Intelligent tightening tool capacity allocation system and method based on deep learning

Abstract

The invention discloses an intelligent tightening tool capacity allocation system and method based on deep learning, and relates to the technical field of tightening tools, multi-modal data is captured in real time, the data weight is dynamically adjusted by using an attention mechanism, the capacity is predicted in combination with a deep learning algorithm, and dynamic production requirements and sudden tasks are quickly responded; in the aspect of exception handling, dual indexes of a variational auto-encoder VAE model, reconstruction error and distribution deviation are introduced, an exception mode is captured, a probability distribution model of normal operation data is established in combination with unsupervised learning, and the false alarm rate and the missing report rate are remarkably reduced; abnormal data are fed back to the productivity prediction module in real time through a closed-loop mechanism, and deep learning model parameters are dynamically optimized, so that the system can adapt to changes of production tasks and attenuation of tool performance; and the allocation module is adopted to allocate tool types and optimize paths through task priorities, so that the resource utilization efficiency is improved, and tool conflicts and station waiting time are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of tightening tools, and in particular to an intelligent tightening tool capacity allocation system and method based on deep learning. Background Art

[0002] In modern industrial production, tightening tools are an important part of the assembly line, and the quality of their capacity allocation method directly affects the overall production efficiency and product quality. In recent years, some allocation schemes have introduced automation and intelligent technologies, but their responsiveness is still insufficient when faced with dynamic production needs. In the assembly line environment, production task requirements often change, such as new tasks, priority adjustments, or equipment failures. However, most of the current allocation strategies are still based on static rules or periodic calculations, which make it difficult to capture and respond to real-time changes in a timely manner. This lag often leads to task queuing, idle resources, or unbalanced capacity allocation, which has an adverse impact on production efficiency.

[0003] To address the problem of delayed response in deployment, some traditional solutions use manual intervention or fixed rules to deal with it. However, this method has limitations such as slow response speed and single rules, and it is difficult to meet the needs of high-frequency adjustment. Therefore, there is an urgent need for an intelligent tightening tool capacity deployment system and method based on deep learning to solve such problems. Summary of the invention

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

[0005] The present invention provides an intelligent tightening tool capacity allocation system and method based on deep learning to solve the problem that traditional solutions have delayed allocation response. Some solutions that use manual intervention or fixed rule processing still have limitations such as slow response speed and single rules, and are difficult to meet the needs of high-frequency adjustment.

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

[0007] In a first aspect, an embodiment of the present invention provides an intelligent tightening tool capacity allocation system based on deep learning, which includes a data acquisition module, a data upload module, a capacity prediction module, an allocation module, an error warning module, a feedback module and a user interface.

[0008] The data acquisition module has a built-in controller, using PLC and PC-based controllers to collect working data of tightening tools, including sensor data, production history records and worker operation records;

[0009] Sensor data includes torque, speed, temperature and humidity;

[0010] Production history records include task sequences and tool usage times;

[0011] Worker operation records include operation frequency and real-time intervention data;

[0012] The data uploading module is used to upload the working data to the capacity prediction module;

[0013] The capacity prediction module uses a deep learning algorithm based on work data to predict capacity and deploy production strategies;

[0014] The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools at each station, as well as the tightening sequence and tightening path;

[0015] The error warning module is used to analyze abnormal data during the tightening process, including torque abnormalities and speed abnormalities, and provide error warnings;

[0016] The feedback module is used to feed back abnormal data to the capacity prediction module in real time;

[0017] The user interface is used by an operator to monitor and adjust the status of a tightening tool.

[0018] As a preferred solution of the deep learning-based intelligent tightening tool capacity allocation system described in the present invention, in which: in the data acquisition module, a multimodal fusion model based on the attention mechanism is adopted to dynamically adjust the weight of the work data and generate comprehensive work features, and the fused work features are input into the allocation module and the error warning module.

[0019] As a preferred solution of the deep learning-based intelligent tightening tool capacity allocation system described in the present invention, wherein: in the error warning module, an anomaly detection model based on variational autoencoder VAE is constructed, and the fused working characteristics are used as input; unsupervised learning is used to establish a probability distribution model of normal operation data, and during operation, the reconstruction error and distribution deviation of real-time data are used as abnormal indicators, and multi-level warnings are performed in combination with business rules.

[0020] In a second aspect, the present invention provides a method for intelligent tightening tool capacity allocation based on deep learning, comprising:

[0021] Step S1, collecting working data during the tightening process through a data acquisition module,

[0022] Step S2: A multimodal fusion model based on the attention mechanism is used for the collected work data to dynamically adjust the weight of the work data and generate comprehensive work features. The fused work features are input into the deployment module and the error warning module.

[0023] Step S3, the capacity prediction module uses a deep learning algorithm to predict capacity based on the integrated work characteristics and deploy production strategies;

[0024] Step S4: The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools for each workstation, as well as the tightening sequence and tightening path; at the same time, the error warning module constructs an anomaly detection model based on the variational autoencoder VAE, takes the fused working features as input, and analyzes the abnormal data in the tightening process, including torque anomalies and speed anomalies; and uses unsupervised learning to establish a probability distribution model for normal operation data. During operation, the reconstruction error and distribution deviation of real-time data are used as abnormal indicators, and multi-level warnings are performed in combination with business rules;

[0025] Step S5: The feedback module feeds back the abnormal data to the capacity prediction module to update the deep learning model in step S3.

[0026] As a preferred solution of the method for capacity allocation of intelligent tightening tools based on deep learning described in the present invention, the steps of using a multimodal fusion model based on an attention mechanism to dynamically adjust the weight of the work data and generate comprehensive work features are as follows:

[0027] After data preprocessing, the multimodal data is standardized and preliminary features are extracted. The extraction formula is:

[0028] H A =f CNN (X A ),H B =f RNN (X B ),H C =f RNN (X C ),

[0029] Among them, H A is the feature matrix of sensor data, H B is the characteristic matrix of production history records, H C is the feature matrix of worker operation records, X A ,X B ,X C is the original data input of the corresponding mode, f CNN ,f RNN are convolutional neural networks and recurrent neural networks respectively.

[0030] Based on the attention mechanism, the weights of each modal feature are dynamically adjusted. The calculation formula is:

[0031]

[0032] Among them, α m is the weight of mode m, m∈{A,B,C}, H m is the preliminary feature of mode m, W Qis the learnable parameter matrix used for weight calculation, k is the modal index of weight normalization, and its value range is {1,2,3},

[0033] After fusing the modal features by weight, the final comprehensive features are generated. The comprehensive feature formula is:

[0034]

[0035] Among them, F final is the comprehensive working characteristic, W F is the weight matrix of the fully connected layer, b F is the bias of the fully connected layer, σ is the nonlinear activation function; the attention mechanism dynamically adjusts the modal weights to effectively generate high-quality comprehensive working features.

[0036] As a preferred solution of the method for capacity allocation of intelligent tightening tools based on deep learning described in the present invention, the steps of predicting capacity based on the fused working characteristics using a deep learning algorithm and allocating production strategies are as follows:

[0037] The capacity prediction module is modeled using a deep feedforward neural network DNN, and the calculation process is as follows:

[0038] Z (l) =σ(W (l) Z (l-1) +b (l) ),l=1,2,…,L-1,

[0039] Y pred =W (L) Z (L-1) +b (L) ,

[0040] Among them, Z (l) is the output feature of the lth layer, Z (0) =F final is the comprehensive working feature of the input layer, W (l) ,b (l) is the weight matrix and bias of the lth layer, Y pred is the final capacity prediction value, L is the total number of layers in the network,

[0041] The mean square error MSE is used as the loss function, which is defined as L:

[0042]

[0043] And update the model parameters by gradient descent method, the update formula is:

[0044]

[0045] Among them, L is the loss value, which represents the average square error between the predicted value and the true value, N is the number of samples, and Y pred,i is the predicted value of the i-th sample, Y true,i is the true capacity value of the u-th sample, θ is the network parameter, including weight and bias, η is the learning rate, and t is the number of iterations. The deep learning model is used here to capture the data pattern and improve the accuracy of capacity prediction.

[0046] As a preferred solution of the method for adjusting the production capacity of intelligent tightening tools based on deep learning described in the present invention, the step of adjusting the production capacity according to the production strategy by the adjustment module is as follows:

[0047] The deployment module adjusts production capacity, including:

[0048] The tools are assigned according to the task priority, and the allocation formula is:

[0049]

[0050] Among them, U j is the type of tool assigned to workstation j, S j,t is the adaptability score of tool t at workstation j,

[0051] The cost function is used to optimize the path. The path optimization formula is:

[0052]

[0053] Among them, C path is the total path cost, d p is the distance of path p, ω p is the task weight of path p, T p The scheduling module improves the capacity utilization through intelligent allocation and path optimization, while the error warning module ensures the stability of production through anomaly detection.

[0054] As a preferred solution of the method for capacity allocation of intelligent tightening tools based on deep learning described in the present invention, the steps of constructing an abnormality detection model based on variational autoencoder VAE, taking the fused working features as input, and analyzing the abnormal data in the tightening process are as follows:

[0055] The potential distribution is constructed through the variational autoencoder VAE, and the construction formula is:

[0056]

[0057] Where z is the latent space representation, μ enc ,σ enc is the mean and standard deviation of the underlying distribution, ∈ is random noise, The features of reconstruction.

[0058] As a preferred solution of the method for capacity allocation of intelligent tightening tools based on deep learning described in the present invention, the steps of using the reconstruction error and distribution deviation of real-time data as abnormal indicators and performing multi-level early warning in combination with business rules are as follows:

[0059] The reconstruction error is calculated as follows:

[0060]

[0061] Among them, E recon is the reconstruction error, which represents the input feature F final Reconstruct output with autoencoder The difference between final is the comprehensive working feature, the input feature matrix fused by the attention mechanism, is the comprehensive feature reconstructed by the autoencoder, the feature matrix generated from the potential representation z by the decoder Decoder(z), ||·|| 2 is the square of the L2 norm;

[0062] Calculate the distribution deviation using the formula:

[0063] D kl =KL(q(z|F final )||p(z)),

[0064] Among them, D kl is the distribution deviation, which represents the approximate posterior distribution q(z|F final ) and the Kullback-Leibler divergence between the prior distribution p(z), q(z|F final ) is the approximate posterior distribution, which is based on the input feature F final The generated potential distribution is a Gaussian distribution: μ enc is the mean vector of the approximate posterior distribution, generated by the encoder, σ enc is the standard deviation vector of the approximate posterior distribution, generated by the encoder, p(z) is the prior distribution, which is the standard normal distribution, 0 is the mean vector of the prior distribution, I is the identity matrix, representing the independent and identically distributed standard normal distribution, and KL(·||·) is the Kullback-Leibler divergence, which is used to measure the relative entropy difference between two probability distributions and is defined as:

[0065]

[0066] By calculating the reconstruction error E recon and distribution deviation D kl, evaluate whether the input features are consistent with the normal operation of the model. Excessive deviation indicates anomaly. The anomaly indicators together constitute the basis for anomaly detection of the autoencoder.

[0067] As a preferred solution of the deep learning-based intelligent tightening tool capacity allocation method described in the present invention, the feedback module compares the abnormal data with the prediction results to generate optimized samples, and the optimized samples are used to update the deep learning model. The feedback module forms a closed-loop system, which enhances the model's ability to handle complex scenarios.

[0068] The beneficial effects of the present invention are as follows: the present invention captures multimodal data in real time, uses the attention mechanism to dynamically adjust data weights, combines deep learning algorithms to predict production capacity, and quickly responds to dynamic production needs and emergency tasks; in terms of exception handling, the variational autoencoder VAE model is introduced, the dual indicators of reconstruction error and distribution deviation are used to capture abnormal patterns, and unsupervised learning is combined to establish a probability distribution model of normal operating data, which significantly reduces the false alarm and missed alarm rates and enhances the robustness of the system; the feedback module feeds back abnormal data to the production capacity prediction module in real time through a closed-loop mechanism, dynamically optimizes the deep learning model parameters, and enables the system to adapt to changes in production tasks and the attenuation of tool performance, thereby achieving long-term self-learning and continuous optimization.

[0069] The present invention adopts a deployment module to allocate tool types and optimize paths through task priorities, thereby improving resource utilization efficiency and reducing tool conflicts and workstation waiting time; multi-modal data fusion is performed to reduce the impact of single-modal data anomalies on system performance, thereby improving overall stability and decision-making accuracy; at the same time, combined with a multi-level early warning mechanism, active early warning is issued before anomalies occur, thereby ensuring the safety of the production process and the consistency of product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0071] Figure 1 A schematic diagram of the capacity allocation and error warning process of the intelligent tightening tool capacity allocation system of the present invention;

[0072] Figure 2 It is a structural schematic diagram of the intelligent tightening tool capacity allocation system of the present invention;

[0073] Figure 3 A workflow diagram of a deployment module of the intelligent tightening tool capacity deployment system of the present invention;

[0074] Figure 4 This is a workflow diagram of the error warning module of the intelligent tightening tool capacity allocation system of the present invention. DETAILED DESCRIPTION

[0075] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0076] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and 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.

[0077] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0078] Example 1, reference Figure 1-Figure 4 , this embodiment provides an intelligent tightening tool capacity allocation system based on deep learning, including:

[0079] Data collection module, data upload module, capacity forecasting module, allocation module, error warning module, feedback module and user interface;

[0080] The data acquisition module has a built-in controller, and PLC and PC-based controllers are selected to collect the working data of the tightening tools, including sensor data, production history records and worker operation records;

[0081] Sensor data includes torque, speed, temperature and humidity;

[0082] Production history records include task sequences and tool usage times;

[0083] Worker operation records include operation frequency and real-time intervention data;

[0084] In the data collection module, a multimodal fusion model based on the attention mechanism is used to dynamically adjust the weight of the work data and generate comprehensive work features. The fused work features are input into the deployment module and the error warning module.

[0085] Data upload module, used to upload work data to the capacity forecasting module;

[0086] The capacity forecasting module uses deep learning algorithms based on work data to forecast capacity and adjust production strategies;

[0087] The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools at each station, as well as the tightening sequence and tightening path;

[0088] Error warning module, used to analyze abnormal data during the tightening process, including torque abnormalities and speed abnormalities, and provide error warnings;

[0089] In the error warning module, an anomaly detection model based on variational autoencoder (VAE) is constructed, with the fused working features as input. Unsupervised learning is used to establish a probability distribution model for normal operation data. During operation, the reconstruction error and distribution deviation of real-time data are used as anomaly indicators, and multi-level warnings are performed in combination with business rules.

[0090] Feedback module, used to feed back abnormal data to the capacity prediction module in real time;

[0091] User interface for operators to monitor and adjust the status of the tightening tool.

[0092] This embodiment also provides a method for allocating production capacity of intelligent tightening tools based on deep learning, including:

[0093] Step S1, collecting working data during the tightening process through a data acquisition module,

[0094] Step S2: A multimodal fusion model based on the attention mechanism is used for the collected work data to dynamically adjust the weight of the work data and generate comprehensive work features. The fused work features are input into the deployment module and the error warning module.

[0095] The multimodal fusion model based on the attention mechanism is used for the collected work data to dynamically adjust the weight of the work data. The steps to generate comprehensive work features are as follows:

[0096] After data preprocessing, the multimodal data is standardized and preliminary features are extracted. The extraction formula is:

[0097] H A =f CNN (X A ),H B =f RNN (X B ),H C =f RNN (X C ),

[0098] Among them, H A is the feature matrix of sensor data, H B is the characteristic matrix of production history records, H C is the feature matrix of worker operation records, X A ,XB ,X C is the original data input of the corresponding mode, f CNN ,f RNN are convolutional neural networks and recurrent neural networks respectively.

[0099] Based on the attention mechanism, the weights of each modal feature are dynamically adjusted. The calculation formula is:

[0100]

[0101] Among them, α m is the weight of mode m, m∈{A,B,C}, H m is the preliminary feature of mode m, W Q is the learnable parameter matrix used for weight calculation, k is the modal index of weight normalization, and its value range is {1,2,3},

[0102] After fusing the modal features by weight, the final comprehensive features are generated. The comprehensive feature formula is:

[0103]

[0104] Among them, F final is the comprehensive working characteristic, W F is the weight matrix of the fully connected layer, b F is the bias of the fully connected layer, and σ is a nonlinear activation function; the attention mechanism dynamically adjusts the modal weights to effectively generate high-quality comprehensive working features;

[0105] Specifically, through the data acquisition module and the data upload module, the multimodal data of the tightening process is captured in real time, and the attention mechanism is used to perform dynamic weight adjustment to generate comprehensive work characteristics; on this basis, the capacity prediction module is modeled based on the deep feedforward neural network DNN to quickly predict future capacity and generate dynamic strategies; the allocation module uses the prediction results to quickly adjust task allocation and path optimization, significantly improving the real-time response capability of the production line.

[0106] Step S3, the capacity prediction module uses a deep learning algorithm to predict capacity based on the integrated work characteristics and deploy production strategies;

[0107] Based on the fused work characteristics, the deep learning algorithm is used to predict production capacity and adjust the production strategy as follows:

[0108] The capacity prediction module uses a deep feedforward neural network (DNN) for modeling, and the calculation process is as follows:

[0109] Z (l) =σ(W (l) Z (l-1 )+b (l)),l=1,2,…,L-1,

[0110] Y pred =W (L) Z (L-1 )+b (L) ,

[0111] Among them, Z (l) is the output feature of the lth layer, Z (0) =F final is the comprehensive working feature of the input layer, W (l) ,b (l) is the weight matrix and bias of the lth layer, Y pred is the final capacity prediction value, L is the total number of layers in the network,

[0112] The mean square error MSE is used as the loss function, which is defined as L:

[0113]

[0114] And update the model parameters by gradient descent method, the update formula is:

[0115]

[0116] Among them, L is the loss value, which represents the average square error between the predicted value and the true value, N is the number of samples, and Y pred,i is the predicted value of the i-th sample, Y true,i is the true capacity value of the i-th sample, θ is the network parameter, including weight and bias, η is the learning rate, and t is the number of iterations. Here, a deep learning model is used to capture data patterns and improve the accuracy of capacity prediction;

[0117] Step S4: The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools for each workstation, as well as the tightening sequence and tightening path; at the same time, the error warning module constructs an anomaly detection model based on the variational autoencoder VAE, takes the fused working features as input, and analyzes the abnormal data in the tightening process, including torque anomalies and speed anomalies; and uses unsupervised learning to establish a probability distribution model for normal operation data. During operation, the reconstruction error and distribution deviation of real-time data are used as abnormal indicators, and multi-level warnings are performed in combination with business rules;

[0118] The steps for the allocation module to adjust production capacity according to the production strategy are as follows:

[0119] The deployment module adjusts production capacity, including:

[0120] The tools are assigned according to the task priority, and the allocation formula is:

[0121]

[0122] Among them, U j is the type of tool assigned to workstation j, S j,t is the adaptability score of tool t on workstation j,

[0123] The cost function is used to optimize the path. The path optimization formula is:

[0124]

[0125] Among them, C path is the total path cost, d p is the distance of path p, ω p is the task weight of path p, T p The scheduling module improves the capacity utilization through intelligent allocation and path optimization, while the error warning module ensures the stability of production through anomaly detection.

[0126] The steps to build an anomaly detection model based on variational autoencoder VAE and use the fused working features as input to analyze the abnormal data in the tightening process are as follows:

[0127] The potential distribution is constructed through the variational autoencoder VAE, and the construction formula is:

[0128]

[0129] Where z is the latent space representation, μ enc ,σ enc is the mean and standard deviation of the underlying distribution, ∈ is random noise, For the reconstructed features;

[0130] Specifically, through multimodal data fusion based on the attention mechanism, key features can be effectively extracted in the presence of noise or incomplete data, reducing the impact of single-modal data anomalies on prediction and deployment; at the same time, the error warning module builds an anomaly detection model based on the variational autoencoder VAE, and accurately captures anomalies in the tightening process through the dual indicators of reconstruction error and distribution deviation. Through unsupervised learning, a probability distribution model of normal operating data is established to flexibly respond to complex abnormal situations, and multi-level warnings are performed in combination with business rules, greatly reducing false alarms and missed alarms.

[0131] The steps of using the reconstruction error and distribution deviation of real-time data as abnormal indicators and combining them with business rules for multi-level warning are as follows:

[0132] The reconstruction error is calculated as follows:

[0133]

[0134] Among them, E reconis the reconstruction error, which represents the input feature F final Reconstruct output with autoencoder The difference between final is the comprehensive working feature, the input feature matrix after fusion by the attention mechanism, is the comprehensive feature reconstructed by the autoencoder, the feature matrix generated from the potential representation z by the decoder Decoder(z), ||·|| 2 is the square of the L2 norm;

[0135] Calculate the distribution deviation using the formula:

[0136] D kl =KL(q(z|F final )||p(z)),

[0137] Among them, D kl is the distribution deviation, which represents the approximate posterior distribution q(z|F final ) and the Kullback-Leibler divergence between the prior distribution p(z), q(z|F final ) is the approximate posterior distribution, which is based on the input feature F final The generated potential distribution is a Gaussian distribution: μ enc is the mean vector of the approximate posterior distribution, generated by the encoder, σ enc is the standard deviation vector of the approximate posterior distribution, generated by the encoder, p(z) is the prior distribution, which is the standard normal distribution, 0 is the mean vector of the prior distribution, I is the identity matrix, representing the independent and identically distributed standard normal distribution, and KL(·||·) is the Kullback-Leibler divergence, which is used to measure the relative entropy difference between two probability distributions and is defined as:

[0138]

[0139] By calculating the reconstruction error E recon and distribution deviation D kl , evaluate whether the input features are consistent with the normal operation of the model. Excessive deviation indicates anomaly. The anomaly indicators together form the basis of anomaly detection in the autoencoder;

[0140] Specifically, the anomaly detection and multi-level early warning mechanism based on the variational autoencoder can proactively detect potential problems before anomalies occur, thereby preventing possible production accidents.

[0141] Step S5, the feedback module feeds back the abnormal data to the capacity prediction module to update the deep learning model in step S3;

[0142] The feedback module compares the abnormal data with the prediction results to generate optimized samples, which are used to update the deep learning model. The feedback module forms a closed-loop system, enhancing the model's ability to handle complex scenarios.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. 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 deep learning-based intelligent tightening tool capacity allocation system, comprising a data acquisition module, a data upload module, a capacity prediction module, an allocation module, an error warning module, a feedback module and a user interface, characterized in that: The data acquisition module has a built-in controller, using PLC and PC-based controllers to collect working data of tightening tools, including sensor data, production history records and worker operation records; Sensor data includes torque, speed, temperature and humidity; Production history records include task sequences and tool usage times; Worker operation records include operation frequency and real-time intervention data; The data uploading module is used to upload the working data to the capacity prediction module; The capacity prediction module uses a deep learning algorithm based on work data to predict capacity and deploy production strategies; The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools at each station, as well as the tightening sequence and tightening path; The error warning module is used to analyze abnormal data during the tightening process, including torque abnormalities and speed abnormalities, and provide error warnings; The feedback module is used to feed back abnormal data to the capacity prediction module in real time; The user interface is used by an operator to monitor and adjust the status of a tightening tool.

2. The deep learning-based intelligent tightening tool capacity allocation system according to claim 1, characterized in that: In the data acquisition module, a multimodal fusion model based on the attention mechanism is adopted to dynamically adjust the weight of the work data and generate comprehensive work features. The fused work features are input into the deployment module and the error warning module.

3. The deep learning-based intelligent tightening tool capacity allocation system according to claim 2, characterized in that: In the error warning module, an anomaly detection model based on variational autoencoder VAE is constructed, with the fused working features as input; Unsupervised learning is used to establish a probability distribution model for normal operation data. During operation, the reconstruction error and distribution deviation of real-time data are used as abnormal indicators, and multi-level warnings are performed in combination with business rules.

4. A method for intelligent tightening tool capacity allocation based on deep learning, based on an intelligent tightening tool capacity allocation system based on deep learning according to any one of claims 1 to 3, characterized in that: include: Step S1, collecting working data during the tightening process through a data acquisition module, Step S2: A multimodal fusion model based on the attention mechanism is used for the collected work data to dynamically adjust the weight of the work data and generate comprehensive work features. The fused work features are input into the deployment module and the error warning module. Step S3, the capacity prediction module uses a deep learning algorithm to predict capacity based on the integrated work characteristics and deploy production strategies; Step S4: The deployment module adjusts the production capacity according to the production strategy, including the type and quantity of tools for each workstation, as well as the tightening sequence and tightening path; at the same time, the error warning module constructs an anomaly detection model based on the variational autoencoder VAE, takes the fused working features as input, and analyzes the abnormal data in the tightening process, including torque anomalies and speed anomalies; and uses unsupervised learning to establish a probability distribution model for normal operation data. During operation, the reconstruction error and distribution deviation of real-time data are used as abnormal indicators, and multi-level warnings are performed in combination with business rules; Step S5: The feedback module feeds back the abnormal data to the capacity prediction module to update the deep learning model in step S3.

5. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 4, characterized in that: The steps of using a multimodal fusion model based on an attention mechanism to dynamically adjust the weight of the work data and generate comprehensive work features are as follows: After data preprocessing, the multimodal data is standardized and preliminary features are extracted. The extraction formula is: H A =f CNN (X A ),H B =f RNN (X B ),H C =f RNN (X C ), Among them, H A is the feature matrix of sensor data, H B is the characteristic matrix of production history records, H C is the feature matrix of worker operation records, X A ,X B ,X C is the original data input of the corresponding mode, f CNN ,f RNN are convolutional neural networks and recurrent neural networks respectively. Based on the attention mechanism, the weights of each modal feature are dynamically adjusted. The calculation formula is: Among them, α m is the weight of mode m, m∈{A,B,C}, H m is the preliminary feature of mode m, W Q is the learnable parameter matrix used for weight calculation, k is the modal index of weight normalization, and its value range is {1,2,3}, After fusing the modal features by weight, the final comprehensive features are generated. The comprehensive feature formula is: Among them, F final is the comprehensive working characteristic, W F is the weight matrix of the fully connected layer, b F is the bias of the fully connected layer, and σ is the nonlinear activation function.

6. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 5, characterized in that: The steps of using a deep learning algorithm to predict production capacity based on the fused work characteristics and deploying production strategies are as follows: The capacity prediction module is modeled using a deep feedforward neural network DNN, and the calculation process is as follows: WITH (l) =σ(W (l) WITH (l-1) +b (l) ),l=1,2,...,L-1, Y pred =W (L) Z (L-1) +b (L) , Among them, Z (l) is the output feature of the lth layer, Z (0) =F final is the comprehensive working feature of the input layer, W (l) ,b (l) is the weight matrix and bias of the lth layer, Y pred is the final capacity prediction value, L is the total number of layers in the network, The mean square error MSE is used as the loss function, which is defined as L: And update the model parameters by gradient descent method, the update formula is: Among them, L is the loss value, which represents the average square error between the predicted value and the true value, N is the number of samples, and Y pred,i is the predicted value of the i-th sample, Y true,i is the true capacity value of the i-th sample, θ is the network parameter, including weight and bias, η is the learning rate, and t is the number of iterations.

7. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 6, characterized in that: The steps of the allocation module adjusting the production capacity according to the production strategy are as follows: The deployment module adjusts production capacity, including: The tools are assigned according to the task priority, and the allocation formula is: Among them, U j is the type of tool assigned to workstation j, S j,t is the adaptability score of tool t at workstation j, The cost function is used to optimize the path. The path optimization formula is: Among them, C path is the total path cost, d p is the distance of path p, ω p is the task weight of path p, T p is the execution time of path p.

8. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 7, characterized in that: The steps of constructing an anomaly detection model based on variational autoencoder VAE, taking the fused working features as input, and analyzing the abnormal data in the tightening process are as follows: The potential distribution is constructed through the variational autoencoder VAE, and the construction formula is: Where z is the latent space representation, μ enc ,σ enc is the mean and standard deviation of the underlying distribution, ∈ is random noise, The features of reconstruction.

9. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 8, characterized in that: The steps of using the reconstruction error and distribution deviation of real-time data as abnormal indicators and performing multi-level warning in combination with business rules are as follows: The reconstruction error is calculated as follows: Among them, E recon is the reconstruction error, which represents the input feature F final Reconstruct output with autoencoder The difference between final is the comprehensive working feature, the input feature matrix fused by the attention mechanism, is the comprehensive feature reconstructed by the autoencoder, the feature matrix generated from the potential representation z by the decoder Decoder(z), ||·|| 2 is the square of the L2 norm; Calculate the distribution deviation using the formula: D kl =KL(q(z|F final )||p(z)), Among them, D kl is the distribution deviation, which represents the approximate posterior distribution q(z|F final ) and the Kullback-Leibler divergence between the prior distribution p(z), q(z|F final ) is the approximate posterior distribution, which is based on the input feature F final The generated potential distribution is a Gaussian distribution: μ enc is the mean vector of the approximate posterior distribution, generated by the encoder, σ enc is the standard deviation vector of the approximate posterior distribution, generated by the encoder, p(z) is the prior distribution, which is the standard normal distribution, 0 is the mean vector of the prior distribution, I is the identity matrix, representing the independent and identically distributed standard normal distribution, and KL(·||·) is the Kullback-Leibler divergence, which is used to measure the relative entropy difference between two probability distributions and is defined as: By calculating the reconstruction error E recon and distribution deviation D kl , evaluate whether the input features are consistent with the normal operation of the model. Excessive deviation indicates anomaly. The anomaly indicators together constitute the basis for anomaly detection of the autoencoder.

10. The method for adjusting the production capacity of intelligent tightening tools based on deep learning according to claim 9, characterized in that: The feedback module compares the abnormal data with the prediction results to generate optimized samples, and the optimized samples are used to update the deep learning model.

Images