A tightening quality assessment method based on deep learning

Abstract

The present invention discloses a tightening quality assessment method based on deep learning, which relates to the technical field of tightening quality assessment. The present invention collects torque, preload, lubrication status and environmental data in real time during the tightening process, uses an LSTM dynamic model to capture the nonlinear fluctuation of the friction factor during repeated tightening, and promptly corrects the torque-preload conversion relationship. Adaptive learning rate and fully connected layer mapping are adopted to enable the model to automatically update parameters under different environmental conditions, dynamically adjust the tightening speed, preset torque and lubrication strategy. The environmental integration module quantifies changes in temperature, humidity and salt concentration, corrects process parameters in real time, and effectively suppresses instantaneous abnormal fluctuations.

Description

Technical Field

[0001] The present invention relates to the technical field of tightening quality assessment, and in particular to a tightening quality assessment method based on deep learning. Background Art

[0002] The current bolt and nut tightening quality assessment system mainly relies on the calibration of the fixed relationship between input torque and preload. However, in actual production, during repeated tightening operations, as the number of tightening times increases, the contact surface morphology of the bolt and nut will change, and the lubrication layer will gradually be consumed, causing the friction coefficient to show irregular fluctuations, and a large deviation will appear between the actual preload and the theoretical value.

[0003] At the same time, different ambient temperatures and humidity levels will also affect the tightening process. For example, in the connection operation of offshore wind turbines, drastic changes in environmental conditions will have unpredictable effects on the surface state of the material and the lubrication effect, and the transmission relationship between torque and preload force will be more complicated. In this case, traditional methods mostly use standardized lubrication schemes and preset process parameters. Obviously, this will cause the preload force of some workpieces to be too low or too high under specific environments; therefore, a tightening quality assessment method based on deep learning is urgently needed to solve this problem. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a tightening quality assessment method based on deep learning to solve the problem that traditional tightening quality assessment schemes cannot effectively assess the nonlinear changes during repeated tightening and have insufficient environmental adaptability assessment.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] An embodiment of the present invention provides a tightening quality assessment method based on deep learning, which includes:

[0008] Step S1, adding environmental simulation, and using a torque sensor, a preload sensor, and temperature, humidity, and salt spray environmental parameter sensors to collect real-time data during the tightening process;

[0009] Step S2: fusing and preprocessing the real-time data to extract quality features; the quality features include changes in contact surface state and fluctuations in friction coefficient caused by repeated tightening;

[0010] Step S3: constructing a dynamic model based on the quality features extracted in step S2. The dynamic model uses an LSTM neural network to fit and predict the changes in the nonlinear friction factor caused by surface state changes and environmental influences during repeated tightening, and to correct the conversion relationship between torque and preload force.

[0011] Step S4, using the output of the dynamic model in step S3, adjusting the tightening process parameters in real time, including adjusting the tightening speed, preset torque, and lubricant replenishment strategy;

[0012] Step S5: Build a digital twin platform, simulate and integrate the collected real-time data with the model prediction results, perform full-cycle status visual monitoring of the tightening process, and generate a real-time evaluation report.

[0013] As a preferred solution of the tightening quality assessment method based on deep learning described in the present invention, in step S1, the collected data includes tightening torque, preload force, contact surface lubrication status and environmental indicators.

[0014] As a preferred solution of the tightening quality assessment method based on deep learning described in the present invention, wherein: the preprocessing includes data denoising, normalization and eigenvalue statistics;

[0015] The data preprocessing step further includes performing statistical analysis on the data of each tightening cycle during the repeated tightening process to quantify the impact of the contact surface morphology and the change of the lubricating layer on the friction coefficient.

[0016] As a preferred solution of the deep learning-based tightening quality assessment method described in the present invention, the training of the dynamic model is based on historical tightening test data, and the dynamic model includes a model adaptive parameter adjustment unit for automatically updating the model weights under different environmental conditions.

[0017] As a preferred solution of the tightening quality assessment method based on deep learning described in the present invention, in step S3, the steps of fitting and predicting are:

[0018] The dynamic model uses the following LSTM unit calculation formula to process time series data:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] ,

[0024] ,

[0025] in, represents the input gate output vector, is the sigmoid activation function, is the input weight matrix of the input gate, For the moment The input feature vector of is the hidden state weight matrix of the input gate, is the hidden state vector at the previous moment, is the bias vector of the input gate, represents the forget gate output vector, is the input weight matrix of the forget gate, is the hidden state weight matrix of the forget gate, is the bias vector of the forget gate, is the candidate memory vector, is the hyperbolic tangent activation function, is the input weight matrix of the candidate memory, is the hidden state weight matrix of the candidate memory, is the bias vector of the candidate memory, is the updated memory state vector, is the memory state vector of the previous moment, represents element-wise multiplication operation, represents the output vector of the output gate, is the input weight matrix of the output gate, is the hidden state weight matrix of the output gate, is the bias vector of the output gate, is the hidden state vector at the current moment;

[0026] The network output is mapped to the friction factor correction through the fully connected layer, and the formula is:

[0027] ,

[0028] in, is the friction factor correction value output by the network, is the fully connected layer mapping matrix, is the hidden state vector at the current moment, is the bias vector of the fully connected layer.

[0029] As a preferred embodiment of the method for tightening quality assessment based on deep learning described in the present invention, in step S3, the step of fitting and predicting further includes:

[0030] The torque is corrected using the prediction results, and the relationship is expressed as:

[0031] ,

[0032] in, is the corrected torque, is the original torque, is the adjustment coefficient, is the friction factor correction;

[0033] The mean square error loss function is used for model training. The loss function formula is:

[0034] ,

[0035] in, is the mean square error loss, is the sample size, is the friction factor correction value predicted by the network, is the actual friction factor change value;

[0036] Adaptive learning rate is introduced. The adaptive learning rate calculation formula is:

[0037] ,

[0038] in, is the adaptive learning rate at the current moment, is the initial learning rate, represents the exponential function, is the attenuation factor, is the current error change rate, which is calculated based on the difference in errors before and after.

[0039] As a preferred solution of the tightening quality assessment method based on deep learning described in the present invention, wherein: in step S5, the method of simulating and integrating the collected real-time data with the model prediction results is: integrating the real-time data collected by the sensor with the model output, and defining the simulation state vector as :

[0040] ,

[0041] in, Indicates time The simulation state vector, is the sensor data fusion weight, For the moment The sensor collects data vector, Predict data fusion weights for the model, The predicted value output by the model;

[0042] The deviation between real-time data and model prediction is calculated to quantify the state difference of the tightening process. The calculation formula is: ,in, For the moment The deviation of represents the norm operation;

[0043] Generate an alarm signal based on the deviation: If ,but ,like ,but ,in, It is an alarm flag. A value of 1 indicates that the deviation exceeds the preset threshold. is the deviation threshold parameter.

[0044] As a preferred solution of the tightening quality assessment method based on deep learning described in the present invention, in step S5, the digital twin platform calibrates the data according to a unified update cycle, and the update cycle is recorded as , expressed as:

[0045] ,

[0046] in, Indicates the data update cycle;

[0047] The platform calculates and reports indicators using the following formula:

[0048] ,

[0049] in, Indicates time The reporting index, is the weight factor of the simulation state norm, is the weight factor of the alarm sign.

[0050] As a preferred embodiment of the deep learning-based tightening quality assessment method described in the present invention, the digital twin platform is constructed using virtual simulation technology to digitally replicate the tightening process based on real-time data feedback, thereby promptly alerting on-site operators when deviations occur between the preload force and the friction state.

[0051] The environmental parameter sensor adopts an integrated module and operates synchronously with the tightening device to dynamically adjust the lubrication and tightening process parameters under conditions of sudden changes in temperature and humidity or salt spray corrosion.

[0052] As a preferred embodiment of the tightening quality assessment method based on deep learning described in the present invention, in the digital twin platform, the environmental parameter sensor integrated module dynamically adjusts the lubrication and tightening process parameters by collecting environmental data in real time and combining it with an adjustment algorithm under conditions of sudden changes in temperature and humidity or salt spray corrosion. Specifically:

[0053] Integrated module to collect temperature ,humidity and salt concentration Environmental data, calculation of environmental severity factors , the calculation formula is:

[0054] ,

[0055] in, Indicates the current temperature. Indicates the current humidity. Indicates the current salt concentration. is the temperature sensitive weight, is the humidity sensitivity weight, is the salt concentration sensitive weight, 、 and are the reference ambient temperature, humidity and salt concentration, respectively. is an exponential function;

[0056] Based on environmental factors, the lubrication parameters are dynamically adjusted. The adjustment formula is:

[0057] ,

[0058] in, is the lubrication parameter after adjustment, is the base lubrication parameter, is the lubrication parameter adjustment coefficient,

[0059] Correct the tightening process parameters, the formula is:

[0060] ,

[0061] in, is the corrected tightening process parameter, As the benchmark tightening process parameters, is the tightening parameter adjustment coefficient;

[0062] The sliding average filter is used to smooth the environmental severity factor in the time domain. The expression is:

[0063] ,

[0064] in, is the smoothed environmental factor, is the time window length, Represents the index of the time series.

[0065] The beneficial effects of the present invention are as follows: the present invention collects torque, preload, lubrication status and environmental data during the tightening process in real time, uses the LSTM dynamic model to capture the nonlinear fluctuations of the friction factor during repeated tightening, and promptly corrects the torque-preload conversion relationship; adopts adaptive learning rate and fully connected layer mapping, so that the model can automatically update parameters under different environmental conditions, dynamically adjust the tightening speed, preset torque and lubrication strategy; the environmental integration module quantifies changes in temperature, humidity and salt concentration, corrects process parameters in real time, and effectively suppresses instantaneous abnormal fluctuations.

[0066] The present invention effectively alleviates the problems of nonlinearity and insufficient environmental adaptability in repeated tightening of traditional methods, and realizes high-precision, real-time and intelligent tightening quality evaluation and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] Figure 1 This is a flow chart of the deep learning-based tightening quality assessment method in Example 1. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0070] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0072] Example 1, reference Figure 1 , this embodiment provides a tightening quality assessment method based on deep learning, including:

[0073] Step S1, adding environmental simulation, and using a torque sensor, a preload sensor, and temperature, humidity, and salt spray environmental parameter sensors to collect real-time data during the tightening process;

[0074] The collected data include tightening torque, preload, contact surface lubrication status and environmental indicators;

[0075] Step S2: fusing and preprocessing the real-time data to extract quality features; quality features include changes in contact surface state and fluctuations in friction coefficient caused by repeated tightening;

[0076] Preprocessing includes data denoising, normalization and eigenvalue statistics;

[0077] The data preprocessing step further includes statistical analysis of the data from each tightening cycle during repeated tightening to quantify the impact of contact surface morphology and lubrication layer changes on the friction factor;

[0078] Step S3: constructing a dynamic model based on the quality features extracted in step S2. The dynamic model uses an LSTM neural network to fit and predict the changes in the nonlinear friction factor caused by surface state changes and environmental influences during repeated tightening, and to correct the conversion relationship between torque and preload force.

[0079] The training of the dynamic model is based on historical tightening test data. The dynamic model includes a model adaptive parameter adjustment unit for automatically updating the model weights under different environmental conditions.

[0080] In step S3, the steps of fitting and prediction are:

[0081] The dynamic model uses the following LSTM unit calculation formula to process time series data:

[0082] ,

[0083] ,

[0084] ,

[0085] ,

[0086] ,

[0087] ,

[0088] in, represents the input gate output vector, is the sigmoid activation function, is the input weight matrix of the input gate, For the moment The input feature vector of is the hidden state weight matrix of the input gate, is the hidden state vector at the previous moment, is the bias vector of the input gate, represents the forget gate output vector, is the input weight matrix of the forget gate, is the hidden state weight matrix of the forget gate, is the bias vector of the forget gate, is the candidate memory vector, is the hyperbolic tangent activation function, is the input weight matrix of the candidate memory, is the hidden state weight matrix of the candidate memory, is the bias vector of the candidate memory, is the updated memory state vector, is the memory state vector of the previous moment, represents element-wise multiplication operation, represents the output vector of the output gate, is the input weight matrix of the output gate, is the hidden state weight matrix of the output gate, is the bias vector of the output gate, is the hidden state vector at the current moment;

[0089] The network output is mapped to the friction factor correction through the fully connected layer, and the formula is:

[0090] ,

[0091] in, is the friction factor correction value output by the network, is the fully connected layer mapping matrix, is the hidden state vector at the current moment, is the bias vector of the fully connected layer;

[0092] In step S3, the steps of fitting and predicting also include:

[0093] The torque is corrected using the prediction results, and the relationship is expressed as:

[0094] ,

[0095] in, is the corrected torque, is the original torque, is the adjustment coefficient, is the friction factor correction;

[0096] The mean square error loss function is used for model training. The loss function formula is:

[0097] ,

[0098] in, is the mean square error loss, is the sample size, is the friction factor correction value predicted by the network, is the actual friction factor change value;

[0099] Adaptive learning rate is introduced. The adaptive learning rate calculation formula is:

[0100] ,

[0101] in, is the adaptive learning rate at the current moment, is the initial learning rate, represents the exponential function, is the attenuation factor, is the current error change rate, calculated based on the error difference before and after;

[0102] Specifically, an LSTM neural network dynamic model is constructed to input the temporal quality features of the tightening process into the network for processing. The model uses a temporal structure consisting of an input gate, a forget gate, a candidate memory, and an output gate to capture the nonlinear fluctuations of the friction factor caused by changes in surface conditions and environmental factors.

[0103] The fully connected layer mapping and subsequent torque correction provide a basis for correcting the actual torque data, adjusting the conversion relationship in real time. The mean square error loss function ensures that the error between the predicted value and the actual value is minimized during the model training process. The adaptive learning rate mechanism dynamically adjusts the weight update rate according to different working conditions, and can respond to subtle changes during multiple tightening operations, thereby improving the accuracy of the overall tightening process assessment.

[0104] Step S4, using the output of the dynamic model in step S3, adjusting the tightening process parameters in real time, including adjusting the tightening speed, preset torque, and lubricant replenishment strategy;

[0105] Step S5: Build a digital twin platform to simulate and integrate the collected real-time data with the model prediction results, perform visual monitoring of the full-cycle state of the tightening process, and generate a real-time evaluation report;

[0106] In step S5, the method of simulating and integrating the collected real-time data with the model prediction results is as follows: the real-time data collected by the sensor is integrated with the model output, and the simulation state vector is defined as :

[0107] ,

[0108] in, Indicates time The simulation state vector, is the sensor data fusion weight, For the moment The sensor collects data vector, Predict data fusion weights for the model, The predicted value output by the model;

[0109] The deviation between real-time data and model prediction is calculated to quantify the state difference of the tightening process. The calculation formula is: ,in, For the moment The deviation of represents the norm operation;

[0110] Generate an alarm signal based on the deviation: If ,but ,like ,but ,in, It is an alarm flag. A value of 1 indicates that the deviation exceeds the preset threshold. is the deviation threshold parameter;

[0111] In step S5, the digital twin platform calibrates the data according to a unified update cycle, and the update cycle is , expressed as:

[0112] ,

[0113] in, Indicates the data update cycle;

[0114] The platform calculates and reports indicators using the following formula:

[0115] ,

[0116] in, Indicates time The reporting index, is the weight factor of the simulation state norm, is the weight factor of the alarm sign;

[0117] Specifically, the digital twin platform organically integrates real-time data with model prediction results, uses data fusion to build simulation states, and achieves dynamic replication of the entire cycle process. Deviation calculation and alarm mechanisms enable the system to promptly reflect deviations from preload and friction states. Data synchronization ensures data consistency across modules, and reporting indicators integrate simulation states and abnormal prompts, providing on-site operators with intuitive monitoring information and evaluation basis.

[0118] The digital twin platform is built using virtual simulation technology to digitally replicate the tightening process based on real-time data feedback, thereby promptly alerting on-site operators when there is a deviation between the preload force and the friction state;

[0119] The environmental parameter sensor uses an integrated module to operate synchronously with the tightening device, dynamically adjusting the lubrication and tightening process parameters under conditions of drastic changes in temperature and humidity or salt spray corrosion;

[0120] In the digital twin platform, the environmental parameter sensor integrated module dynamically adjusts lubrication and tightening process parameters by collecting real-time environmental data and combining it with adjustment algorithms under conditions of drastic temperature and humidity changes or salt spray corrosion. Specifically:

[0121] Integrated module to collect temperature ,humidity and salt concentration Environmental data, calculation of environmental severity factors , the calculation formula is:

[0122] ,

[0123] in, Indicates the current temperature. Indicates the current humidity. Indicates the current salt concentration. is the temperature sensitive weight, is the humidity sensitivity weight, is the salt concentration sensitive weight, 、 and are the reference ambient temperature, humidity and salt concentration, respectively. is an exponential function;

[0124] Based on environmental factors, the lubrication parameters are dynamically adjusted. The adjustment formula is:

[0125] ,

[0126] in, is the lubrication parameter after adjustment, is the base lubrication parameter, is the lubrication parameter adjustment coefficient,

[0127] Correct the tightening process parameters, the formula is:

[0128] ,

[0129] in, is the corrected tightening process parameter, As the benchmark tightening process parameters, is the tightening parameter adjustment coefficient;

[0130] The sliding average filter is used to smooth the environmental severity factor in the time domain. The expression is:

[0131] ,

[0132] in, is the smoothed environmental factor, is the time window length, Represents the index of the time series.

[0133] Specifically, the integrated module collects environmental parameters in real time and constructs an environmental severity factor using an exponential function to quantify the environmental impact under drastic changes in temperature, humidity, and salt concentration. Based on this factor, the lubrication and tightening process parameters are dynamically adjusted using a linear correction method. The sliding average filter can suppress instantaneous abnormal fluctuations and automatically respond to environmental changes under special working conditions, effectively improving the stability and safety of lubrication and tightening operations. At the same time, it provides intuitive data support for on-site operators, enhancing the adaptability of the overall process and real-time monitoring effects.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A tightening quality assessment method based on deep learning, characterized by: include, Step S1, adding environmental simulation, and using a torque sensor, a preload sensor, and temperature, humidity, and salt spray environmental parameter sensors to collect real-time data during the tightening process; Step S2: fusing and preprocessing the real-time data to extract quality features; the quality features include changes in contact surface state and fluctuations in friction coefficient caused by repeated tightening; Step S3: constructing a dynamic model based on the quality features extracted in step S2. The dynamic model uses an LSTM neural network to fit and predict the changes in the nonlinear friction factor caused by surface state changes and environmental influences during repeated tightening, and to correct the conversion relationship between torque and preload force. Step S4, using the output of the dynamic model in step S3, adjusting the tightening process parameters in real time, including adjusting the tightening speed, preset torque, and lubricant replenishment strategy; Step S5: Build a digital twin platform to simulate and integrate the collected real-time data with the model prediction results, perform visual monitoring of the full-cycle state of the tightening process, and generate a real-time evaluation report; The digital twin platform is built using virtual simulation technology to digitally replicate the tightening process based on real-time data feedback; The environmental parameter sensor uses an integrated module to operate synchronously with the tightening device, dynamically adjusting the lubrication and tightening process parameters under conditions of drastic changes in temperature and humidity or salt spray corrosion; In the digital twin platform, the environmental parameter sensor integrated module dynamically adjusts lubrication and tightening process parameters by collecting environmental data in real time and combining it with an adjustment algorithm under conditions of drastic temperature and humidity changes or salt spray corrosion. Specifically: Integrated module to collect temperature ,humidity and salt concentration Environmental data, calculation of environmental severity factors , the calculation formula is: ， in, Indicates the current temperature. Indicates the current humidity. Indicates the current salt concentration. is the temperature sensitive weight, is the humidity sensitivity weight, is the salt concentration sensitive weight, 、 and are the reference ambient temperature, humidity and salt concentration, respectively. is an exponential function; Based on environmental factors, the lubrication parameters are dynamically adjusted. The adjustment formula is: ， in, is the lubrication parameter after adjustment, is the base lubrication parameter, is the lubrication parameter adjustment coefficient, Correct the tightening process parameters, the formula is: ， in, is the corrected tightening process parameter, As the benchmark tightening process parameters, is the tightening parameter adjustment coefficient; The sliding average filter is used to smooth the environmental severity factor in the time domain. The expression is: ， in, is the smoothed environmental factor, is the time window length, Represents the index of the time series.

2. The method for evaluating tightening quality based on deep learning according to claim 1, wherein: In step S1 , the collected data include tightening torque, preload force, contact surface lubrication status and environmental indicators.

3. The method for tightening quality assessment based on deep learning according to claim 1, wherein: The preprocessing includes data denoising, normalization and eigenvalue statistics; The pre-processing further includes statistically analyzing the data of each tightening cycle during the repeated tightening process to quantify the impact of the contact surface morphology and the change of the lubricating layer on the friction factor.

4. The method for evaluating tightening quality based on deep learning according to claim 1, wherein: The training of the dynamic model is based on historical tightening test data. The dynamic model includes a model adaptive parameter adjustment unit for automatically updating the model weight under different environmental conditions.

5. The method for tightening quality assessment based on deep learning according to claim 4, characterized in that: In step S3, the steps of fitting and predicting are: The dynamic model uses the following LSTM unit calculation formula to process time series data: ， ， ， ， ， ， in, represents the input gate output vector, is the sigmoid activation function, is the input weight matrix of the input gate, For the moment The input feature vector of is the hidden state weight matrix of the input gate, is the hidden state vector at the previous moment, is the bias vector of the input gate, represents the forget gate output vector, is the input weight matrix of the forget gate, is the hidden state weight matrix of the forget gate, is the bias vector of the forget gate, is the candidate memory vector, is the hyperbolic tangent activation function, is the input weight matrix of the candidate memory, is the hidden state weight matrix of the candidate memory, is the bias vector of the candidate memory, is the updated memory state vector, is the memory state vector of the previous moment, represents element-wise multiplication operation, represents the output vector of the output gate, is the input weight matrix of the output gate, is the hidden state weight matrix of the output gate, is the bias vector of the output gate, is the hidden state vector at the current moment; The network output is mapped to the friction factor correction through the fully connected layer, and the formula is: ， in, is the friction factor correction value output by the network, is the fully connected layer mapping matrix, is the hidden state vector at the current moment, is the bias vector of the fully connected layer.

6. The method for tightening quality assessment based on deep learning according to claim 5, wherein: In step S3, the steps of fitting and predicting further include: The torque is corrected using the prediction results, and the relationship is expressed as: ， in, is the corrected torque, is the original torque, is the adjustment coefficient, is the friction factor correction; The mean square error loss function is used for model training. The loss function formula is: ， in, is the mean square error loss, is the sample size, is the friction factor correction value predicted by the network, is the actual friction factor change value; Adaptive learning rate is introduced. The adaptive learning rate calculation formula is: ， in, is the adaptive learning rate at the current moment, is the initial learning rate, represents the exponential function, is the attenuation factor, is the current error change rate, which is calculated based on the difference in errors before and after.

7. The method for tightening quality assessment based on deep learning according to claim 1, wherein: In step S5, the method of simulating and integrating the collected real-time data with the model prediction results is as follows: the real-time data collected by the sensor is integrated with the model output, and the simulation state vector is defined as : ， in, Indicates time The simulation state vector, is the sensor data fusion weight, For the moment The sensor collects data vector, Predict data fusion weights for the model, The predicted value output by the model; The deviation between real-time data and model prediction is calculated to quantify the state difference of the tightening process. The calculation formula is: ,in, For the moment The deviation of represents the norm operation; Generate an alarm signal based on the deviation: If ,but ,like ,but ,in, It is an alarm flag. A value of 1 indicates that the deviation exceeds the preset threshold. is the deviation threshold parameter.

8. The method for tightening quality assessment based on deep learning according to claim 7, wherein: In step S5, the digital twin platform calibrates the data according to a unified update cycle, and the update cycle is , expressed as: ， in, Indicates the data update cycle; The platform calculates and reports indicators using the following formula: ， in, Indicates time The reporting index, is the weight factor of the simulation state norm, is the weight factor of the alarm sign.

Images