Vehicle electronic control assembly workshop monitoring management method

Abstract

The invention relates to a vehicle electronic control assembly workshop monitoring and management method, and belongs to the technical field of workshop monitoring and management. The method comprises the following steps: collecting qualified and fault torque-angle curves through tightening experiments covering screws of all heights, calculating a slope safety threshold, a slope attenuation coefficient and a stroke coefficient as physical parameters, and establishing a parameter library through working condition parameter vectorization and by taking the physical parameters as labels; establishing an identification model and a compensation model based on the constructed parameter library, and optimizing the models through transfer learning; a static feature vector is generated in real time by reading bolt specifications and adopting an identification model, a compensation model is adopted to fuse static features and a real-time torque sequence to update physical parameters, the tightening state is monitored through double strategies, and abnormal types are classified when faults occur. According to the method, a screw physical parameter library is constructed, static working condition recognition and a dynamic friction compensation model are combined, bolt specifications are matched in real time, a torque curve is monitored through double strategies, and highly-adaptive tightening control is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of workshop monitoring and management, and in particular relates to a monitoring and management method for a vehicle electronic control assembly workshop. Background Art

[0002] In the traditional electronic assembly technology, most circuit board tightening processes still rely on manual tightening guns, relying on operators to visually confirm whether the screws on the circuit board have been correctly tightened. Some semi-automated equipment controls by counting the number of screws tightened to reduce the risk of missed tightening. However, manual visual and counting confirmation methods have problems such as low accuracy and susceptibility to human interference, making it difficult to effectively guarantee assembly quality. To improve the stability and reliability of the tightening process, current methods have introduced fixed torque thresholds to monitor the quality of the tightening process, thereby making up for the shortcomings of traditional technologies in quality control to a certain extent.

[0003] However, there are often structures such as connectors on circuit boards, which leads to differences in the tightening depth of screws at different locations. Although some tightening guns can identify torque values, it is still possible that the torque meets the requirements but the screws are not tightened in place. Under the same program, the monitoring of the tightening angle relies on a static threshold, and this method cannot effectively distinguish the tightening conditions of screws of different depths. At the same time, the actual workshop working environment is relatively complex, and there are screws of various specifications applied to workstations at different heights of the circuit board. In addition, factors such as oil pollution and temperature changes can cause fluctuations in the friction coefficient, and transient faults such as floating nails and thread damage often occur, resulting in a single threshold monitoring method that is difficult to cope with these differentiated working scenarios. Summary of the Invention

[0004] To solve the above-mentioned problems existing in the prior art, the present invention provides a monitoring and management method for a vehicle electronic control assembly workshop. By constructing a physical parameter library of screws, combining static working condition identification with a dynamic friction compensation model, using real-time matching of bolt specifications and adopting a dual-strategy monitoring torque curve, highly adaptive tightening control is achieved.

[0005] The purpose of the present invention can be achieved through the following technical solutions: The present disclosure provides a monitoring and management method for a vehicle electronic control assembly workshop, comprising the following steps: S1. Parameter library construction: Through tightening experiments covering all height screws, qualified and failed torque-angle curves are collected, and the slope safety threshold, slope attenuation coefficient, and stroke coefficient are calculated as physical parameters. The parameter library is established by vectorizing the working condition parameters and using the physical parameters as labels; S2. Machine learning model training: Establish recognition and compensation models based on the constructed parameter library, and optimize the models through transfer learning; S3. Tightening status monitoring: By reading the bolt specifications, the recognition model is used to generate static feature vectors in real time. The compensation model is used to fuse the static features with the real-time torque sequence to update the physical parameters. The tightening status is monitored through a dual strategy, and the abnormal type is classified in case of failure.

[0006] Furthermore, the parameter library is constructed, comprising the following steps: S11. Data Collection: Tightening tests are conducted on screws of all heights, and torque-angle curves are collected for each type of screws, including those that pass and fail. Failures include floating screws, thread damage, metallic foreign matter, skewness, and incorrect model. S12. Parameter initialization: Based on the qualified torque-angle curve, the bolt deformation stage is divided into the screw-in stage, the stretching stage, and the final tightening stage. The slope of each stage of the qualified curve is calculated. The safety threshold range is obtained by eliminating 5% of extreme values. At the same time, strain gauges are attached to the bolt shaft. During the process of tightening the bolt to fracture, the point where the strain suddenly increases is defined as the yield point. The slope attenuation coefficient and stroke coefficient are calculated. S13. Dataset integration: Build a parameter library by vectorizing the operating parameters and generating datasets with physical parameters as labels.

[0007] Furthermore, the calculation formula of the slope attenuation coefficient β is: ; Where K e is the average slope of the stretching section, K y is the instantaneous slope of the yield point, H is the bolt height; The calculation formula of the stroke coefficient γ is: ; Where θ y is the yield point angle, θ e is the end angle of the stretch segment.

[0008] Furthermore, the machine learning model training includes the following steps: S21. Working condition identification: The working condition identification is completed by establishing an identification model, and the ratio threshold and slope attenuation coefficient are obtained through the static feature mapping of the working condition; S22, dynamic compensation: by establishing a compensation model to obtain friction fluctuations and transient faults, dynamic compensation of ratio threshold and slope attenuation coefficient is achieved; S23. Performance optimization: Enhance fault feature extraction through transfer learning, input the output features into the recognition model and compensation model, and optimize the model.

[0009] Furthermore, the operating condition identification includes the following steps: S211. Feature engineering: Convert the original features of bolt height, thread specification, material hardness, mounting surface roughness, and circuit board temperature into feature vectors by encoding; S212. Establish a CNN-Transformer model: Establish a model architecture including a convolutional layer and a Transformer encoder, wherein the convolutional layer is used to extract local feature combinations, and the Transformer encoder is used to calculate the global dependency between features, and the output is a parameter group of a ratio threshold and a slope attenuation coefficient; S213, Model training: Input the feature vector into the CNN-Transformer model to obtain the output prediction, calculate the mean square error between the predicted value and the working condition label in the parameter library, and use the optimizer to update the weight.

[0010] Furthermore, the dynamic compensation includes the following steps: S221. Establish an LSTM-GRU model: Take the torque sequence of the screw-in stage as input and the correction value of the ratio threshold and slope attenuation coefficient as output, build a compensation model for LSTM and GRU dual-path parallel processing, and splice the LSTM and GRU outputs through the feature fusion layer; S222, Model training: Through the torque sequence input and deviation label data, apply the time series Dropout forward propagation, calculate the loss and inject gradient noise for reverse update, and finally obtain a dynamic correction model that can predict the working condition deviation.

[0011] Furthermore, the performance optimization comprises the steps of: The screw-in torque sequence is converted into a time-frequency spectrum through STFT transformation. Using the ResNet architecture, the first-layer convolution kernel is modified to adapt to the long strip structure of the spectrum. The number of input channels is adjusted from three RGB channels to a single channel to match the grayscale torque spectrum input. The ImageNet pre-trained weights of the first three ResBlocks are retained to maintain general feature extraction capabilities.

[0012] Furthermore, the performance optimization also includes the steps of: unfreezing and training the last ResBlock and the adaptive pooling layer to enable them to learn the specific features of the screw working condition; compressing the features through the global average pooling layer to output a 512-dimensional high-semantic fault feature vector.

[0013] Furthermore, the tightening status monitoring includes the following steps: S31. Parameter preloading: When the device starts, the machine learning model and parameter library are preloaded to establish a sensor communication link. The parameter library data includes the ratio threshold, slope attenuation coefficient, and altitude parameter mapping table. S32, screw identification: Use RFID to read the bolt specifications, and generate static feature vectors in real time through the identification model to obtain the initial ratio threshold and slope attenuation coefficient; S33, real-time dynamic monitoring: Torque angle data is collected at 0.5° intervals, and moving average filtering and outlier removal are performed. A compensation model is used to fuse static features with real-time torque sequences, output corrections for the ratio threshold and slope attenuation coefficient, and update the actual ratio threshold and slope attenuation coefficient. S34, fault diagnosis: when the alarm mechanism sounds an alarm, freeze the data, call the fault classifier to identify the abnormality type, input the abnormal segment curve and static feature vector into the fault classifier to obtain the fault type; S35, data closed-loop collection: store all tightening curves and model decision data, and mark low-confidence samples for self-evolution.

[0014] Furthermore, the real-time dynamic monitoring further includes the steps of: Abnormal alarm is performed through dual-strategy monitoring of ratio monitoring and gradient method monitoring, where: Ratio monitoring: segmented verification to see if the slope is within the actual ratio threshold range; Gradient method monitoring: Real-time detection of whether the slope decay triggers the yield point or floating nail condition.

[0015] The beneficial effects of the present invention are: The present invention optimizes the quality control of bolts during the tightening process through systematic parameter library construction, machine learning model training, and real-time monitoring. First, by covering the tightening experiments of all height screws, the torque-angle curves of failure and qualification are collected, and the slope safety threshold, slope attenuation coefficient, and stroke coefficient are calculated to establish a database containing working condition parameters and physical parameter labels; then the CNN-Transformer model is used to perform static feature recognition on different working conditions, and the LSTM-GRU model is used to dynamically compensate for torque fluctuations and transient faults, thereby optimizing the control accuracy during the bolt tightening process; then the bolt specifications are identified through RFID technology, torque data is collected in real time, and anomalies are detected through dual-strategy monitoring to ensure timely alarm and classification when a fault occurs; finally, the entire system collects tightening curves and model decision data through a data closed-loop mechanism for self-optimization and fault diagnosis. This method effectively overcomes problems such as height differences and oversensitivity of small-stroke screws through machine learning optimization and highly adaptive process parameter adjustment, improves the accuracy of bolt tightening and fault diagnosis, and ensures the stability and reliability of the assembly process.

[0016] This method quantifies the essential parameters of bolt deformation through physical experiments and constructs a monitoring architecture that integrates static and dynamic AI. A CNN-Transformer architecture identifies static operating conditions and generates adaptive thresholds. An LSTM-GRU architecture integrates real-time torque sequences to compensate for dynamic friction fluctuations. Time-frequency spectrum transfer learning is introduced to extract high-dimensional fault signatures, enhancing the generalization capabilities of the dual models and achieving a dual closed-loop drive based on physical laws and data intelligence. This integration of material mechanical properties, operating condition dynamic response, and fault spectrum fingerprints forms a complete closed-loop from physical essence to data intelligence, fundamentally solving the industrial monitoring challenges of highly differentiated and complex fault scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0018] Figure 1 A schematic diagram of the steps of a vehicle electronic control assembly workshop monitoring and management method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0020] This invention is based on the process of tightening screws on circuit boards during the assembly of vehicle electronic controls. This method proposes a method for determining the tightening status based on the combined monitoring of torque and angle, addressing the problem of connectors and other structures often present on circuit boards, which results in different screw tightening depths at different locations and causes the tightening gun to fail to identify the tightening process by identifying torque. Specifically, the present invention describes a monitoring and management method for vehicle electronic control assembly workshops in detail through the following content: This embodiment provides a monitoring and management method for a vehicle electronic control assembly workshop, such as Figure 1 As shown, the following steps are included: S1. Parameter library construction: Through tightening experiments covering all height screws, qualified and fault torque-angle curves are collected, and the slope safety threshold, slope attenuation coefficient, and stroke coefficient are calculated as physical parameters. The parameter library is established by vectorizing the working condition parameters and using the physical parameters as labels, including the following steps: S11. Data collection: Tightening tests are conducted on screws of all heights, and torque-angle curves of qualified and failed screws are collected for each type. Failures include floating screws, thread damage, metal foreign matter, skewness, and model errors.

[0021] It should be noted that in the fault curve experiment, a 0.05mm polyimide film was pre-placed at the bottom of the threaded hole to simulate a misalignment, a tap with a wear loss of 0.2mm was used to produce burrs to simulate thread damage, chips were injected to simulate metal foreign matter, the tightening gun was installed at a 5° tilt to simulate skew, and screws with mixed length deviations were used to simulate model errors. Among them, the torque-angle curve is the core data carrier for characterizing the bolt tightening process. By simulating faults, it is ensured that the collected data can cover scenarios with abnormal curve morphology and yield point offset. The data acquisition equipment includes a torque sensor and an optical encoder, which respectively record changes of 0.1N·m and the screw rotation trajectory.

[0022] S12. Parameter initialization: Based on the qualified torque-angle curve, the bolt deformation stage is divided into the screw-in stage, the stretching stage, and the final tightening stage. The slope of each stage of the qualified curve is calculated. The safety threshold range is obtained by eliminating 5% of extreme values. At the same time, strain gauges are attached to the bolt shaft. When the bolt is tightened to fracture, the point where the strain suddenly increases is defined as the yield point. The slope attenuation coefficient and stroke coefficient are calculated.

[0023] It should be noted that the end point of the screw-in segment is determined by the inflection point of the first-order derivative of the torque curve, the end point of the stretching segment is determined by the zero crossing of the second-order derivative of the torque curve, and the end point of the final tightening segment is determined when the bolt enters the plastic deformation zone (irreversible damage). The slope attenuation coefficient is determined by the ratio of the yield point slope drop rate to the bolt height. The slope drop rate is the ratio of the difference between the average slope of the stretching segment and the instantaneous slope of the yield point to the average slope of the stretching segment. The stroke coefficient is the ratio of the difference between the yield point angle and the end point angle of the stretching segment to the bolt height.

[0024] Specifically, the calculation formula of the slope attenuation coefficient β is: ; Where K e is the average slope of the stretching section, K y is the instantaneous slope at the yield point, and H is the bolt height.

[0025] The calculation formula of stroke coefficient γ is: ; Where θ y is the yield point angle, θ e is the end angle of the stretch segment.

[0026] S13. Dataset integration: Build a parameter library by vectorizing the operating parameters and generating datasets with physical parameters as labels.

[0027] It should be noted that parameter initialization establishes a ratio monitoring threshold based on the statistical slope of the qualified curve, and destructive testing is used to determine the yield point and calculate the gradient method β and γ parameters, thereby digitizing the physical properties of the bolt. The operating parameters include bolt height, thread size, material hardness, mounting surface roughness, and circuit board temperature.

[0028] S2. Machine learning model training: Build the recognition model and compensation model based on the constructed parameter library, and optimize the model through transfer learning, including the following steps: S21. Working condition identification: The working condition identification is completed by establishing an identification model, and the ratio threshold and slope attenuation coefficient are obtained through the static feature mapping of the working condition, including the following steps: S211. Construct feature engineering: Convert the original features of bolt height, thread specification, material hardness, mounting surface roughness and circuit board temperature into feature vectors by encoding.

[0029] It can be understood that by converting the original feature numerical vector, the original features are input into the subsequent machine learning model through structured encoding.

[0030] S212. Establish a CNN-Transformer model: Establish a model architecture including convolutional layers and Transformer encoders, where the convolutional layers are used to extract local feature combinations (such as the combined effect of height and hardness), and the Transformer encoder is used to calculate the global dependency between features (such as the amplification effect of temperature on roughness). The output is a parameter set of ratio threshold and slope attenuation coefficient.

[0031] S213, Model training: Input the feature vector into the CNN-Transformer model to obtain the output prediction, calculate the mean square error between the predicted value and the working condition label in the parameter library, and use the optimizer to update the weight.

[0032] The CNN-Transformer model is a hybrid model that combines a convolutional neural network (CNN) and a transformer (Transformer) structure. It uses CNN for feature extraction and Transformer for capturing sequences.

[0033] S22. Dynamic compensation: A compensation model is established to obtain friction fluctuations and transient faults, and to achieve dynamic compensation of ratio threshold and slope attenuation coefficient, including the following steps: S221. Establish an LSTM-GRU model: take the torque sequence of the screw-in segment as input and the correction amount of the ratio threshold and slope attenuation coefficient as output, build a compensation model with LSTM (long short-term memory network) and GRU (gated recurrent unit) dual-path parallel processing, and splice the LSTM and GRU outputs through the feature fusion layer.

[0034] S222, Model training: Through the torque sequence input and deviation label data, apply the time series Dropout forward propagation, calculate the loss and inject gradient noise for reverse update, and finally obtain a dynamic correction model that can predict the working condition deviation.

[0035] It should be noted that during the training process, the training data also includes ratio threshold and slope attenuation deviation label data. The deviation label data is obtained through the fault curve test experiment during the construction of the parameter library. The golden label is obtained for different fault conditions. During the model training stage, the deviation mapping is completed by learning the fault characteristics, thereby solving the generalization problem.

[0036] S23. Performance Optimization: Enhance fault feature extraction through transfer learning, input the output features into the recognition model and compensation model, and optimize the model, including the following steps: The screw insertion torque sequence is converted into a time-frequency spectrum through STFT (short-time Fourier transform) transformation (window length = 16 points, overlap rate 75%). The ResNet (residual network) architecture is adopted, and the first-layer convolution kernel is modified to 3×15 to adapt to the long strip structure of the spectrum. The number of input channels is adjusted from three RGB channels to a single channel to match the grayscale torque spectrum input. The ImageNet pre-trained weights of the first three ResBlocks (residual blocks) are retained to maintain general feature extraction capabilities. The last ResBlock and adaptive pooling layer are unfrozen and trained to learn the screw working condition-specific features. The features are compressed through the global average pooling layer to output a 512-dimensional high-semantic fault feature vector.

[0037] It should be noted that the time-spectrogram contains more fault mode information (such as high-frequency transients) than the original sequence. Furthermore, pre-trained ResNet can extract features that are more robust than manually designed ones. Finally, the 512-dimensional vector serves as a shared feature, simultaneously optimizing the structural flaws of both models. For the CNN-Transformer, static features may have omitted dynamic features related to the working conditions (such as torque fluctuation patterns caused by oil contamination). Now, the 512-dimensional vector supplements this information through the concat operation. This is even more critical for the LSTM-GRU. The original time series has a low signal-to-noise ratio for short-term mutations, while the spectral features significantly enhance the ability to identify periodic fault patterns. For example, a screw crack can cause energy anomalies in a specific frequency band, which is submerged in the noise in the original torque data. The essence of the 512-dimensional feature is to encode physical fault phenomena into machine-readable high-dimensional fingerprints, providing downstream models with fault identification capabilities that exceed the raw data, thereby achieving breakthroughs in prediction accuracy.

[0038] S3. Tightening status monitoring: By reading the bolt specifications and using the recognition model to generate static feature vectors in real time, a compensation model is used to fuse the static features with the real-time torque sequence to update the physical parameters. The tightening status is monitored through a dual strategy, and the abnormal type is classified in the event of a fault. The following steps are included: S31. Parameter preloading: When the device starts, the machine learning model and parameter library are preloaded to establish a sensor communication link. The parameter library data includes the ratio threshold, slope attenuation coefficient, and altitude parameter mapping table.

[0039] It should be noted that in this embodiment, a detour controller is used to establish a millisecond-level communication link with the torque / angle sensor, and an RFID reader is started at the same time to identify the bolt identity tag.

[0040] S32. Screw identification: RFID is used to read the bolt specifications (height / diameter / material), and the static feature vector is generated in real time through the identification model to obtain the initial ratio threshold and slope attenuation coefficient.

[0041] S33, real-time dynamic monitoring: Torque angle data is collected at 0.5° intervals, and moving average filtering and outlier removal are performed. A compensation model is used to fuse static features with real-time torque sequences, output corrections for the ratio threshold and slope attenuation coefficient, and update the actual ratio threshold and slope attenuation coefficient. Abnormal alarm is performed through dual-strategy monitoring of ratio monitoring and gradient method monitoring, where: Ratio monitoring: segmented verification to see if the slope is within the actual ratio threshold range; Gradient method monitoring: Real-time detection of whether the slope decay triggers the yield point or floating nail condition.

[0042] S34, fault diagnosis: when the alarm mechanism sounds an alarm, freeze the data, call the fault classifier to identify the abnormality type, and input the abnormal segment curve and static feature vector into the fault classifier to obtain the fault type.

[0043] It should be noted that the fault classifier is trained through the fault torque-angle curve in the parameter library and the static feature vector obtained by the recognition model, which will not be described in detail here.

[0044] S35, data closed-loop collection: store all tightening curves and model decision data, and mark low-confidence samples for self-evolution.

[0045] It should be noted that ratio monitoring divides the complete torque-angle curve into multiple monitoring intervals. By calculating the torque change rate of each interval, if the actual ratio of any interval exceeds the threshold range during real-time tightening, an alarm is triggered. By analyzing the changes in the curve shape in real time, process anomalies (such as sudden changes in resistance during the screw-in stage) are captured. Gradient method monitoring is that the bolt undergoes elastic deformation, yield point, and plastic deformation stages during the tightening process. In the elastic stage, the torque and angle are linearly related (the slope is constant). At the yield point, the slope drops significantly (the material begins to plastically deform). The gradient method monitoring calculates the rate of change of the curve slope to accurately capture the yield point and terminate tightening, avoiding failure of the bolt due to over-tightening. This embodiment addresses the problems of ratio monitoring being unable to solve the end point judgment deviation caused by height differences and the gradient method monitoring being overly sensitive to small-stroke screws. By highly adaptively matching process parameters, it solves the problem of tightening quality control in highly differentiated scenarios.

[0046] This invention optimizes bolt quality control during the tightening process through systematic parameter library construction, machine learning model training, and real-time monitoring. First, through tightening experiments covering all screw heights, failed and qualified torque-angle curves are collected, and the slope safety threshold, slope attenuation coefficient, and stroke coefficient are calculated to establish a database containing operating parameters and physical parameter labels. A CNN-Transformer model is then used to perform static feature recognition for different operating conditions. An LSTM-GRU model is then used to dynamically compensate for torque fluctuations and transient faults, optimizing control accuracy during the bolt tightening process. RFID technology is then used to identify bolt specifications, collect torque data in real time, and detect anomalies through dual-strategy monitoring (ratio monitoring and gradient method monitoring), ensuring timely alarm and classification when faults occur. Finally, the entire system collects tightening curves and model decision data through a closed-loop data mechanism for self-optimization and fault diagnosis. Through machine learning optimization and highly adaptive process parameter adjustment, this method effectively overcomes issues such as height variations and oversensitivity of small-stroke screws, improving bolt tightening accuracy and fault diagnosis accuracy, and ensuring the stability and reliability of the assembly process.

[0047] This method quantifies the essential parameters of bolt deformation (slope attenuation coefficient β and stroke coefficient γ) through physical experiments, constructing a monitoring architecture that integrates static and dynamic AI. A CNN-Transformer model identifies static operating conditions and generates adaptive thresholds. An LSTM-GRU model integrates real-time torque sequences to compensate for dynamic friction fluctuations. Time-frequency spectrum transfer learning is introduced to extract high-dimensional fault signatures, enhancing the generalization capabilities of the dual models and achieving a dual closed-loop driven by physical laws and data intelligence. This integration of material mechanical properties (β / γ), operating condition dynamic response (LSTM-GRU compensation), and fault spectrum fingerprints (ResNet features) forms a complete closed-loop from physical essence to data intelligence, fundamentally solving the industrial monitoring challenges of highly differentiated and complex fault scenarios.

[0048] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A monitoring and management method for a vehicle electronic control assembly workshop, characterized by: The following steps are involved: Parameter library construction: Through tightening experiments covering screws of all heights, qualified and failed torque-angle curves are collected. The slope safety threshold, slope attenuation coefficient, and stroke coefficient are calculated as physical parameters. The parameter library is established by vectorizing the working condition parameters and using the physical parameters as labels. Machine learning model training: Build recognition and compensation models based on the constructed parameter library, and optimize the models through transfer learning; Tightening status monitoring: By reading the bolt specifications, the recognition model is used to generate static feature vectors in real time. The compensation model is used to fuse the static features with the real-time torque sequence to update the physical parameters. The tightening status is monitored through a dual strategy, and the abnormal type is classified in the event of a fault.

2. A vehicle electronic control assembly workshop monitoring and management method according to claim 1, characterized in that: The parameter library construction includes the following steps: Data Collection: Tightening tests are conducted on screws of all heights, and torque-angle curves are collected for both qualified and failed screws of each type. Failures include floating screws, thread damage, metallic foreign matter, skewness, and incorrect model. Parameter initialization: Based on the qualified torque-angle curve, the bolt deformation stage is divided into the screw-in stage, the stretching stage, and the final tightening stage. The slope of each stage of the qualified curve is calculated, and the safety threshold range is obtained by eliminating 5% of extreme values. Strain gauges are also attached to the bolt shaft. During the process of tightening the bolt to fracture, the point where the strain suddenly increases is defined as the yield point. The slope attenuation coefficient and stroke coefficient are calculated. Dataset integration: Build a parameter library by vectorizing operating condition parameters and generating datasets with physical parameters as labels.

3. The vehicle electronic control assembly workshop monitoring and management method according to claim 2, characterized in that: The calculation formula of the slope attenuation coefficient β is: ； Where K e is the average slope of the stretching section, K y is the instantaneous slope of the yield point, H is the bolt height; The calculation formula of the stroke coefficient γ is: ； Where θ y is the yield point angle, θ e is the end angle of the stretch segment.

4. The vehicle electronic control assembly workshop monitoring and management method according to claim 1, characterized in that: The machine learning model training includes the following steps: Working condition identification: The working condition identification is completed by establishing an identification model, and the ratio threshold and slope attenuation coefficient are obtained through the static feature mapping of the working condition; Dynamic compensation: By establishing a compensation model to obtain friction fluctuations and transient faults, dynamic compensation of ratio threshold and slope attenuation coefficient is achieved; Performance optimization: Enhance fault feature extraction through transfer learning, input the output features into the recognition model and compensation model to optimize the model.

5. The vehicle electronic control assembly workshop monitoring and management method according to claim 4, characterized in that: The working condition identification, The following steps are involved: Feature engineering: Bolt height, thread specification, material hardness, mounting surface roughness, and circuit board temperature are encoded to convert the original features into feature vectors. Build a CNN-Transformer model: Build a model architecture consisting of convolutional layers and a Transformer encoder. The convolutional layers are used to extract local feature combinations, and the Transformer encoder is used to calculate the global dependencies between features. The output is a parameter set of ratio thresholds and slope decay coefficients. Model training: Input the feature vector into the CNN-Transformer model to obtain the output prediction, calculate the mean square error between the predicted value and the working condition label in the parameter library, and use the optimizer to update the weight.

6. The vehicle electronic control assembly workshop monitoring and management method according to claim 5, characterized in that: The dynamic compensation comprises the following steps: Establish an LSTM-GRU model: Using the torque sequence of the screw-in stage as input and the correction value of the ratio threshold and slope attenuation coefficient as output, a compensation model with LSTM and GRU dual-path parallel processing is constructed, and the LSTM and GRU outputs are spliced ​​through a feature fusion layer; Model training: Through torque sequence input and deviation label data, time series Dropout forward propagation is applied, and after calculating the loss, gradient noise is injected for reverse update, ultimately obtaining a dynamic correction model that can predict operating condition deviations.

7. The vehicle electronic control assembly workshop monitoring and management method according to claim 6, characterized in that: The performance optimization comprises the steps of: The screw-in torque sequence is converted into a time-frequency spectrum through STFT transformation. Using the ResNet architecture, the first-layer convolution kernel is modified to adapt to the long strip structure of the spectrum. The number of input channels is adjusted from three RGB channels to a single channel to match the grayscale torque spectrum input. The ImageNet pre-trained weights of the first three ResBlocks are retained to maintain general feature extraction capabilities.

8. The vehicle electronic control assembly workshop monitoring and management method according to claim 7, characterized in that: The performance optimization also includes the steps of: unfreezing and training the last ResBlock and the adaptive pooling layer to enable them to learn the specific features of the screw working condition; compressing the features through the global average pooling layer to output a 512-dimensional high-semantic fault feature vector.

9. The vehicle electronic control assembly workshop monitoring and management method according to claim 1, characterized in that: The tightening status monitoring comprises the following steps: Parameter preloading: When the device starts up, it preloads the machine learning model and parameter library to establish a sensor communication link. The parameter library data includes ratio thresholds, slope attenuation coefficients, and altitude parameter mapping tables. Screw identification: RFID is used to read the bolt specifications, and the static feature vector is generated in real time through the identification model to obtain the initial ratio threshold and slope attenuation coefficient; Real-time dynamic monitoring: Torque angle data is collected at 0.5° intervals, and moving average filtering and outlier removal are performed. A compensation model is used to fuse static features with real-time torque sequences, outputting corrections to the ratio threshold and slope attenuation coefficient, and updating the actual ratio threshold and slope attenuation coefficient. Fault diagnosis: When the alarm mechanism sounds, freeze the data, call the fault classifier to identify the abnormality type, and input the abnormal segment curve and static feature vector into the fault classifier to obtain the fault type; Closed-loop data collection: Store all tightening curves and model decision data, and mark low-confidence samples for self-evolution.

10. The vehicle electronic control assembly workshop monitoring and management method according to claim 9, characterized in that: The real-time dynamic monitoring further comprises the steps of: Abnormal alarm is performed through dual-strategy monitoring of ratio monitoring and gradient method monitoring, where: Ratio monitoring: segmented verification to see if the slope is within the actual ratio threshold range; Gradient method monitoring: Real-time detection of whether the slope decay triggers the yield point or floating nail condition.

Images