monitoring, fault diagnosis, and predictive maintenance to the user.

[0012] According to the above solution, the data acquisition module includes an environmental monitoring unit, a vibration monitoring unit, a sound monitoring unit, a displacement monitoring unit, and a data preprocessing unit.

[0013] The environmental monitoring unit is used to collect temperature, humidity, and pressure data of the environment around the expansion joint in real time; the vibration monitoring unit is used to collect vibration signals of the expansion joint in real time; the sound monitoring unit is used to collect sound signals of the expansion joint in real time; the displacement monitoring unit is used to collect deformation and displacement data of the expansion joint in real time; according to the changes in multimodal data, the sampling frequency is dynamically adjusted, and the sensors are calibrated regularly to ensure the accuracy of the collected data.

[0014] The data preprocessing unit preprocesses the collected multimodal data regarding the expansion joint; the preprocessing includes missing value processing, outlier detection and correction, and data standardization processing for the temperature, humidity, and pressure data of the environment around the expansion joint, time domain processing and frequency domain processing for the vibration signals of the expansion joint, noise reduction processing and feature extraction for the sound signals of the expansion joint, and smoothing processing and data difference processing for the deformation and displacement data of the expansion joint; the data preprocessing unit also includes interpolating and filling in the missing data to ensure data integrity; detecting outliers using statistical methods and correcting or removing them; and standardizing different modal data to make the different modal data within the same numerical range.

[0015] According to the above solution, the multimodal fusion module includes a multimodal deep learning model and an attention mechanism unit.

[0016] The multimodal deep learning model includes a Transformer model sub-unit and a graph neural network model sub-unit; the Transformer model sub-unit generates sequence data features representing the comprehensive features of time series data; the graph neural network model sub-unit generates graph structure data features representing the comprehensive relationship between multimodal data.

[0017] The attention mechanism unit receives the sequence data features of the Transformer model subunit and the graph structure data features of the graph neural network model subunit, calculates the correlation scores of each modality data with other modality data, where each modality data includes the sequence data features processed by the Transformer model subunit and the graph structure data features processed by the graph neural network model subunit; converts the correlation scores into attention weights through the Softmax function, and the higher the attention weight, the greater the importance of the modality data in the current environment; multiplies the features of each modality data by the corresponding attention weight to obtain weighted features; concatenates or sums the weighted features to generate a fused comprehensive feature.

[0018] According to the above solution, the Transformer model subunit receives the preprocessed multi-modal data, arranges it in chronological order, embeds the data of each modality into a high-dimensional vector space respectively, and uses the self-attention mechanism to calculate the attention weights between different time steps to capture the dependencies within the sequence; through multiple attention heads, captures the relationships between different modality data, and generates sequence data features representing the comprehensive features of the time series data.

[0019] The graph neural network model subunit receives the preprocessed multi-modal data, takes the data of each modality as a node, takes the relationships between different modality data as edges, aggregates the information of nodes and edges through a graph convolutional layer, and generates graph structure data features representing the comprehensive relationships between multi-modal data.

[0020] According to the above solution, the adaptive perception module includes a reinforcement learning unit and a perception model unit.

[0021] The reinforcement learning unit includes:

[0022] Receives the fused comprehensive feature, and takes the fused comprehensive feature as the input state of the reinforcement learning agent; takes the adjustable parameters of the perception model as the action space of the reinforcement learning agent, where the adjustable parameters include the sampling frequency of the multi-modal data of the telescopic joint, the weight coefficient of multi-modal data fusion, and the threshold of the perception model; gives a reward according to the judgment accuracy of the perception model on the telescopic joint environment state and its own state.

[0023] The reinforcement learning agent selects an action to execute from the policy network according to the current state; the policy network guides the reinforcement learning agent to make action selections by learning the mapping relationship between states and actions.

[0024] Based on the action selection, combined with the feedback of the environment, a new state and a reward value are obtained.

[0025] Form an experience tuple consisting of the state, action, reward, and new state, and store it in the experience replay buffer; periodically sample a number of experience tuples randomly from the experience replay buffer for updating the parameters of the policy network and value network, and learning the optimal action selection under different states;

[0026] Pass the optimal action selection under the different states to the perception model unit;

[0027] The perception model unit includes:

[0028] Based on the optimal action selection under different states learned by the reinforcement learning unit, adjust the adjustable parameters of the perception model to construct a perception model for evaluating and predicting the telescopic joint environment state and its own state.

[0029] According to the above solution, the reward includes giving a positive reward when the perception model accurately predicts the normal operation state, abnormal state, or fault type of the telescopic joint, and giving a negative reward when the perception model incorrectly predicts the normal operation state, abnormal state, or fault type of the telescopic joint; the reward is obtained by calculating the prediction accuracy and recall rate through comparison with the actual state.

[0030] According to the above solution, the adaptive optimization module includes an online learning unit and a model performance evaluation unit;

[0031] The online learning unit includes:

[0032] Receive the perception model constructed by the adaptive perception module;

[0033] Receive the fused comprehensive features of the multi-modal fusion module in real time, fuse them with historical data to form a dynamic data set;

[0034] Add the fused comprehensive features in the recent period of the dynamic data set to the sliding window to ensure that the perception model can quickly adapt to the latest environmental changes;

[0035] Adopt an online learning algorithm to incrementally update the perception model. At the same time, introduce a forgetting factor to reduce the impact of old data on the perception model and avoid the perception model from becoming obsolete;

[0036] Real-time detect whether there is data abnormality in the fused comprehensive features, and dynamically correct the perception model parameters according to the anomaly detection results to avoid the perception model from drifting;

[0037] The model performance evaluation unit includes:

[0038] Based on the online learning unit, adjust the adjustable parameters of the perception model in real time;

[0039] Verify the perception model using the dataset, calculate the evaluation metric values, and evaluate the generalization ability and accuracy of the model;

[0040] Feed back the evaluation metric values to the adaptive perception module to further adjust the adjustable parameters of the perception model;

[0041] When the evaluation metric values no longer show significant improvement in consecutive multiple optimization cycles, stop the optimization;

[0042] Output the perception model optimized by the adaptive optimization module for real-time evaluation of the environmental state and the state of the expansion joint.

[0043] According to the above solution, the state evaluation module includes a state evaluation unit and a fault prediction unit;

[0044] The state evaluation unit includes:

[0045] Input the comprehensively collected and fused features in real time into the perception model;

[0046] Based on the perception data of the perception model, classify the state of the expansion joint and predict the environmental state;

[0047] The state of the expansion joint includes normal, slightly abnormal, and severely abnormal; the prediction of the environmental state includes predicted values of temperature, humidity, and pressure data;

[0048] The state evaluation unit outputs the current state of the expansion joint and the prediction result of the environmental state;

[0049] The fault prediction unit includes:

[0050] Based on the state evaluation unit, classify the fault type of the expansion joint, output the fault type, and predict the time of fault occurrence;

[0051] The fault prediction unit outputs the fault type and the fault prediction result;

[0052] The fault types include deformation exceeding limit, vibration abnormality, sound abnormality, temperature abnormality, humidity abnormality, pressure abnormality, displacement abnormality, material fatigue, and seal failure.

[0053] According to the above solution, the application interface module includes a real-time monitoring unit, a fault diagnosis unit, and a predictive maintenance unit;

[0054] The real-time monitoring unit uses dynamic graphs to display the multi-modal data of the expansion joint, updates the expansion joint data in real time, allowing users to observe the changing trend of the data in real time; it displays the current state of the expansion joint and differentiates it using different colors, including green indicating normal, yellow indicating slight abnormality, and red indicating severe abnormality.

[0055] When the expansion joint is in an abnormal state, the fault diagnosis unit details the fault information; the fault information includes the fault type, fault level, and fault location, and also displays the occurrence situation, handling measures, and handling effects of historical faults of the same type; it shows the time prediction of the fault occurrence and the cause analysis of the fault, and the cause analysis is obtained based on the correlation analysis of the output of the perception model and historical fault data.

[0056] The predictive maintenance unit formulates a personalized predictive maintenance plan by combining the operation history and maintenance records of the expansion joint; the predictive maintenance plan includes the time nodes for maintenance, maintenance equipment, and maintenance components; it sets maintenance reminders, and according to the set maintenance reminders, before the time nodes for maintenance, it displays the predictive maintenance plan in the form of an interface pop-up window to remind users to perform maintenance operations.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. The present invention more comprehensively reflects the operating state of the expansion joint through multi-modal data.

[0059] 2. The present invention deeply fuses multi-modal data to generate more representative comprehensive features, improving the comprehensiveness and accuracy of the state monitoring of the expansion joint.

[0060] 3. The present invention uses reinforcement learning for environmental state perception and optimizes in real time according to the actual situation, improving the ability to judge the state of the expansion joint under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic structural diagram of a dynamic perception system for an expansion joint based on GIL multi-dimensional perception of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] Embodiment: As Figure 1As shown, the present invention provides a technical solution, a dynamic sensing system for expansion joints based on GIL multi-dimensional sensing. The system includes: a data acquisition module, a multi-modal fusion module, an adaptive sensing module, an adaptive optimization module, a state evaluation module, and an application interface module;

[0064] The data acquisition module is used to collect multi-modal data about the expansion joint. The multi-modal data includes environmental data, vibration data, sound data, and displacement data; and preprocess the multi-modal data;

[0065] Specifically, the data acquisition module includes an environmental monitoring unit, a vibration monitoring unit, a sound monitoring unit, a displacement monitoring unit, and a data preprocessing unit;

[0066] Furthermore, the environmental monitoring unit is used to collect the temperature, humidity, and pressure data of the environment around the expansion joint in real time; the vibration monitoring unit is used to collect the vibration signal of the expansion joint in real time; the sound monitoring unit is used to collect the sound signal of the expansion joint in real time; the displacement monitoring unit is used to collect the deformation and displacement data of the expansion joint in real time; according to the change of multi-modal data, dynamically adjust the sampling frequency, and regularly calibrate the sensors to ensure the accuracy of the collected data; the data preprocessing unit preprocesses the multi-modal data about the expansion joint collected. The preprocessing includes missing value processing, outlier detection and correction, and data standardization processing for the temperature, humidity, and pressure data of the environment around the expansion joint, time-domain processing and frequency-domain processing for the vibration signal of the expansion joint, noise reduction processing and feature extraction for the sound signal of the expansion joint, and smoothing processing and data difference processing for the deformation and displacement data of the expansion joint; the data preprocessing unit also includes interpolating and filling the missing data to ensure the integrity of the data; detecting outliers using statistical methods and correcting or removing them; standardizing the data of different modalities to make the data of different modalities in the same numerical range.

[0067] The multi-modal fusion module performs fusion of different modality data based on the preprocessed multi-modal data and outputs the integrated features after fusion;

[0068] Specifically, the multi-modal fusion module includes a multi-modal deep learning model and an attention mechanism unit;

[0069] Furthermore, the multimodal deep learning model includes a Transformer model subunit and a graph neural network model subunit; the Transformer model subunit generates sequence data features representing the comprehensive features of time series data; the graph neural network model subunit generates graph structure data features representing the comprehensive relationship between multimodal data; the attention mechanism unit receives the sequence data features of the Transformer model subunit and the graph structure data features of the graph neural network model subunit, calculates the correlation scores between each modality data and other modality data, and each modality data includes the sequence data features processed by the Transformer model subunit and the graph structure data features processed by the graph neural network model subunit; the correlation scores are converted into attention weights through the Softmax function, and the higher the attention weight, the greater the importance of the modality data in the current environment; the features of each modality data are multiplied by the corresponding attention weights to obtain weighted features; the weighted features are concatenated or summed to generate the fused comprehensive features; for example: the attention mechanism unit calculates that the attention weight of vibration data in the current environment is 0.4, the sound data is 0.3, the displacement data is 0.2, and the environmental data is 0.1; the features of each modality are multiplied by the attention weights and then concatenated to obtain the fused comprehensive feature.

[0070] Specifically, the Transformer model subunit receives the preprocessed multimodal data, arranges it in chronological order, embeds the data of each modality into a high-dimensional vector space respectively, and uses the self-attention mechanism to calculate the attention weights between different time steps to capture the dependencies within the sequence; through multiple attention heads, the relationships between different modality data are captured to generate sequence data features representing the comprehensive features of time series data; for example: the Transformer model subunit sets 8 attention heads, each with a dimension of 64, and captures the data dependencies through the self-attention mechanism; the graph neural network model subunit receives the preprocessed multimodal data, takes the data of each modality as a node, takes the relationships between different modality data as edges, and aggregates the information of nodes and edges through graph convolutional layers to generate graph structure data features representing the comprehensive relationship between multimodal data; for example: the graph neural network model subunit adopts the GraphSAGE algorithm and sets 3 graph convolutional layers, and each layer aggregates the information of neighbor nodes.

[0071] The adaptive perception module, based on the fused comprehensive features, uses reinforcement learning for environmental state perception and constructs a perception model.

[0072] Specifically, the adaptive perception module includes a reinforcement learning unit and a perception model unit.

[0073] Further, the reinforcement learning unit includes: receiving the fused comprehensive features and using the fused comprehensive features as the input state of the reinforcement learning agent; taking the adjustable parameters of the perception model as the action space of the reinforcement learning agent, where the adjustable parameters include the sampling frequency of the multi-modal data of the expansion joint, the weight coefficient of the multi-modal data fusion, and the threshold of the perception model; giving a reward according to the judgment accuracy of the perception model on the environmental state and its own state of the expansion joint; the reinforcement learning agent selects an action to execute from the policy network according to the current state; the policy network guides the reinforcement learning agent to make action selections by learning the mapping relationship between states and actions; based on the action selection and combined with the feedback of the environment, a new state and a reward value are obtained; forming an experience tuple with the state, action, reward, and new state and storing it in the experience replay buffer; randomly sampling several experience tuples from the experience replay buffer regularly to update the parameters of the policy network and the value network and learn the optimal action selections in different states; transmitting the optimal action selections in different states to the perception model unit;

[0074] Specifically, the reward includes giving a positive reward when the perception model accurately predicts the normal operation state, abnormal state, or fault type of the expansion joint, and giving a negative reward when the perception model incorrectly predicts the normal operation state, abnormal state, or fault type of the expansion joint; the reward is obtained by calculating the prediction accuracy and recall rate through comparison with the actual state; for example, when the perception model accurately predicts the normal operation state of the expansion joint, a positive reward of +1 is given, and a negative reward of -1 is given when the prediction is incorrect;

[0075] Further, the perception model unit includes: based on the optimal action selections in different states learned by the reinforcement learning unit, adjusting the adjustable parameters of the perception model to construct a perception model for evaluating and predicting the environmental state and its own state of the expansion joint.

[0076] The adaptive optimization module, based on the adaptive perception module, uses online learning technology to combine historical data and real-time input to optimize the parameters of the perception model in real time;

[0077] Specifically, the adaptive optimization module includes an online learning unit and a model performance evaluation unit;

[0078] Furthermore, the online learning unit includes: receiving the perception model constructed by the adaptive perception module; receiving in real time the fused comprehensive features of the multi-modal fusion module, fusing them with historical data to form a dynamic data set; adding the fused comprehensive features in the dynamic data set for a recent period of time to a sliding window to ensure that the perception model can quickly adapt to the latest environmental changes; using an online learning algorithm to incrementally update the perception model. At the same time, a forgetting factor is introduced to reduce the impact of old data on the perception model and prevent the perception model from becoming outdated; detecting in real time whether there are data anomalies in the fused comprehensive features, and dynamically correcting the parameters of the perception model according to the anomaly detection results to prevent the perception model from drifting; for example: the online learning unit uses the Adam algorithm, with a learning rate of 0.001 and a forgetting factor of 0.95; receiving in real time the fused comprehensive features, fusing them with 1000 historical data to form a dynamic data set; putting the data for the most recent 100 time steps into the sliding window and updating the perception model in real time; when it is found that the vibration data is abnormal at a certain moment (the vibration amplitude far exceeds the normal range), dynamically correcting the vibration-related parameters in the perception model according to the anomaly detection results to prevent model drift.

[0079] Furthermore, the model performance evaluation unit includes: based on the online learning unit, adjusting in real time the adjustable parameters of the perception model; using a data set to verify the perception model, calculating the evaluation metric values, and evaluating the generalization ability and accuracy of the model; feeding back the evaluation metric values to the adaptive perception module to further adjust the adjustable parameters of the perception model; when the evaluation metric values no longer show significant improvement in a continuous number of optimization cycles, stop the optimization; for example: using a validation set containing 500 data to verify the perception model, and calculating metrics such as accuracy, recall rate, and F1 value. Stop the optimization when the F1 value improvement is less than 0.01 in 10 consecutive optimization cycles.

[0080] Furthermore, output the perception model optimized by the adaptive optimization module for real-time evaluation of the environmental state and the expansion joint state.

[0081] The state evaluation module performs state evaluation and prediction of the expansion joint based on the optimized perception data;

[0082] Specifically, the state evaluation module includes a state evaluation unit and a fault prediction unit;

[0083] Furthermore, the state evaluation unit includes: inputting the comprehensively collected and fused real-time features into the perception model; classifying the state of the expansion joint based on the perception data of the perception model and predicting the environmental state; the state of the expansion joint includes normal, slightly abnormal, and severely abnormal; predicting the environmental state includes predicted values of temperature, humidity, and pressure data; the state evaluation unit outputs the current state of the expansion joint and the prediction result of the environmental state; for example, at a certain moment, the perception model determines that the expansion joint is in a slightly abnormal state, the predicted value of the environmental temperature is 32 °C, the predicted value of the humidity is 62% RH, and the predicted value of the pressure is 101.5 kPa; after prediction, there is a 20% probability of a displacement abnormal fault within the next 24 hours, and the fault may occur between 18 - 22 hours from the current time;

[0084] Furthermore, the fault prediction unit includes: classifying the fault types of the expansion joint based on the state evaluation unit, outputting the fault types, and predicting the time of fault occurrence; the fault prediction unit outputs the fault types and the fault prediction results; the fault types include deformation exceeding the limit, abnormal vibration, abnormal sound, abnormal temperature, abnormal humidity, abnormal pressure, abnormal displacement, material fatigue, and seal failure.

[0085] The application interface module is used to provide real-time monitoring, fault diagnosis, and predictive maintenance to the user.

[0086] Specifically, the application interface module includes a real-time monitoring unit, a fault diagnosis unit, and a predictive maintenance unit;

[0087] Furthermore, the real-time monitoring unit uses a dynamic graph to display the multi-modal data of the expansion joint, updates the expansion joint data in real time, allows the user to observe the change trend of the data in real time; displays the current state of the expansion joint and differentiates it with different colors, including green indicating normal, yellow indicating slightly abnormal, and red indicating severely abnormal;

[0088] Furthermore, when the expansion joint is in an abnormal state, the fault diagnosis unit details the fault information; the fault information includes the fault type, fault level, and fault location, and displays the occurrence situation, handling measures, and handling effects of historical faults of the same type; displays the predicted time of fault occurrence and the cause analysis of the fault, and the cause analysis is obtained based on the correlation analysis of the output of the perception model and historical fault data; for example, when the expansion joint is abnormal, such as a severely abnormal state marked in red, the fault diagnosis unit displays the fault type, such as abnormal vibration, the fault level is severe, the fault location is at the A expansion joint of the GIL line, and at the same time displays that the historical fault of the same type occurred 3 months ago, the handling measure was to repair the expansion joint, and the handling effect was good; the cause analysis is that due to the sudden temperature change recently, equipment resonance occurred.

[0089] Further, the predictive maintenance unit formulates a personalized predictive maintenance plan by combining the operation history and maintenance records of the expansion joint; the predictive maintenance plan includes the time nodes for maintenance, the equipment to be maintained, and the components to be maintained; a maintenance reminder is set, and according to the set maintenance reminder, before the time node for maintenance, the predictive maintenance plan is displayed in the form of an interface pop-up window to remind the user to perform maintenance operations; for example: according to the prediction that a seal failure may occur within the next week, it is planned to inspect and maintain the seal components on the third day, and a maintenance reminder is set to remind the maintenance personnel through an interface pop-up window before the third day.

[0090] The present invention provides another technical solution, a dynamic sensing system for expansion joints based on GIL multi-dimensional sensing;

[0091] The user monitors the GIL expansion joints in an important transmission line through a dynamic sensing system for expansion joints based on GIL multi-dimensional sensing;

[0092] Specifically, the user opens the real-time monitoring unit in the application interface module of the system and views the multi-modal data in the form of a visual dynamic graph. The data is updated once per second. For example, a line graph shows the changes in temperature and vibration amplitude over time, and a bar graph presents the displacement change amounts during different time periods, etc.; and the normal range is given. For example, under normal conditions, the ambient temperature around the expansion joint remains at about 25°C, the vibration amplitude is within 5 m / s 2 and the displacement change amount is within the range of ±2 mm.

[0093] Further, the real-time monitoring unit visually distinguishes the current state of the expansion joint with colors. Green indicates normal, yellow indicates slight abnormality, and red indicates serious abnormality; for example: the line representing the temperature gradually rises and reaches 35°C at a certain moment, and at the same time the vibration amplitude also rises to 8 m / s 2 At this time, the state of the expansion joint in the interface becomes yellow, indicating a slight abnormality;

[0094] Further, after the state becomes yellow, the user clicks on the fault diagnosis unit, and the system details the fault information; for example: the fault types are temperature abnormality and vibration abnormality, the fault level is slight, and the fault location is determined to be at the 15th expansion joint of the GIL line through positioning technology;

[0095] At the same time, the occurrence situation, treatment measures, and treatment effects of historical faults of the same type are displayed; for example: a similar situation of temperature abnormality and vibration abnormality occurred half a year ago. At that time, the treatment measure was to dissipate heat from the surrounding environment and check and tighten some connection components of the expansion joint. After the treatment, the equipment returned to normal operation;

[0096] Meanwhile, based on the correlation analysis of the output of the perception model and historical fault data, a cause analysis is given. For example: Cause analysis: The recent increase in environmental temperature and the increase in load have led to an increase in the working pressure of the expansion joint, resulting in abnormal vibration.

[0097] Furthermore, the predictive maintenance unit formulates a personalized maintenance plan by combining the operation history and maintenance records of the expansion joint. For example: It is planned to conduct a comprehensive inspection and maintenance of the expansion joint after 30 hours. The main maintenance components are the connecting bolts and seals of the expansion joint; the system reminds the user through an interface pop-up window 30 hours in advance.

[0098] Furthermore, after receiving the reminder, the user conducts an inspection and maintenance of the expansion joint on time after 30 hours. Through actual measurement, it is found that the displacement of the expansion joint has started to change abnormally. The user tightens the connecting bolts and checks and replaces the seals according to the maintenance plan. After the maintenance is completed, the system monitors that data such as temperature, vibration, and displacement gradually return to normal, and the status of the expansion joint turns green again, and the fault is eliminated.

[0099] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

Claims

1. A telescopic joint dynamic perception system based on GIL multi-dimensional perception, characterized in that: The system includes: a data acquisition module, a multimodal fusion module, an adaptive perception module, an adaptive optimization module, a state evaluation module, and an application interface module; The data acquisition module is used to acquire multimodal data about the expansion joint. The multimodal data includes environmental data, vibration data, sound data, and displacement data; and preprocess the multimodal data. The multimodal fusion module performs fusion of different modal data based on the preprocessed multimodal data and outputs the integrated features after fusion. The adaptive perception module performs environmental state perception using reinforcement learning based on the integrated features after fusion and constructs a perception model. The adaptive optimization module, based on the adaptive perception module, uses online learning technology to combine historical data and real-time input to optimize the parameters of the perception model in real time. The state evaluation module performs state evaluation and prediction of the expansion joint based on the optimized perception data. The application interface module is used to provide real-time monitoring, fault diagnosis, and predictive maintenance to the user.

2. The dynamic perception system for expansion joints based on GIL multi-dimensional perception according to claim 1, characterized in that: The data acquisition module includes an environmental monitoring unit, a vibration monitoring unit, a sound monitoring unit, a displacement monitoring unit, and a data preprocessing unit; The environmental monitoring unit is used to acquire the temperature, humidity, and pressure data of the environment around the expansion joint in real time; the vibration monitoring unit is used to acquire the vibration signal of the expansion joint in real time; the sound monitoring unit is used to acquire the sound signal of the expansion joint in real time; the displacement monitoring unit is used to acquire the deformation and displacement data of the expansion joint in real time. The data preprocessing unit preprocesses the acquired multimodal data about the expansion joint; the preprocessing includes missing value processing, outlier detection and correction, and data standardization processing for the temperature, humidity, and pressure data of the environment around the expansion joint, time domain processing and frequency domain processing for the vibration signal of the expansion joint, noise reduction processing and feature extraction for the sound signal of the expansion joint, and smoothing processing and data difference processing for the deformation and displacement data of the expansion joint.

3. The dynamic perception system for expansion joints based on GIL multi-dimensional perception according to claim 1, characterized in that: The multimodal fusion module includes a multimodal deep learning model and an attention mechanism unit; The multimodal deep learning model includes a Transformer model subunit and a graph neural network model subunit; the Transformer model subunit generates sequence data features representing the integrated features of time series data; the graph neural network model subunit generates graph structure data features representing the integrated relationship between multimodal data. The attention mechanism unit receives the sequence data features of the Transformer model subunit and the graph structure data features of the graph neural network model subunit, and calculates the correlation scores between each modality data and other modality data. Each modality data includes the sequence data features processed by the Transformer model subunit and the graph structure data features processed by the graph neural network model subunit; The correlation scores are converted into attention weights through the Softmax function. The higher the attention weight, the greater the importance of the modality data in the current environment. Multiply the features of each modality data by the corresponding attention weights to obtain weighted features. Concatenate or sum the weighted features to generate the fused comprehensive features.

4. The dynamic perception system of an expansion joint based on GIL multi-dimensional perception according to claim 3, wherein: The Transformer model subunit receives the preprocessed multi-modal data, arranges it in chronological order, embeds the data of each modality into a high-dimensional vector space respectively, and uses the self-attention mechanism to calculate the attention weights between different time steps to capture the dependencies within the sequence. Through multiple attention heads, capture the relationships between different modality data and generate sequence data features representing the comprehensive features of time series data; The graph neural network model subunit receives the preprocessed multi-modal data, takes the data of each modality as a node, takes the relationships between different modality data as edges, and aggregates the information of nodes and edges through graph convolutional layers to generate graph structure data features representing the comprehensive relationships between multi-modal data.

5. The dynamic perception system of an expansion joint based on GIL multi-dimensional perception according to claim 1, wherein: The adaptive perception module includes a reinforcement learning unit and a perception model unit; The reinforcement learning unit includes: Receives the fused comprehensive features and uses the fused comprehensive features as the input state of the reinforcement learning agent. Takes the adjustable parameters of the perception model as the action space of the reinforcement learning agent. The adjustable parameters include the sampling frequency of the multi-modal data of the expansion joint, the weight coefficient of multi-modal data fusion, and the threshold of the perception model. Gives rewards according to the judgment accuracy of the perception model on the environmental state and its own state of the expansion joint; The reinforcement learning agent selects an action to execute from the policy network according to the current state. The policy network guides the reinforcement learning agent to make action selections by learning the mapping relationship between states and actions; Based on the action selection, combined with the feedback of the environment, obtain new states and reward values; Form an experience tuple with the state, action, reward, and new state, and store it in the experience replay buffer. Periodically randomly sample several experience tuples from the experience replay buffer to update the parameters of the policy network and the value network, and learn the optimal action selections under different states; Transmit the optimal action selections under different states to the perception model unit; The perception model unit includes: Based on the optimal action selection in different states learned by the reinforcement learning unit, adjust the adjustable parameters of the perception model to construct a perception model for evaluating and predicting the state of the expansion joint environment and its own state.

6. The dynamic perception system of an expansion joint based on GIL multi-dimensional perception according to claim 4, characterized in that: The rewards include giving positive rewards when the perception model accurately predicts the normal operation state, abnormal state or fault type of the expansion joint, and giving negative rewards when the perception model wrongly predicts the normal operation state, abnormal state or fault type of the expansion joint; the rewards are obtained by calculating the accuracy rate and recall rate of the prediction through comparison with the actual state.

7. The dynamic perception system of an expansion joint based on GIL multi-dimensional perception according to claim 1, characterized in that: The adaptive optimization module includes an online learning unit and a model performance evaluation unit; The online learning unit includes: Receiving the perception model constructed by the adaptive perception module; Receiving in real time the fused comprehensive features of the multi-modal fusion module, fusing them with historical data to form a dynamic data set; Adding the fused comprehensive features in the dynamic data set in the recent period to a sliding window to ensure that the perception model can quickly adapt to the latest environmental changes; Using an online learning algorithm to incrementally update the perception model. At the same time, introduce a forgetting factor to reduce the influence of old data on the perception model and avoid the perception model from becoming outdated; Detecting in real time whether there are data anomalies in the fused comprehensive features, and dynamically correcting the parameters of the perception model according to the anomaly detection results to avoid the drift of the perception model; The model performance evaluation unit includes: Based on the online learning unit, adjusting the adjustable parameters of the perception model in real time; Using a data set to verify the perception model, calculating evaluation metric values, and evaluating the generalization ability and accuracy of the model; Feeding back the evaluation metric values to the adaptive perception module to further adjust the adjustable parameters of the perception model; When the evaluation metric values no longer have significant improvement in consecutive multiple optimization cycles, stop the optimization; Outputting the perception model optimized by the adaptive optimization module for real-time evaluation of the environmental state and the state of the expansion joint.

8. The dynamic perception system of an expansion joint based on GIL multi-dimensional perception according to claim 1, characterized in that: The state evaluation module includes a state evaluation unit and a fault prediction unit; The state evaluation unit includes: Inputting the real-time collected and fused comprehensive features into the perception model; Based on the perception data of the perception model, classifying the state of the expansion joint and predicting the environmental state; The state of the expansion joint includes normal, slightly abnormal and severely abnormal; the prediction of the environmental state includes predicted values of temperature, humidity and pressure data; The state evaluation unit outputs the current state of the expansion joint and the prediction result of the environmental state; The fault prediction unit includes: Based on the state evaluation unit, classifying the fault type of the expansion joint, outputting the fault type, and predicting the time of fault occurrence; The fault prediction unit outputs the fault type and the fault prediction result.

9. A telescopic joint dynamic perception system based on GIL multi-dimensional perception according to claim 1, characterized in that: The application interface module includes a real-time monitoring unit, a fault diagnosis unit, and a predictive maintenance unit; The real-time monitoring unit uses a dynamic graph to display the multi-modal data of the telescopic joint, updates the telescopic joint data in real time, allows users to observe the change trend of the data in real time; displays the current state of the telescopic joint and differentiates it with different colors, including green for normal, yellow for slight abnormality, and red for serious abnormality; The fault diagnosis unit, when the telescopic joint is in an abnormal state, details the fault information; the fault information includes the fault type, fault level, and fault location, and displays the occurrence, handling measures, and handling effects of historical faults of the same type; Displays the time prediction of the fault occurrence and the cause analysis of the fault, and the cause analysis is obtained based on the correlation analysis of the output of the perception model and historical fault data; The predictive maintenance unit formulates a personalized predictive maintenance plan in combination with the operation history and maintenance records of the telescopic joint; the predictive maintenance plan includes the maintenance time node, maintenance equipment, and maintenance components; Sets a maintenance reminder, and according to the set maintenance reminder, before the maintenance time node, displays the predictive maintenance plan in the form of an interface pop-up window to remind the user to perform maintenance operations.

Images