Cable accessory installation process traceability system and method based on time sequence attention and digital twinning

Abstract

The invention relates to the technical field of cable installation, and particularly discloses a cable accessory installation technology traceability system and method based on time sequence attention and digital twinning, and the system comprises a multi-modal data collection module which captures the hand motion time sequence data of a cable installation worker in real time, generates local three-dimensional point cloud data of a cable accessory, and transmits the local three-dimensional point cloud data to a data processing module; obtaining defect data of each process; the time sequence process fingerprint extraction module performs feature extraction on hand action time sequence data through a time sequence attention network and generates a causal model of an action sequence and a process quality index; the digital twinning dynamic mapping module generates a defect tracing thermodynamic diagram based on the local three-dimensional point cloud data; the cross-process correlation analysis module identifies a cross-process analysis result; and the defect root cause tracing module carries out root cause tracing and defect evolution prediction and generates an optimal optimization scheme. According to the invention, the problem source can be accurately positioned.

Description

Technical Field

[0001] The present invention relates to the technical field of cable installation, and particularly to a cable accessory installation process traceability system and method based on temporal attention and digital twin. Background Art

[0002] In the power system, as a key carrier for power transmission, the installation quality of cable accessories directly relates to the stability and safety of power supply. With the rapid development of the power industry, the scale of cable laying is continuously expanding, and the complexity of cable accessory installation technology is also increasing day by day. Ensuring the high-quality execution of cable accessory installation technology is crucial for reducing power failures and ensuring the reliable operation of the power grid. The cable accessory installation process involves multiple processes and numerous operation details. Any problem in any link may lead to potential safety hazards. Traditional cable accessory installation process management mainly relies on manual experience and post-event inspection, making it difficult to comprehensively and real-time monitor and analyze the installation process. This makes it difficult to quickly and accurately trace the root cause of quality problems when they occur, and thus it is impossible to take effective improvement measures in a timely manner. With the continuous development of digital twin technology and artificial intelligence algorithms, new opportunities have been brought to the management of cable accessory installation technology, which can promote the power industry to develop towards the direction of intelligence and refinement, and has broad application prospects in ensuring the safe and stable operation of the power system.

[0003] However, there are many deficiencies in the existing cable accessory installation process management technology. Due to the lack of effective data processing and analysis means, the understanding of the factors affecting process quality is not deep enough, and the defect distribution and potential causes cannot be accurately determined, resulting in the inability to comprehensively grasp the potential problems in the process, and thus it is impossible to conduct root cause traceability and defect evolution prediction and generate the best optimization plan, making it difficult to effectively improve the quality of cable accessory installation technology.

[0004] Therefore, the present invention proposes a cable accessory installation process traceability system based on temporal attention and digital twin. Summary of the Invention

[0005] This invention provides a cable accessory installation process traceability system and method based on temporal attention and digital twins. It can capture real-time temporal data of cable installers' hand movements, generate local 3D point cloud data of cable accessories, and acquire defect data for each process, providing a comprehensive and real-time data foundation for subsequent analysis. By extracting features from the temporal data of hand movements through a temporal attention network and generating a causal model based on the correlation between construction actions and process quality, it can deeply explore the intrinsic relationship between actions and process quality, providing a strong basis for process evaluation. Based on the local 3D point cloud data, the causal model, and defect data, a defect tracing heatmap is generated, visually displaying the possible sources of defects and facilitating rapid location of problem areas. Correlation analysis is performed on the output of the causal model of the preceding process and the real-time defect data to identify cross-process analysis results, helping to discover potential influence relationships between different processes and comprehensively grasp the process. Root cause tracing is performed based on the defect tracing heatmap, cross-process analysis results, and defect data of each process.

[0006] The first aspect of this invention provides a cable accessory installation process traceability system based on temporal attention and digital twins, comprising:

[0007] The multimodal data acquisition module is used to capture real-time timing data of the hand movements of cable installers, generate local three-dimensional point cloud data of cable accessories, and acquire defect data of each process.

[0008] The temporal process fingerprint extraction module is used to extract features from the temporal data of hand movements through a temporal attention network and combine the correlation between construction actions and process quality to generate a causal model of action sequences and process quality indicators.

[0009] The digital twin dynamic mapping module is used to generate a defect source tracing heat map based on local 3D point cloud data;

[0010] The cross-process correlation analysis module is used to perform correlation analysis on the relationships between processes and identify the cross-process analysis results;

[0011] The defect root cause tracing module is used to trace the root causes based on the defect tracing heatmap, cross-process analysis results, and defect data of each process.

[0012] Preferably, the multimodal data acquisition module includes:

[0013] The hand motion data acquisition module is used to capture the timing data of the hand motions of cable installers in real time through a wearable operation recorder equipped with an inertial measurement unit and motion sensors.

[0014] The 3D point cloud data acquisition module is used to capture image data of the cable accessory installation location in real time through a high-precision binocular vision terminal equipped with a depth camera and an image processor, and generate local 3D point cloud data based on the real-time captured image data.

[0015] The defect data acquisition module is used to acquire defect data for each process based on dedicated sensor units and local 3D point cloud data of each process.

[0016] Preferably, the defect data acquisition module includes:

[0017] The first defect data acquisition submodule is used to detect insulation defects, temperature anomalies and assembly pressure deviations using a dedicated sensor unit equipped with a partial discharge sensor, an infrared thermal imager and a pressure sensor. It also associates the detected insulation defects, temperature anomalies and assembly pressure deviations with the corresponding process timestamps and spatial coordinates to obtain the first defect data for each process.

[0018] The second defect data acquisition submodule is used to identify the second defect data of each process by comparing the local three-dimensional point cloud data of each process.

[0019] Preferably, the timing process fingerprint extraction module includes:

[0020] The temporal feature extraction submodule is used to encode the hand movement temporal data and extract temporal features through a bidirectional long short-term memory network to obtain hand movement temporal features;

[0021] The attention mechanism submodule is used to calculate the attention weights of the hand action temporal features at different time steps, and to fuse the hand action temporal features based on the attention weights at different time steps to obtain process fingerprint features focused on key actions.

[0022] The causal model construction submodule is used to establish a causal relationship model between action sequences and process quality indicators based on process fingerprint features and corresponding process quality indicator data.

[0023] Preferably, the digital twin dynamic mapping module includes:

[0024] The geometric model construction submodule is used to construct the geometric model of cable accessories based on local 3D point cloud data;

[0025] The physical property mapping submodule is used to map process parameters to the cable accessory geometry model and define material properties and boundary conditions;

[0026] The dynamic simulation submodule is used to simulate the installation process and quality evolution under different process parameters based on the geometric model, material properties and boundary conditions of cable accessories, and generate a defect source heat map that reflects the risk of process defects at different locations.

[0027] Preferably, the cross-process correlation analysis module includes:

[0028] The process diagram construction submodule is used to decompose the cable accessory installation process into multiple processes and construct a directed graph that reflects the dependencies between processes.

[0029] The graph neural network submodule is used to perform convolution operations on the directed graph and learn the relationships between nodes to obtain the relationships between processes.

[0030] The early warning generation submodule is used to identify the impact path of the preceding process on subsequent defects based on the relationship between processes, and to serve as the result of cross-process analysis.

[0031] Preferably, the defect root cause tracing module includes:

[0032] The defect parameter parsing submodule is used to extract the risk density value of all defect locations from the defect source heat map, extract the transmission probability of each influence path from the cross-process analysis results, and obtain the physical parameters of each defect location from the defect data of each process.

[0033] The Defect Contribution Accumulation Submodule is used to calculate the posterior probability of each influence path in the cross-process analysis results, and to label the defect evolution contribution of the preceding process for each defect location based on the posterior probability of each influence path.

[0034] The root cause reasoning submodule is used to generate the root cause reasoning chain for each defect location based on the defect evolution contribution of all preceding processes at each defect location and the transmission probability of all influence paths of each defect location in the cross-process analysis results.

[0035] The preferred defect contribution accumulation submodule includes:

[0036] The path posterior probability calculation unit is used to calculate the posterior probability that a corresponding influence path causes all defect types involved in a single influence path, given that all defect types involved in a single influence path are known in the cross-process analysis results.

[0037] The same source dependency path analysis unit is used to filter out all the same source dependency path combinations from all the influence paths involved in the cross-process analysis results, and calculate the dependency coefficient of each same source dependency path combination.

[0038] The homogeneous posterior probability calculation unit is used to calculate the posterior probability of each homogeneous dependent path combination based on the premise that the corresponding influence path is the posterior probability of all defect types involved in the corresponding influence path and the dependency coefficient of each homogeneous dependent path combination, given that all defect types involved in a single influence path are known in the cross-process analysis results.

[0039] The contribution transmission factor definition unit is used to back-transmit the posterior probability of each homologous dependent path combination to the preceding process node and define all contribution transmission factors for each influencing path.

[0040] The transmission factor tracing and accumulation unit is used to trace and accumulate all contribution transmission factors of all influence paths to obtain the defect evolution contribution of the preceding process at each defect location.

[0041] Preferably, the posterior probability calculation unit includes:

[0042] The prior probability calculation subunit is used to calculate the prior probability of each influencing path in the cross-process analysis results;

[0043] The likelihood calculation subunit is used to calculate the likelihood of all defect types involved in the corresponding influence path under the given condition of the corresponding influence path;

[0044] The posterior probability calculation subunit is used to calculate the posterior probability that the corresponding influence path causes all defect types involved in the corresponding influence path, based on the prior probability of each influence path in the cross-process analysis results, the likelihood of all defect types involved in the corresponding influence path, and the prior probability of all defect types involved in the corresponding influence path, given that all defect types involved in the corresponding influence path are known.

[0045] Preferably, the system also includes:

[0046] The optimization scheme generation module is used to predict the evolution of defects and generate the best optimization scheme.

[0047] Preferably, the optimization scheme generation module includes:

[0048] The comprehensive risk calculation submodule is used to calculate the comprehensive risk coefficient of each defect location based on the risk density value, root cause inference chain, and physical parameters of each defect location.

[0049] The defect evolution prediction submodule is used to build the current defect evolution network based on the root cause tracing reasoning chain of all defect locations, and to perform defect evolution prediction based on the current defect evolution network and the comprehensive risk coefficient of each defect location, so as to obtain a complete defect evolution prediction network.

[0050] The optimization scheme generation submodule is used to generate multiple hypothetical optimization schemes based on the complete defect evolution prediction network, and simulate the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme. The hypothetical optimization scheme with the largest risk reduction rate among all hypothetical optimization schemes is regarded as the best optimization scheme.

[0051] The preferred optimization scheme generation submodule includes:

[0052] The defect optimization evolution prediction unit is used to input each hypothetical optimization scheme into the complete defect evolution prediction network and update the state parameters of all network nodes to generate the defect optimization evolution prediction network.

[0053] The pre- and post-optimization risk index analysis unit is used to analyze the pre-optimization risk index of the complete defect evolution prediction network, and simulate the defect development trend in the future period based on the defect optimization evolution prediction network to generate the post-optimization risk index of each hypothetical optimization scheme.

[0054] The risk reduction rate analysis unit is used to analyze the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme based on the risk index before optimization and the risk index after optimization for each hypothetical optimization scheme.

[0055] The optimal optimization scheme selection unit is used to select the optimal optimization scheme as the one with the largest risk reduction rate among all the hypothetical optimization schemes.

[0056] A second aspect of this invention provides a method for tracing the installation process of cable accessories based on temporal attention and digital twins, comprising the following steps:

[0057] The system acquires real-time captured time-series data of the hand movements of cable installers, generates local 3D point cloud data of cable accessories, and acquires defect data for each process.

[0058] By using a temporal attention network to extract features from the temporal data of hand movements and combining the correlation between construction actions and process quality, a causal model of action sequences and process quality indicators is generated.

[0059] Generate a defect source heat map based on local 3D point cloud data;

[0060] Perform correlation analysis on the relationships between processes to identify cross-process analysis results;

[0061] Root cause tracing is performed based on defect source heatmaps, cross-process analysis results, and defect data from each process.

[0062] Preferably, the method further includes: predicting the evolution of defects and generating the optimal optimization scheme.

[0063] The beneficial effects of this invention compared to existing technologies are as follows: It can capture real-time timing data of cable installation workers' hand movements and acquire defect data for each process, providing a comprehensive and real-time data foundation for subsequent analysis. By extracting features from the timing data of hand movements using a temporal attention network and generating a causal model based on the correlation between construction actions and process quality, it can deeply explore the intrinsic connection between actions and process quality, providing a strong basis for process evaluation. Based on the causal model and defect data, a defect source heatmap is generated, visually displaying the possible sources of defects and facilitating rapid location of problem areas. Correlation analysis is performed on the output of the causal model of previous processes and real-time defect data to identify cross-process analysis results, helping to discover potential influence relationships between different processes and comprehensively grasp the process. Based on the defect source heatmap, cross-process analysis results, and defect data for each process, root cause tracing and defect evolution prediction are performed to generate the best optimization solution, accurately locating the root cause of the problem and predicting the defect development trend, providing targeted optimization strategies. In summary, this system can combine multi-source data to achieve in-depth analysis and source tracing of cable accessory installation processes. By collecting and processing multimodal data during the installation process, using temporal attention networks to explore the relationship between actions and process quality, and leveraging digital twin technology to achieve visualized traceability of process defects, the quality of cable accessory installation process can be effectively improved, ensuring the stable and reliable operation of the cable system and helping to enhance the management level of cable accessory installation process.

[0064] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0065] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0067] Figure 1 This is a schematic diagram of a cable accessory installation process traceability system based on temporal attention and digital twins in an embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the multimodal data acquisition module in an embodiment of the present invention;

[0069] Figure 3 This is a schematic diagram of the timing process fingerprint extraction module in an embodiment of the present invention;

[0070] Figure 4 This is a schematic diagram of the digital twin dynamic mapping module in an embodiment of the present invention;

[0071] Figure 5 This is a schematic diagram of the cross-process correlation analysis module in an embodiment of the present invention;

[0072] Figure 6 This is a schematic diagram of the defect root cause tracing module in an embodiment of the present invention;

[0073] Figure 7 This is a schematic diagram of the optimization scheme generation module in an embodiment of the present invention;

[0074] Figure 8 This is a flowchart of the cable accessory installation process traceability method based on temporal attention and digital twin in an embodiment of the present invention. Detailed Implementation

[0075] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0076] like Figure 1 As shown, this embodiment of the invention provides a cable accessory installation process traceability system based on temporal attention and digital twins, comprising:

[0077] The multimodal data acquisition module is used to capture real-time timing data of the hand movements of cable installers, generate local three-dimensional point cloud data of cable accessories, and acquire defect data of each process.

[0078] The temporal process fingerprint extraction module is used to extract features from the temporal data of hand movements through a temporal attention network and combine the correlation between construction actions and process quality to generate a causal model of action sequences and process quality indicators.

[0079] The digital twin dynamic mapping module is used to generate a defect source tracing heat map based on local 3D point cloud data;

[0080] The cross-process correlation analysis module is used to perform correlation analysis between the causal model output of the preceding process and the defect data of each process collected in real time, and to identify the cross-process analysis results.

[0081] The defect root cause tracing module is used to trace the root causes based on the defect tracing heatmap, cross-process analysis results, and defect data of each process.

[0082] In this embodiment, the hand movement timing data is a sequence of data showing the changes in the hand movements of cable installers over time, recording the sequence, interval, and amplitude of the movements.

[0083] In this embodiment, cable accessories are components used in power system cable lines to connect cables with other electrical equipment or two sections of cables to ensure reliable power transmission, such as cable terminations and intermediate joints.

[0084] In this embodiment, the local three-dimensional point cloud data of the cable accessory is a set of points with spatial coordinates generated after processing the image data of the cable accessory installation location, which is used to describe the geometry and surface features of the installation location.

[0085] In this embodiment, a temporal attention network is used to process time-series data. Specifically, it's used for extracting features from hand movement temporal data, allowing the model to focus on key information at different time steps and automatically assign attention weights. It is specifically designed for processing time-series data like the hand movements of cable installers. It mainly consists of components with special functions. For example, a bidirectional long short-term memory network first encodes the worker's hand movement temporal data, essentially organizing this movement information into a format easy for the network to understand. This allows for the extraction of temporal features of the hand movements, such as the sequence of movements, duration, and pause intervals, resulting in hand movement temporal features. Then, an attention mechanism focuses on key movements. For instance, in the critical step of cable connection, the time step in which the worker's hand precisely connects to the cable will have a higher attention weight because it significantly impacts the quality of the connection process. Finally, based on these attention weights, the hand movement temporal features are weighted and fused.

[0086] In this embodiment, the relationship between construction actions and process quality refers to the inherent connection between the actions of cable installers and the process quality of cable accessory installation. For example, the action of stripping the cable insulation layer affects the insulation performance.

[0087] In this embodiment, the causal model between action sequences and process quality indicators is established based on the process fingerprint features extracted by the temporal attention network and the process quality indicator data, and is used to describe how action sequences lead to changes in process quality indicators.

[0088] In this embodiment, a defect tracing heatmap is used to visually represent the risk level of process defects at different locations during the installation of cable accessories.

[0089] In this embodiment, the cross-process analysis results are obtained by associating the causal model output of the preceding process with the real-time defect data of each process, which shows the potential impact path of the preceding process on the defects of the subsequent process.

[0090] In this embodiment, the optimal solution is the one with the highest risk reduction rate selected from multiple hypothetical optimization solutions.

[0091] like Figure 2As shown, in order to collect real-time data on the hand movements of cable installers, local 3D point cloud data of cable accessories, and defect data of each process using a wearable operation recorder, a high-precision binocular vision terminal, and a dedicated sensor unit, a multimodal data acquisition module is further proposed, including:

[0092] The hand motion data acquisition module is used to capture the timing data of the hand motions of cable installers in real time through a wearable operation recorder equipped with an inertial measurement unit and motion sensors.

[0093] The 3D point cloud data acquisition module is used to capture image data of the cable accessory installation location in real time through a high-precision binocular vision terminal equipped with a depth camera and an image processor, and generate local 3D point cloud data based on the real-time captured image data.

[0094] The defect data acquisition module is used to acquire defect data for each process based on dedicated sensor units and local 3D point cloud data of each process.

[0095] In this embodiment, the inertial measurement unit (IMU) is a device for measuring the motion state of an object. It can detect the acceleration and angular velocity of a wearable operation recorder, thereby obtaining information such as the acceleration and rotation of the cable installer's hand movements, and assisting in accurately capturing the timing data of hand movements.

[0096] In this embodiment, the motion sensor is used to sense changes in the hand movements of cable installers and convert the motion information into processable data forms such as electrical signals.

[0097] In this embodiment, a wearable operation recorder can capture the timing data of a worker's hand movements in real time after the worker wears it.

[0098] In this embodiment, the depth camera is capable of acquiring depth information of objects in the scene and obtaining an image containing distance information by photographing the installation location of the cable accessory.

[0099] In this embodiment, the image processor processes the image data of the cable accessory installation location acquired by the depth camera, such as performing noise reduction and feature extraction operations.

[0100] In this embodiment, the high-precision binocular vision terminal integrates a depth camera and an image processor, which can capture and process image data of the cable accessory installation location in real time.

[0101] In this embodiment, the image data of the cable accessory installation location is captured in real time by a high-precision binocular vision terminal, and includes visual information of the location when the cable accessory is installed, such as shape, color, and texture.

[0102] In this embodiment, a dedicated sensor unit is used to detect defect data such as insulation defects, temperature anomalies, and assembly pressure deviations in each process of cable accessory installation.

[0103] To detect insulation, temperature, and pressure-related defects using dedicated sensor units configured with specific sensors and associate them with timestamps and coordinates, and to identify defects through point cloud comparison, thereby comprehensively acquiring defect data from each process, a defect data acquisition module is further proposed, including:

[0104] The first defect data acquisition submodule is used to detect insulation defects, temperature anomalies and assembly pressure deviations using a dedicated sensor unit equipped with a partial discharge sensor, an infrared thermal imager and a pressure sensor. It also associates the detected insulation defects, temperature anomalies and assembly pressure deviations with the corresponding process timestamps and spatial coordinates to obtain the first defect data for each process.

[0105] The second defect data acquisition submodule is used to identify the second defect data of each process by comparing the local three-dimensional point cloud data of each process.

[0106] The defect data for each process includes the first defect data and the second defect data for each process.

[0107] In this embodiment, the partial discharge sensor is a device capable of detecting weak signals generated by partial discharge caused by air gaps, impurities, etc. in the insulation structure inside the cable accessory.

[0108] In this embodiment, the infrared thermal imager is an instrument that converts infrared radiation emitted by cable accessories into thermal images to present the surface temperature distribution.

[0109] In this embodiment, the pressure sensor is a device that can sense the pressure during the assembly of cable accessories and convert the pressure signal into an electrical signal that is easy to process.

[0110] In this embodiment, during each installation process, a partial discharge sensor is used to find insulation defects, an infrared thermal imager is used to check for temperature anomalies, and a pressure sensor is used to measure assembly pressure deviations. At the same time, the time of the process in which the problem was found and the location of the defect in the cable accessory are recorded. The data are then integrated to obtain the first defect data from these three aspects for each process.

[0111] In this embodiment, by comparing the local three-dimensional point cloud data of each process, the second defect data of each process is identified: the local three-dimensional point cloud data of cable accessories generated in each process is compared with the standard or expected point cloud data, and based on the differences in shape, position, size, etc. of the point cloud, defects such as component position offset and size discrepancy in the process are found.

[0112] like Figure 3As shown, in order to extract the temporal features of hand movements through a bidirectional long short-term memory network, a weighted fusion focusing on key movements is utilized using an attention mechanism to establish a causal relationship model between movement sequences and process quality indicators. Furthermore, a temporal process fingerprint extraction module is proposed, including:

[0113] The temporal feature extraction submodule is used to encode the hand movement temporal data and extract temporal features through a bidirectional long short-term memory network to obtain hand movement temporal features;

[0114] The attention mechanism submodule is used to calculate the attention weights of the hand action temporal features at different time steps, and to fuse the hand action temporal features based on the attention weights at different time steps to obtain process fingerprint features focused on key actions.

[0115] The causal model construction submodule is used to establish a causal relationship model between action sequences and process quality indicators based on process fingerprint features and corresponding process quality indicator data.

[0116] In this embodiment, a bidirectional long short-term memory network processes temporal data recording the hand movements of cable installers over time. The network extracts features related to the movements in the time dimension from the data, such as the sequence of movements and duration. These extracted features are the temporal features of the hand movements. The specific structure of the bidirectional long short-term memory network is as follows:

[0117] Network layers: A 3-layer stacked Bi-LSTM structure is adopted, with 1 input layer and 2 hidden layers. The number of neurons in each layer is 256, 128 and 64 respectively. The input layer receives hand action time-series data (sampling frequency 10Hz, single time-series data length is 512 time steps, feature dimensions include acceleration, angular velocity, and attitude angle, a total of 9 dimensions). After linear transformation, the 9-dimensional features are mapped to 256-dimensional feature vectors and then input into the hidden layer.

[0118] Activation functions: The hidden layer neurons use the tanh activation function, and the output layer uses the sigmoid activation function to output the feature weight coefficients at each time step.

[0119] Forget gate / input gate / output gate parameters: The forget gate weights are initialized using a Xavier normal distribution, with the initial value of the bias term set to 0.1; the input and output gate weights are initialized using a He normal distribution, with the initial value of the bias term set to 0; the dropout probability of each layer is set to 0.2 to prevent the model from overfitting.

[0120] Training parameters: The Adam optimizer was used, the initial learning rate was set to 0.001, and it was decayed to 0.8 times the previous stage every 50 epochs; the mean squared error (MSE) loss function was used, the number of training iterations was 300 epochs, and the batch size was set to 32.

[0121] In this embodiment, the attention weights of hand action timing features at different time steps are calculated, and the hand action timing features are weighted and fused based on the attention weights at different time steps to obtain process fingerprint features focusing on key actions: the hand action timing features at each time step are weighted and added according to their corresponding weights, and the fused result highlights the key actions that have an important impact on the process.

[0122] Suppose that when a worker installs cable accessories, their hand actions are performed in the following sequence: picking up insulating tape (Action A), wrapping the tape (Action B), and inspecting the connection points (Action C). A bidirectional long short-term memory network has extracted the temporal characteristics of these hand actions. Different time steps have different degrees of impact on the process. For example, inspecting the connection points (Action C) is crucial to ensuring the quality of the cable connection, and its time step has a significant impact on the process; while picking up the insulating tape (Action A) has a relatively smaller impact.

[0123] Suppose we have a set of temporal features of hand movements, where each time step can be viewed as a vector. Taking the scaled dot product attention algorithm as an example, the temporal features of hand movements are first transformed linearly to obtain a query vector (Q), a key vector (K), and a value vector (V). When calculating the attention weights, the query vector Q and the key vector K are first multiplied to obtain a score matrix. To make the calculation more stable, the score matrix is ​​divided by a scaling factor (usually the square root of the key vector dimension) and then normalized using a softmax function to obtain the attention weight for each time step. Finally, the attention weight for each time step is multiplied by the corresponding value vector V, and then accumulated to achieve a weighted fusion of the temporal features of hand movements, resulting in a process fingerprint feature focused on key actions.

[0124] In this embodiment, a causal relationship model between action sequences and process quality indicators is established based on process fingerprint features and corresponding process quality index data. Process fingerprint features reflect key action characteristics, while corresponding process quality index data quantifies process quality performance such as connection strength and insulation performance. Assuming that in a cable installation scenario, process fingerprint features and corresponding process quality index data have already been acquired, the process fingerprint features reflect the key action characteristics of workers installing cables, such as the duration and force changes of specific actions. The process quality index data quantifies performance such as connection strength and insulation performance; for example, connection strength is measured by the tensile force that the joint can withstand, and insulation performance is represented by the insulation resistance value. A common way to construct this model is using a linear regression algorithm. Besides linear regression, decision tree algorithms can also be used to describe how action sequences cause changes in process quality indicators.

[0125] like Figure 4 As shown, in order to construct a geometric model of cable accessories based on local 3D point cloud data, map process parameters and define attributes and conditions, and generate a defect source heat map based on multi-factor simulation of the installation process and quality evolution, a digital twin dynamic mapping module is further proposed, including:

[0126] The geometric model construction submodule is used to construct the geometric model of cable accessories based on local 3D point cloud data;

[0127] The physical property mapping submodule is used to map process parameters to the cable accessory geometry model and define material properties and boundary conditions;

[0128] The dynamic simulation submodule is used to simulate the installation process and quality evolution under different process parameters based on the geometric model, material properties and boundary conditions of cable accessories, and generate a defect source heat map that reflects the risk of process defects at different locations.

[0129] In this embodiment, the cable accessory geometric model is a model that abstracts and digitally represents the geometric features of the cable accessory, such as its shape, size, and spatial relationship between its components.

[0130] In this embodiment, process parameters refer to various parameters that affect the installation quality and process during the installation of cable accessories. These include, for example, the temperature, humidity, operating force, and operating time during installation, as well as the specifications of each component of the cable accessories.

[0131] In this embodiment, process parameters are mapped to the geometric model of cable accessories and material properties and boundary conditions are defined: various process parameters are associated with the constructed geometric model of cable accessories, and the material properties (such as conductivity, insulation, flexibility, etc.) of each part of the model and the boundary constraints of the model in the simulation environment (such as the connection relationship with surrounding objects, force boundaries, etc.) are clarified, so that the model is closer to the actual installation scenario.

[0132] In this embodiment, material properties and boundary conditions are defined as follows: material properties describe the physical and chemical characteristics of the materials used in the cable accessories, such as the conductivity of metallic materials and the insulation strength of insulating materials; boundary conditions define the external constraints of the model during the simulation process, such as installation space constraints and connection methods between components.

[0133] In this embodiment, based on the cable accessory's geometric model, material properties, and boundary conditions, the installation process and quality evolution under different process parameters are simulated, generating a defect tracing heatmap reflecting the risk of process defects at different locations. Using the constructed cable accessory geometric model as a foundation, combined with defined material properties and boundary conditions, and based on the causal relationship between action sequences and process quality indicators, while also referencing real-time collected defect data for each process step, the installation process and quality changes of the cable accessory are simulated under different process parameter settings. The simulation results are presented in the form of a heatmap, using different colors or shades to represent the degree of risk of process defects occurring at different locations of the cable accessory.

[0134] like Figure 5 As shown, in order to decompose the cable accessory installation process into steps and construct a directed graph, a graph neural network is used to learn the node relationships and identify the influence path of the preceding steps on subsequent defects as the result of cross-step analysis. Furthermore, a cross-step correlation analysis module is proposed, including:

[0135] The process diagram construction submodule is used to decompose the cable accessory installation process into multiple processes and construct a directed graph that reflects the dependencies between processes. Nodes represent processes and associated process parameters and quality indicators, while edges represent the dependencies between processes.

[0136] The graph neural network submodule is used to perform convolution operations on the directed graph and learn the relationships between nodes to obtain the relationships between processes.

[0137] The early warning generation submodule is used to identify the impact path of the preceding process on subsequent defects based on the relationship between processes, and to serve as the result of cross-process analysis.

[0138] In this embodiment, a directed graph reflecting the dependencies between processes is used: each process in the cable accessory installation process is considered as a node, and the order, constraints, and other dependencies between processes are connected by directed edges to form a directed graph. For example, the conductor connection process can only be carried out after the cable stripping process is completed, so a directed edge is drawn from the cable stripping process node to the conductor connection process node, thus visually demonstrating the logical order and dependencies between processes.

[0139] In this embodiment, a convolution operation is performed on the directed graph to learn the relationships between nodes, thereby obtaining the relationships between processes: the convolution operation in graph neural networks is used to process this directed graph reflecting the dependencies between processes. Through convolution operations, the model automatically learns the mutual influence and interaction information between each node (process). For example, suppose the cable accessory installation process constitutes a directed graph, where nodes represent different processes, such as cable stripping, conductor connection, insulation treatment, and sealing treatment, and directed edges represent the sequential dependencies between processes, such as conductor connection only after cable stripping is completed. When performing a convolution operation on this directed graph, each process node carries its own information. For example, the cable stripping process node carries process parameter information such as stripping length and stripping speed, while the conductor connection process node carries information such as connection method and connection strength.

[0140] Taking the conductor connection process node as an example, convolution operations collect information from its preceding process node (cable stripping) and itself. Using a specific convolution kernel (a pre-defined calculation rule), the parameters of the cable stripping process (such as the impact of stripping length on the conductor connection length adaptability) are weighted and calculated with the parameters of the conductor connection process itself. For example, the insulation treatment process node integrates information from the conductor connection process node (such as the impact of connection tightness on insulation treatment effectiveness) as well as its own insulation material type, coating thickness, and other information. Through multiple convolution operations, the model can automatically learn the mutual influence between each process node. For instance, it learns how an inappropriate cable stripping length affects the stability of the conductor connection, which in turn affects the reliability of the insulation treatment, ultimately deriving the specific relationships between the various processes in cable accessory installation.

[0141] In this embodiment, the potential impact paths of preceding processes on subsequent defects are identified based on the inter-process relationships as the results of cross-process analysis: based on the inter-process relationships obtained above, it is analyzed which preceding processes may cause defects in subsequent processes, and these potential impact paths are identified. For example, improper operation of the preceding cable insulation process may, through a series of intermediate steps, ultimately affect the subsequent cable sealing process, resulting in a defect of incomplete sealing.

[0142] In this embodiment, the interrelationships between processes refer to the connections between various processes during cable accessory installation, including sequential relationships, process parameter transfer relationships, and resource sharing relationships. For example, the operational precision of a certain process can affect the installation difficulty of the next process.

[0143] In this embodiment, the potential impact path of preceding processes on subsequent defects represents a series of related links and steps in the cable accessory installation process that may lead to defects in subsequent processes due to factors such as the operation and process parameters of preceding processes. For example, the impact path could be from insufficient pressure in the cable conductor crimping process to the discovery of abnormal resistance in the subsequent electrical performance testing process.

[0144] like Figure 6 As shown, in order to extract relevant parameters from multi-source data, calculate the posterior probability of the impact path to label the defect evolution contribution, and generate a root cause tracing inference chain, a defect root cause tracing module is further proposed, including:

[0145] The defect parameter parsing submodule is used to extract the risk density value of all defect locations from the defect source heat map, extract the transmission probability of each influence path from the cross-process analysis results, and obtain the physical parameters of each defect location from the defect data of each process.

[0146] The Defect Contribution Accumulation Submodule is used to calculate the posterior probability of each influence path in the cross-process analysis results, and to label the defect evolution contribution of the preceding process for each defect location based on the posterior probability of each influence path.

[0147] The root cause reasoning submodule is used to generate the root cause reasoning chain for each defect location based on the defect evolution contribution of all preceding processes at each defect location and the transmission probability of all influence paths of each defect location in the cross-process analysis results.

[0148] In this embodiment, risk density values ​​for all defect locations are extracted from the defect tracing heatmap. Risk density values ​​are quantitative indicators of the risk level at each defect location. Using specific algorithms or tools, the risk density values ​​corresponding to each defect location are read and extracted from the heatmap to measure the likelihood of a process defect occurring at that location.

[0149] Suppose we have a heatmap for tracing defects in cable accessories, using different colors to represent the risk level of process defects at different locations on the cable accessories. Each pixel has a corresponding color value (in the RGB color space, each pixel is represented by values ​​from three channels: red, green, and blue). We can set a rule to convert color values ​​into risk density values. Assume the red channel value is between 0 and 255, and that a higher red channel value corresponds to a greater risk density. For a specific defect location, find the pixel in the image corresponding to that location. For example, this pixel might have a red channel value of 200. We can divide the red channel value by 255 to obtain a value between 0 and 1; this value will be used as the risk density value for that defect location.

[0150] In this embodiment, the transmission probability of each influence path is extracted from the cross-process analysis results: the cross-process analysis results contain the potential influence paths of the preceding process on subsequent defects. The transmission probability corresponding to each influence path is extracted from the data records of the cross-process analysis results. These probabilities reflect the likelihood of different influence paths causing defects to occur.

[0151] In this embodiment, physical parameters of each defect location are obtained from the defect data of each process, such as temperature, pressure, and resistance at the defect location.

[0152] In this embodiment, the posterior probability of each influence path is: given that all defect types involved in a single influence path are known in the cross-process analysis results, the probability that the influence path leads to these defect types.

[0153] In this embodiment, the defect location refers to the specific part of the cable accessory where the manufacturing defect occurs.

[0154] In this embodiment, the defect evolution contribution of the preceding process at each defect location is a quantitative assessment of the role of the preceding process at each defect location in the defect formation and development process, based on the posterior probability of each influence path.

[0155] In this embodiment, based on the defect evolution contribution of all preceding processes at each defect location and the transmission probability of all influence paths at each defect location in the cross-process analysis results, a root cause tracing inference chain is generated for each defect location: the defect evolution contribution of the preceding processes at each defect location and the transmission probability of the related influence paths are combined, and a series of reasoning steps from the process that may initially cause the defect to the final defect location are sorted out in logical order to form a root cause tracing inference chain.

[0156] In the cross-process analysis results, several influence paths related to the defect location and their propagation probabilities were identified. For example, path one starts from the cable stripping process (process A), passes through the conductor cleaning process (process B), and finally leads to the defect in the cable connection process (process C), with a propagation probability of 0.6. Path two leads directly from the insulation material preparation process (process D) to the cable connection process (process C), resulting in a defect with a propagation probability of 0.4. Simultaneously, the defect evolution contribution of each preceding process to the defect location was calculated. Assuming process A's contribution is 0.3, process B's contribution is 0.2, and process D's contribution is 0.5, a root cause tracing inference chain for the defect location (deteriorated insulation performance at the cable connection) is generated. First, considering path one, process A (cable stripping), due to its 0.3 defect evolution contribution, may be the starting point for subsequent problems due to improper operation, affecting process B (conductor cleaning) with a propagation probability of 0.6. Process B then propagates the effect to process C (cable connection) with a corresponding influence and a propagation probability of 0.6. In path two, process D (insulation material preparation) directly affects process C with a contribution of 0.5 and a transmission probability of 0.4. Organizing this information in logical order forms a root cause tracing reasoning chain.

[0157] To calculate the posterior probability of the influence path, we analyze the combination of co-originating dependent paths, calculate the co-originating posterior probability, define the contribution transmission factor and accumulate it tracing its origins to obtain the defect evolution contribution of the preceding process at each defect location. Furthermore, we propose a defect contribution accumulation submodule, including:

[0158] The path posterior probability calculation unit is used to calculate the posterior probability that a corresponding influence path causes all defect types involved in a single influence path, given that all defect types involved in a single influence path are known in the cross-process analysis results.

[0159] The same source dependency path analysis unit is used to filter out all the same source dependency path combinations that share a common starting node from all the influence paths involved in the cross-process analysis results, and calculate the dependency coefficient of each same source dependency path combination based on the number of shared nodes of all influence paths contained in each same source dependency path combination and the total number of nodes.

[0160] The homogeneous posterior probability calculation unit is used to calculate the posterior probability of each homogeneous dependent path combination based on the premise that the corresponding influence path is the posterior probability of all defect types involved in the corresponding influence path and the dependency coefficient of each homogeneous dependent path combination, given that all defect types involved in a single influence path are known in the cross-process analysis results.

[0161] The contribution transmission factor definition unit is used to convert all the influence paths in the cross-process analysis results into a multi-level directed weighted graph. Based on the posterior probability of each homologous dependent path combination and the length of each influence path, the defect process node in the multi-level directed weighted graph is introduced. At the same time, the path decay factor, time decay factor, and spatial decay factor are introduced to transmit back to the preceding process node and define all contribution transmission factors for each influence path.

[0162] The transmission factor tracing and accumulation unit is used to trace and accumulate all contribution transmission factors of all influence paths to obtain the defect evolution contribution of the preceding process at each defect location.

[0163] In this embodiment, all combinations of dependent paths sharing a common starting node are selected from all the influence paths involved in the cross-process analysis results: from the numerous influence paths obtained from the cross-process analysis, paths with the same starting point are identified and grouped together. For example, if path A and path B both start from process A, then they constitute such a combination.

[0164] In this embodiment, the dependency coefficient of each same-source dependency path combination is calculated based on the number of shared nodes and the total number of nodes of all affected paths contained in each same-source dependency path combination: for each same-source dependency path combination, the dependency coefficient is obtained by dividing the number of shared nodes by the total number of nodes.

[0165] In this embodiment, based on the premise that all defect types involved in a single influence path are known in the cross-process analysis results, the posterior probability of each co-source dependent path combination is calculated as the posterior probability of all defect types involved in the corresponding influence path and the dependency coefficient of each co-source dependent path combination: that is, the sum of the product of the posterior probabilities of all influence paths in the co-source dependent path combination and the product of the product and the dependency coefficient is taken as the posterior probability of the corresponding co-source dependent path combination.

[0166] In this embodiment, all influence paths in the cross-process analysis results are converted into a multi-level directed weighted graph: the influence paths in the cross-process analysis are represented by directed edges, which connect process or defect nodes, and the weight of the edge represents the path attribute, thus constructing a multi-level graph structure.

[0167] In this embodiment, based on the posterior probability of each source-dependent path combination and the length of each influencing path, the defective process node in the multi-level directed weighted graph is introduced. At the same time, path decay factor, time decay factor, and spatial decay factor are introduced and propagated backward to the preceding process node, and all contribution transmission factors of each influencing path are defined: starting from the defective process node in the graph, based on the posterior probability of the source-dependent path combination and the length of the influencing path, considering path, time, and spatial decay factors, the contribution transmission factor of each influencing path is determined by deriving from the preceding process node.

[0168] Suppose we are analyzing a complex circuit board assembly process, which can be represented by a multi-level directed weighted graph. There is a defect process node, representing the process where a short circuit is finally detected in a certain area of ​​the circuit board. We have two path combinations with the same source dependency: path combination one contains paths P1 and P2, and path combination two contains path P3. Each path has a length; for example, P1 has a length of 3 (representing 3 intermediate processes from the starting process to the defect process), P2 has a length of 4, and P3 has a length of 5. First, we determine the posterior probability of each path combination with the same source dependency. Assume the posterior probability of path combination one is 0.6, and the posterior probability of path combination two is 0.4. Next, we introduce attenuation factors. Assume the path attenuation factor is set to decrease in influence by 0.1 for each process; the time attenuation factor is set to decrease in influence by 0.05 for each additional day since the defect occurred (assuming days are the unit of time); and the spatial attenuation factor is set to decrease in influence by 0.02 for each centimeter further from the defect location (assuming centimeters are the unit of space). We then work backward from the short circuit defect process node. For path P1, with a length of 3, passing through 3 intermediate processes, path decay reduces its influence to the initial (1-0.1)^3 = 0.729. Assuming 5 days elapse from the starting process to the defective process, time decay reduces the influence to 0.729 × (1-0.05 × 5) = 0.729 × 0.75 = 0.54675. If the starting process is 10 cm from the defect location, spatial decay reduces the influence to 0.54675 × (1-0.02 × 10) = 0.54675 × 0.8 = 0.4374. Finally, combining the posterior probability of 0.6 for path combination one, the contribution transmission factor of P1 is 0.4374 × 0.6 = 0.26244. The same method is used to calculate the contribution transmission factors for P2 and P3, respectively.

[0169] In this embodiment, the contribution transmission factors of all influence paths are traced and accumulated to obtain the defect evolution contribution of the preceding process at each defect location: the contribution transmission factors of the influence paths related to each defect location are accumulated in order from the defect location to the preceding process to obtain the degree of contribution of the preceding process to the defect formation at that defect location.

[0170] Suppose that a poor contact defect is found at the cable termination joint (the defect location) during cable accessory installation. Analysis identifies three paths related to this defect location: path one begins with the conductor cleaning process, path two begins with the crimping process, and path three begins with the sealing process.

[0171] The contribution transmission factor for each influence path has been obtained through the preceding calculations. We assume the contribution transmission factor for the conductor cleaning process is 0.2, for the crimping process it is 0.3, and for the sealing process it is 0.1.

[0172] The contribution of the conductor cleaning process to the poor contact at the terminal joint is accumulated sequentially from the defect location (poor contact at the terminal joint) to the preceding process. Therefore, the contribution of the conductor cleaning process to the formation of this defect is 0.2; the contribution of the crimping process is 0.3; and the contribution of the sealing process is 0.1.

[0173] The sum of the contribution transmission factors of these three influence paths is 0.2 + 0.3 + 0.1 = 0.6, which is the total contribution of all preceding processes to the formation of this defect at the terminal joint.

[0174] To calculate the prior probability and likelihood of an impact path based on a historical process database, and further derive the posterior probability of the impact path leading to a defect type given a known defect type, a posterior probability calculation unit is proposed, comprising:

[0175] The prior probability calculation subunit is used to calculate the occurrence probability of each historical high-frequency influence path in the cross-process analysis results based on the historical process database as the prior probability of each historical high-frequency influence path, and to perform smooth correction based on the historical occurrence count of each historical low-frequency influence path in the cross-process analysis results to obtain the prior probability of each historical low-frequency influence path in the cross-process analysis results.

[0176] The likelihood calculation subunit is used to calculate the likelihood of all defect types involved in the corresponding influence path occurring under the given condition of the corresponding influence path, based on the number of times each influence path in the cross-process analysis results co-occurs with all defect types involved in the corresponding influence path in the historical process database and the number of times the corresponding influence path occurs alone in the historical process database.

[0177] The posterior probability calculation subunit is used to calculate the posterior probability that the corresponding influence path causes all defect types involved in the corresponding influence path, based on the prior probability of each influence path in the cross-process analysis results, the likelihood of all defect types involved in the corresponding influence path, and the prior probability of all defect types involved in the corresponding influence path, given that all defect types involved in the corresponding influence path are known.

[0178] In this embodiment, the historical process database stores a collection of data related to the installation process of cable accessories in the past. It includes detailed records of the procedures, defect types, impact paths, and corresponding process parameters in each installation process, providing historical reference for current process analysis.

[0179] In this embodiment, historical high-frequency impact paths refer to the impact paths of frequently occurring preceding processes on subsequent defects, recorded in the historical process database. These paths repeatedly lead to similar defects during multiple cable accessory installations, exhibiting a high frequency of occurrence and reflecting common process problem associations.

[0180] In this embodiment, the probability of occurrence of each historical high-frequency influence path in the cross-process analysis results is calculated based on the historical process database: all records involving cross-process analysis results are extracted from the historical process database, the number of occurrences of each historical high-frequency influence path is counted, and then divided by the total number of records to obtain the probability of occurrence of each historical high-frequency influence path in the historical data.

[0181] In this embodiment, the prior probability of each historical low-frequency impact path in the cross-process analysis results is obtained by smoothing and correcting the historical occurrence count of each path. Since historical low-frequency impact paths occur infrequently, directly calculating the probability by counting the occurrences may be inaccurate. Therefore, the Laplace smoothing correction algorithm is used, combined with the historical occurrence count and other relevant factors (such as overall process conditions, similar impact path probabilities, etc.), to adjust the path. The correction process is expressed as: P(path)_correction = (historical occurrence count + 1) / (total count + total number of paths), thus obtaining the prior probability of each historical low-frequency impact path.

[0182] In this embodiment, the number of times each influence path in the cross-process analysis results co-occurs with all defect types involved in the corresponding influence path in the historical process database is calculated: in the historical process database, each influence path obtained from the cross-process analysis is searched, and the number of times the influence path co-occurs with all defect types involved in it is counted. This number reflects the co-occurrence frequency of the influence path and related defect types in the historical installation process.

[0183] In this embodiment, the number of times each influence path appears individually in the historical process database is counted. In the historical process database, the number of times each influence path appears individually in the cross-process analysis results is counted. That is, without considering the related defect types, only the frequency of the influence path itself in the records is calculated.

[0184] In this embodiment, the likelihood of all defect types involved in the corresponding influence path occurring under the given influence path is calculated based on the number of times each influence path co-occurs with all defect types involved in the corresponding influence path in the historical process database and the number of times the corresponding influence path occurs alone in the historical process database: the number of times each influence path co-occurs with the corresponding defect type is divided by the number of times the influence path occurs alone. The resulting value is the likelihood of the corresponding defect type occurring given this influence path. It represents the probability of the relevant defect type occurring when the influence path is known to have occurred.

[0185] In this embodiment, based on the prior probability of each influence path in the cross-process analysis results, the likelihood of all defect types involved in the corresponding influence path, and the prior probability of all defect types involved in the corresponding influence path, the posterior probability that the corresponding influence path leads to all defect types involved in the corresponding influence path is calculated under the premise that all defect types involved in the corresponding influence path are known: using Bayes' theorem, the prior probability of the influence path, the likelihood of the corresponding defect type, and the prior probability of the defect type are substituted into the calculation to obtain the posterior probability that the influence path leads to these defect types under the premise that the corresponding defect types are known to occur, that is, the posterior probability of the path = (prior probability of the influence path × likelihood of the corresponding defect type) / prior probability of the defect type.

[0186] In some preferred embodiments, the system further includes:

[0187] The optimization scheme generation module is used to predict the evolution of defects and generate the best optimization scheme.

[0188] like Figure 7 As shown, in order to calculate the comprehensive risk coefficient, predict defect evolution to obtain the complete network, and generate the optimal optimization scheme, an optimization scheme generation module is further proposed, including:

[0189] The comprehensive risk calculation submodule is used to calculate the comprehensive risk coefficient of each defect location based on the risk density value, root cause inference chain, and physical parameters of each defect location.

[0190] The defect evolution prediction submodule is used to build the current defect evolution network based on the root cause tracing reasoning chain of all defect locations, and to perform defect evolution prediction based on the current defect evolution network and the comprehensive risk coefficient of each defect location, so as to obtain a complete defect evolution prediction network.

[0191] The optimization scheme generation submodule is used to generate multiple hypothetical optimization schemes based on the complete defect evolution prediction network, and simulate the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme. The hypothetical optimization scheme with the largest risk reduction rate among all hypothetical optimization schemes is regarded as the best optimization scheme.

[0192] In this embodiment, the risk density value, root cause inference chain, and weights of physical parameters for each defect location are determined using the analytic hierarchy process (AHP). The comprehensive risk coefficient for each defect location is then calculated by combining these factors. This process involves decomposing the complex problem into multiple levels, establishing a hierarchical model, comparing the importance of elements at the same level pairwise, constructing a judgment matrix, calculating weight vectors, and performing consistency checks. These steps determine the relative importance (i.e., weights) of the risk density value, root cause inference chain, and physical parameters in assessing defect risk. Then, based on these weights, the risk density value, root cause inference chain, and physical parameters corresponding to each defect location are comprehensively calculated to obtain a numerical value that fully reflects the risk level of that defect location—the comprehensive risk coefficient.

[0193] In this embodiment, the comprehensive risk coefficient for each defect location is used to comprehensively assess the risk level of each defect location in the cable accessory installation process.

[0194] In this embodiment, the current defect evolution network is constructed based on the root cause tracing reasoning chains of all defect locations: the root cause tracing reasoning chains of each defect location are used as basic units, and a network structure is constructed according to the possible logical connections and mutual influence relationships between them. In this network, nodes can represent different defect locations or processes, and edges represent the evolutionary relationships between defects or the influence paths between processes.

[0195] In this embodiment, defect evolution prediction is performed based on the current defect evolution network and the comprehensive risk coefficient of each defect location to obtain a complete defect evolution prediction network: by utilizing the relationship between defects reflected by the current defect evolution network and combining the comprehensive risk coefficient of each defect location, the future development trend of defects is predicted through simulation, reasoning and other methods.

[0196] Now, we can predict future defect development trends using methods such as simulation and reasoning. A reasoning approach similar to causal analysis can be employed, based on known defect relationships and risk coefficients. Because defect location A has a high risk coefficient and is associated with defect location B, it is predicted that defect location A is highly likely to worsen further. For example, as current flows, loose conductor connections may worsen, leading to increased local resistance and temperature. Simultaneously, considering the impact of defect location A on defect location B, and the risk coefficient of defect location B itself, it is predicted that the deterioration of A may cause further expansion of the damage at defect location B, and may even lead to new defects, such as increased leakage risk due to severe insulation damage. This potential new defect is marked as defect location C.

[0197] In this embodiment, the complete defect evolution prediction network is a network model obtained by predicting defect evolution based on the current defect evolution network and considering the comprehensive risk coefficient of each defect location. It not only includes the existing relationships between defects during the current cable accessory installation process but also predicts the possible development direction and trend of future defects.

[0198] In this embodiment, the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme is used to evaluate the effectiveness of each hypothetical optimization scheme in reducing the risk of cable accessory installation process, so as to screen out the best optimization scheme.

[0199] To extract modifiable variables from a complete defect evolution prediction network to generate hypothetical optimization schemes, the network node state parameters are updated based on the input schemes. The risk index before and after optimization is analyzed to derive the risk reduction rate. The scheme with the largest risk reduction rate is selected as the optimal optimization scheme. Further, an optimization scheme generation submodule is proposed, including:

[0200] The hypothesis optimization scheme generation unit is used to extract all the modifiable variables from the complete defect evolution prediction network and generate a variety of hypothesis optimization schemes based on all the modifiable variables.

[0201] The defect optimization evolution prediction unit is used to input each hypothetical optimization scheme into the complete defect evolution prediction network and update the state parameters of all network nodes to generate the defect optimization evolution prediction network.

[0202] The pre- and post-optimization risk index analysis unit is used to analyze the pre-optimization risk index of the complete defect evolution prediction network, and simulate the defect development trend in the future period based on the defect optimization evolution prediction network to generate the post-optimization risk index of each hypothetical optimization scheme.

[0203] The risk reduction rate analysis unit is used to analyze the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme based on the risk index before optimization and the risk index after optimization for each hypothetical optimization scheme.

[0204] The optimal optimization scheme selection unit is used to select the optimal optimization scheme as the one with the largest risk reduction rate among all the hypothetical optimization schemes.

[0205] In this embodiment, all modifiable variables are extracted from the complete defect evolution prediction network: within the complete defect evolution prediction network, all factors that can be changed through human intervention or operational adjustments are identified and selected. These factors will subsequently be used to generate improvement plans. Modifiable variables refer to parameters, conditions, or operational steps in the cable accessory installation process that can be adjusted or controlled manually. Changes in these parameters may affect the evolution trend of defects, such as the force, temperature, and type of tools used during installation.

[0206] In this embodiment, multiple hypothetical optimization schemes are generated based on all the modifiable variables: by using the extracted modifiable variables, multiple hypothetical schemes are created through different combinations and value settings.

[0207] Suppose that three controllable variables are extracted in the cable accessory installation process: installation temperature, crimping force, and insulation material coating thickness.

[0208] Step 1: Create a solution using combination methods:

[0209] Option 1: Change the installation temperature and pressing force. Set the installation temperature to a higher value (e.g., increase it from the usual 25℃ to 35℃), and at the same time increase the pressing force (from the usual 50N to 70N). Observe the impact of this combination on the installation process. The aim is to make the material more malleable by increasing the temperature, and with greater pressing force, make the connection tighter and reduce the risk of defects in the connection.

[0210] Option 2: Adjust the crimping force and insulation material coating thickness. Reduce the crimping force to 40N, while increasing the insulation material coating thickness (from the usual 2mm to 3mm). This option takes into account that a smaller crimping force may avoid damaging the cable, while increasing the insulation material thickness can improve insulation performance, thereby reducing the risk of insulation-related defects.

[0211] The second step is to set the creation scheme by retrieving values:

[0212] Option 3: Change only the installation temperature. Set the installation temperature to 15℃, keeping the crimping force at 50N and the insulation material coating thickness at 2mm unchanged. Explore the impact of lower temperature environments on the installation process. Low temperatures may alter certain material properties, thus affecting installation quality, thereby determining the specific impact of temperature on defect risk.

[0213] In this embodiment, each hypothetical optimization scheme is input into the complete defect evolution prediction network and the state parameters of all network nodes are updated to generate a defect optimization evolution prediction network: each hypothetical optimization scheme is applied to the complete defect evolution prediction network, and the parameters corresponding to the process, defect and other states represented by each node in the network are adjusted according to the scheme, so as to obtain a new network that reflects the defect evolution under the optimization scheme.

[0214] Assuming the complete defect evolution prediction network describes the cable accessory installation process, network nodes represent different procedures (such as cable stripping, conductor connection, and insulation treatment) and potential defects (such as conductor damage and poor insulation). Each node has corresponding state parameters; for example, the parameters for the cable stripping procedure node include stripping length and speed; the parameters for the conductor damage defect node include damage degree and probability of occurrence.

[0215] Take a hypothetical optimization scheme as an example: the scheme proposes to increase the temperature of the conductor connection process and replace the insulation material.

[0216] Input this scheme into the complete defect evolution prediction network:

[0217] Process node parameter adjustment: Locate the node representing the conductor connection process and increase its temperature parameter from the original normal value (let's say 25°C) to a new setting (e.g., 35°C). This will change the state of the process node, as temperature changes can affect the quality and speed of the connection.

[0218] Defect node parameter adjustment: Changes in conductor connection temperature may affect defect nodes related to conductor connections. For example, the probability of a poor conductor connection defect node occurring may decrease with increasing temperature, say from 10% to 5%. Similarly, replacing insulation materials may affect defect nodes with poor insulation, such as adjusting their risk level parameters from higher to lower values ​​because the new insulation material may have better performance.

[0219] After adjusting the parameters of all affected processes and defect nodes in the network according to the optimization scheme, a defect optimization evolution prediction network is generated. This new network reflects the possible evolution of defects during cable accessory installation under the optimization scheme of "increasing the temperature of the conductor connection process and replacing the insulation material," such as which defects have a reduced probability of occurrence and which processes may have different effects on the overall process due to parameter changes.

[0220] In this embodiment, the state parameters of the network nodes are numerical values ​​or indicators that describe the characteristics and states of each node in the complete defect evolution prediction network. For example, a node representing a process may have parameters such as operation time and operation accuracy, while a node representing a defect may have parameters such as risk level and development speed. These parameters determine the state of the node in the network and its interaction with other nodes.

[0221] In this embodiment, the pre-optimization risk index of the complete defect evolution prediction network is analyzed: by comprehensively evaluating and calculating the state parameters of each node, the relationship between nodes, and the defect evolution trend in the complete defect evolution prediction network, a quantitative value is obtained. This value is used to characterize the defect risk level faced by the entire cable accessory installation process before any hypothetical optimization scheme is implemented.

[0222] Assume the complete defect evolution prediction network includes multiple process nodes and defect nodes in cable accessory installation. Process nodes include cable cutting, conductor stranding, and insulation treatment; defect nodes include conductor damage and insulation layer breakage.

[0223] Step 1, Node state parameter evaluation:

[0224] Process Node: Cable cutting nodes have status parameters such as cutting speed and cutting accuracy. Assuming a standard cutting speed of 10cm / s and an accuracy error of ±0.5mm, if the cutting speed increases to 15cm / s, the accuracy error increases to ±1mm, potentially increasing the risk of conductor damage. This change has a significant impact on the overall risk and is assigned a high weight (e.g., 0.4). Based on the assessment, the current combined score for cutting speed and accuracy is 80 points (out of 100). Therefore, this node contributes 80 × 0.4 = 32 points to the risk index.

[0225] Defect Node: Conductor damage nodes have parameters such as damage degree and probability of occurrence. If the damage degree is divided into minor, moderate, and severe, the current assessment is moderate damage with a probability of occurrence of 20%, and the comprehensive assessment score is 60 points (out of 100). Assigning a weight of 0.3, this node contributes 60 × 0.3 = 18 points to the risk index.

[0226] The second step is to assess the relationship between nodes: the conductor stranding process node and the conductor damage defect node are related. If improper conductor stranding increases the risk of conductor damage, assuming the correlation between the two is 80% (out of 100), a weight of 0.2 is assigned. The assessment shows that this correlation scores 70 points (out of 100) on the risk impact, so the correlation contributes 70 × 0.2 = 14 points to the risk index.

[0227] The third step is to assess the defect evolution trend: It was observed that the insulation layer damage defect tends to worsen over time. Assuming that this worsening trend scores 75 points (out of 100), a weight of 0.1 is assigned. Therefore, this defect evolution trend contributes 75 × 0.1 = 7.5 points to the risk index.

[0228] The fourth step is to calculate the pre-optimization risk index: Add up the contribution scores of each item above, resulting in 32 + 18 + 14 + 7.5 = 71.5 points. This 71.5 points is the pre-optimization risk index of the complete defect evolution prediction network, used to characterize the defect risk level faced by the entire cable accessory installation process before implementing any hypothetical optimization schemes. A higher score indicates a higher risk level.

[0229] In this embodiment, the defect development trend in the future period is simulated based on the defect optimization evolution prediction network, and the post-optimization risk index of each hypothetical optimization scheme is generated: with the help of the defect optimization evolution prediction network, the possible development and changes of defects in the future period are simulated according to certain rules and algorithms. Then, based on the simulation results, the defect risk quantification value of the cable accessory installation process after the implementation of the hypothetical optimization scheme is obtained by using an evaluation calculation method similar to that used in the analysis of the pre-optimization risk index. That is, the post-optimization risk index.

[0230] Suppose there is a defect optimization evolution prediction network for cable accessory installation, which involves process nodes such as cable stripping, conductor connection, and insulation treatment, as well as defect nodes such as conductor damage and poor insulation.

[0231] The first step is to simulate the development trend of defects in the future:

[0232] Suppose an optimization strategy involves increasing the temperature of the conductor connection process and improving the insulation material. In the defect optimization evolution prediction network, rules and algorithms are set based on physical principles and past experience. For example, for the conductor connection process node, increasing the temperature improves the connection tightness and reduces the probability of conductor damage defects according to a certain functional relationship. Assuming the original probability of conductor damage was 20%, according to the algorithm, increasing the temperature reduces the probability of conductor damage to 10%. Following these rules, the development and changes of various defects over the next six months are simulated.

[0233] The second step is to generate the optimized risk index: just like the previous analysis of the pre-optimization risk index, we evaluate the node state parameters, the relationship between nodes, and the defect evolution trend, and then add up the contribution scores of the above items to obtain the optimized risk index under the assumed optimization scheme.

[0234] In this embodiment, the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme is analyzed based on the risk index before optimization and the risk index after optimization for each hypothetical optimization scheme: the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme is obtained by subtracting the risk index after optimization from the risk index before optimization and then dividing the difference by the risk index before optimization. This percentage is used to measure the effectiveness of each hypothetical optimization scheme in reducing the risk of cable accessory installation process.

[0235] The cable accessory installation process traceability system based on temporal attention and digital twins provided in the above embodiments can capture real-time temporal data of cable installers' hand movements and acquire defect data for each process, providing a comprehensive and real-time data foundation for subsequent analysis. By extracting features from the temporal data of hand movements through a temporal attention network and generating a causal model based on the correlation between construction actions and process quality, it can deeply explore the intrinsic connection between actions and process quality, providing a strong basis for process evaluation. A defect source heatmap is generated based on the causal model and defect data, displaying the possible sources of defects in an intuitive and visual way, facilitating rapid location of problem areas. Correlation analysis is performed on the output of the causal model of the preceding process and real-time defect data to identify cross-process analysis results, helping to discover potential influence relationships between different processes and comprehensively grasp the process. Based on the defect source heatmap, cross-process analysis results, and defect data of each process, root cause tracing and defect evolution prediction are performed to generate the best optimization solution, accurately locating the root cause of the problem and predicting the defect development trend, providing targeted optimization strategies. In summary, this system can combine multi-source data to achieve in-depth analysis and traceability of the cable accessory installation process. By collecting and processing multimodal data during the installation process, using temporal attention networks to explore the relationship between actions and process quality, and leveraging digital twin technology to achieve visualized traceability of process defects, the quality of cable accessory installation process can be effectively improved, ensuring the stable and reliable operation of the cable system and helping to enhance the management level of cable accessory installation process.

[0236] like Figure 8 As shown, this embodiment of the invention also provides a method for tracing the installation process of cable accessories based on temporal attention and digital twins, including the following steps:

[0237] S1: Acquire real-time captured timing data of cable installers' hand movements, simultaneously generate local 3D point cloud data of cable accessories, and acquire defect data for each process. Specifically, this includes:

[0238] S11: Acquire real-time timing data of the hand movements of cable installers captured by a wearable operation recorder equipped with an inertial measurement unit and motion sensors.

[0239] S12: Acquire image data of the cable accessory installation location captured in real time by a high-precision binocular vision terminal equipped with a depth camera and image processor, and generate local three-dimensional point cloud data based on the real-time captured image data;

[0240] S13: Obtain defect data for each process based on dedicated sensor units and local 3D point cloud data of each process.

[0241] S2: A causal model of action sequences and process quality indicators is generated by extracting features from temporal data of hand movements using a temporal attention network and combining this with the correlation between construction actions and process quality. Specifically, this includes:

[0242] S21: Encode the hand movement timing data and extract timing features through a bidirectional long short-term memory network to obtain hand movement timing features;

[0243] S22: Calculate the attention weights of the hand action temporal features at different time steps, and fuse the hand action temporal features based on the attention weights at different time steps to obtain process fingerprint features focused on key actions;

[0244] S23: Based on process fingerprint features and corresponding process quality index data, establish a causal relationship model between action sequences and process quality indicators.

[0245] S3: Generate a defect source heatmap based on local 3D point cloud data. Specifically, this includes:

[0246] S31: Construct a geometric model of cable accessories based on local 3D point cloud data;

[0247] S32: Map process parameters to the cable accessory geometry model and define material properties and boundary conditions;

[0248] S33: Based on the geometric model, material properties and boundary conditions of cable accessories, simulate the installation process and quality evolution under different process parameters, and generate a defect source heat map reflecting the risk of process defects at different locations.

[0249] S4: Perform correlation analysis on the relationships between processes to identify cross-process analysis results. Specifically, this includes:

[0250] S41: Decompose the cable accessory installation process into multiple steps and construct a directed graph that reflects the dependencies between the steps;

[0251] S42: Perform a convolution operation on the directed graph and learn the relationships between nodes to obtain the relationships between processes;

[0252] S43: Identify the influence path of the preceding process on subsequent defects based on the relationship between processes as the result of cross-process analysis.

[0253] S5: Root cause analysis is performed based on defect source heatmaps, cross-process analysis results, and defect data from each process. Specifically, this includes:

[0254] S51: Extract the risk density values ​​of all defect locations from the defect source heat map, extract the transmission probability of each influence path from the cross-process analysis results, and obtain the physical parameters of each defect location from the defect data of each process.

[0255] S52: Calculate the posterior probability of each influencing path in the cross-process analysis results, and label the defect evolution contribution of the preceding process for each defect location based on the posterior probability of each influencing path.

[0256] S53: Based on the defect evolution contribution of all preceding processes at each defect location and the transmission probability of all influence paths of each defect location in the cross-process analysis results, generate the root cause tracing inference chain for each defect location.

[0257] In some preferred embodiments, the method further includes:

[0258] S6: Predict defect evolution and generate the optimal solution. Specifically, this includes:

[0259] S61: Calculate the comprehensive risk coefficient of each defect location based on the risk density value, root cause inference chain, and physical parameters.

[0260] S62: Construct the current defect evolution network based on the root cause tracing reasoning chain of all defect locations, and predict the defect evolution based on the current defect evolution network and the comprehensive risk coefficient of each defect location to obtain a complete defect evolution prediction network.

[0261] S63: Generate multiple hypothetical optimization schemes based on the complete defect evolution prediction network, and simulate the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme. The hypothetical optimization scheme with the largest risk reduction rate among all hypothetical optimization schemes is regarded as the best optimization scheme.

[0262] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0263] It should be understood that the functional unit modules in the various embodiments of the present invention can be concentrated in one processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated into one unit module, and can be implemented in hardware or software.

[0264] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A cable accessory installation process traceability system based on temporal attention and digital twins, characterized in that, include: The multimodal data acquisition module is used to capture real-time timing data of the hand movements of cable installers, generate local three-dimensional point cloud data of cable accessories, and acquire defect data of each process. The temporal process fingerprint extraction module is used to extract features from the temporal data of hand movements through a temporal attention network and combine the correlation between construction actions and process quality to generate a causal model of action sequences and process quality indicators. The digital twin dynamic mapping module is used to generate a defect source tracing heat map based on local 3D point cloud data; The cross-process correlation analysis module is used to perform correlation analysis on the relationships between processes and identify the cross-process analysis results; The defect root cause tracing module is used to trace the root causes based on the defect tracing heatmap, cross-process analysis results, and defect data of each process.

2. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 1, characterized in that, The multimodal data acquisition module includes: The hand motion data acquisition module is used to capture the timing data of the hand motions of cable installers in real time through a wearable operation recorder equipped with an inertial measurement unit and motion sensors. The 3D point cloud data acquisition module is used to capture image data of the cable accessory installation location in real time through a high-precision binocular vision terminal equipped with a depth camera and an image processor, and generate local 3D point cloud data based on the real-time captured image data. The defect data acquisition module is used to acquire defect data for each process based on dedicated sensor units and local 3D point cloud data of each process.

3. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 1, characterized in that, The timing process fingerprint extraction module includes: The temporal feature extraction submodule is used to encode the hand movement temporal data and extract temporal features through a bidirectional long short-term memory network to obtain hand movement temporal features; The attention mechanism submodule is used to calculate the attention weights of the hand action temporal features at different time steps, and to fuse the hand action temporal features based on the attention weights at different time steps to obtain process fingerprint features focused on key actions. The causal model construction submodule is used to establish a causal relationship model between action sequences and process quality indicators based on process fingerprint features and corresponding process quality indicator data.

4. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 1, characterized in that, The digital twin dynamic mapping module includes: The geometric model construction submodule is used to construct the geometric model of cable accessories based on local 3D point cloud data; The physical property mapping submodule is used to map process parameters to the cable accessory geometry model and define material properties and boundary conditions; The dynamic simulation submodule is used to simulate the installation process and quality evolution under different process parameters based on the geometric model, material properties and boundary conditions of cable accessories, and generate a defect source heat map that reflects the risk of process defects at different locations.

5. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 1, characterized in that, The cross-process correlation analysis module includes: The process diagram construction submodule is used to decompose the cable accessory installation process into multiple processes and construct a directed graph that reflects the dependencies between processes. The graph neural network submodule is used to perform convolution operations on the directed graph and learn the relationships between nodes to obtain the relationships between processes. The early warning generation submodule is used to identify the impact path of the preceding process on subsequent defects based on the relationship between processes, and to serve as the result of cross-process analysis.

6. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 1, characterized in that, The defect root cause tracing module includes: The defect parameter parsing submodule is used to extract the risk density value of all defect locations from the defect source heat map, extract the transmission probability of each influence path from the cross-process analysis results, and obtain the physical parameters of each defect location from the defect data of each process. The Defect Contribution Accumulation Submodule is used to calculate the posterior probability of each influence path in the cross-process analysis results, and to label the defect evolution contribution of the preceding process for each defect location based on the posterior probability of each influence path. The root cause reasoning submodule is used to generate the root cause reasoning chain for each defect location based on the defect evolution contribution of all preceding processes at each defect location and the transmission probability of all influence paths of each defect location in the cross-process analysis results.

7. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 6, characterized in that, The defect contribution accumulation submodule includes: The path posterior probability calculation unit is used to calculate the posterior probability that a corresponding influence path causes all defect types involved in a single influence path, given that all defect types involved in a single influence path are known in the cross-process analysis results. The same source dependency path analysis unit is used to filter out all the same source dependency path combinations from all the influence paths involved in the cross-process analysis results, and calculate the dependency coefficient of each same source dependency path combination. The homogeneous posterior probability calculation unit is used to calculate the posterior probability of each homogeneous dependent path combination based on the premise that the corresponding influence path is the posterior probability of all defect types involved in the corresponding influence path and the dependency coefficient of each homogeneous dependent path combination, given that all defect types involved in a single influence path are known in the cross-process analysis results. The contribution transmission factor definition unit is used to back-transmit the posterior probability of each homologous dependent path combination to the preceding process node and define all contribution transmission factors for each influencing path. The transmission factor tracing and accumulation unit is used to trace and accumulate all contribution transmission factors of all influence paths to obtain the defect evolution contribution of the preceding process at each defect location.

8. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 7, characterized in that, The posterior probability calculation unit includes: The prior probability calculation subunit is used to calculate the prior probability of each influencing path in the cross-process analysis results; The likelihood calculation subunit is used to calculate the likelihood of all defect types involved in the corresponding influence path under the given condition of the corresponding influence path; The posterior probability calculation subunit is used to calculate the posterior probability that the corresponding influence path causes all defect types involved in the corresponding influence path, based on the prior probability of each influence path in the cross-process analysis results, the likelihood of all defect types involved in the corresponding influence path, and the prior probability of all defect types involved in the corresponding influence path, given that all defect types involved in the corresponding influence path are known.

9. The cable accessory installation process traceability system based on temporal attention and digital twins according to any one of claims 1 to 8, characterized in that, Also includes: The optimization scheme generation module is used to predict the evolution of defects and generate the best optimization scheme.

10. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 9, characterized in that, The optimization scheme generation module includes: The comprehensive risk calculation submodule is used to calculate the comprehensive risk coefficient of each defect location based on the risk density value, root cause inference chain, and physical parameters of each defect location. The defect evolution prediction submodule is used to build the current defect evolution network based on the root cause tracing reasoning chain of all defect locations, and to perform defect evolution prediction based on the current defect evolution network and the comprehensive risk coefficient of each defect location, so as to obtain a complete defect evolution prediction network. The optimization scheme generation submodule is used to generate multiple hypothetical optimization schemes based on the complete defect evolution prediction network, and simulate the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme. The hypothetical optimization scheme with the largest risk reduction rate among all hypothetical optimization schemes is regarded as the best optimization scheme.

11. The cable accessory installation process traceability system based on temporal attention and digital twin as described in claim 10, characterized in that, The optimization scheme generation submodule includes: The defect optimization evolution prediction unit is used to input each hypothetical optimization scheme into the complete defect evolution prediction network and update the state parameters of all network nodes to generate the defect optimization evolution prediction network. The pre- and post-optimization risk index analysis unit is used to analyze the pre-optimization risk index of the complete defect evolution prediction network, and simulate the defect development trend in the future period based on the defect optimization evolution prediction network to generate the post-optimization risk index of each hypothetical optimization scheme. The risk reduction rate analysis unit is used to analyze the risk reduction rate of the complete defect evolution prediction network under each hypothetical optimization scheme based on the risk index before optimization and the risk index after optimization for each hypothetical optimization scheme. The optimal optimization scheme selection unit is used to select the optimal optimization scheme as the one with the largest risk reduction rate among all the hypothetical optimization schemes.

12. A method for tracing the installation process of cable accessories based on temporal attention and digital twins, characterized in that, Includes the following steps: The system acquires real-time captured time-series data of the hand movements of cable installers, generates local 3D point cloud data of cable accessories, and acquires defect data for each process. By using a temporal attention network to extract features from the temporal data of hand movements and combining the correlation between construction actions and process quality, a causal model of action sequences and process quality indicators is generated. Generate a defect source heat map based on local 3D point cloud data; Perform correlation analysis on the relationships between processes to identify cross-process analysis results; Root cause tracing is performed based on defect source heatmaps, cross-process analysis results, and defect data from each process.

13. The cable accessory installation process traceability method based on temporal attention and digital twin as described in claim 12, characterized in that, Also includes: Predict the evolution of defects and generate the best optimization scheme.

Images