A tightening tool parameter optimization system and method based on deep learning

Abstract

The present invention discloses a tightening tool parameter optimization system and method based on deep learning, which relates to the technical field of tool parameter optimization. The present invention first uses an edge computing module to clean, standardize and extract features of real-time data, and uses an RNN model to predict the optimal tightening parameters. The process parameter optimization module dynamically adjusts parameters in combination with real-time working conditions, and can quickly adapt to complex environmental changes; secondly, the cloud uses a Transformer model to analyze the load balancing and operation sequence between multiple tools, and optimizes the task allocation strategy. At the same time, the IIoT platform realizes the sharing of optimization parameters between multiple tools, significantly improving the collaborative efficiency of the production line; in addition, the feedback module collects tightening operation performance and operating parameters, and uploads the data to the cloud, uses a deep learning model to continuously analyze the collaboration between tools and historical data, dynamically updates the optimization algorithm, and improves the system's adaptability to dynamic working conditions and continuous improvement capabilities.

Description

Technical Field

[0001] The present invention relates to the technical field of tool parameter optimization, and in particular to a system and method for optimizing tightening tool parameters based on deep learning. Background Art

[0002] In modern industrial production, tightening operations are a key step in the assembly process, and their parameter optimization has a significant impact on product quality and production efficiency. Traditional parameter optimization methods for tightening tools mostly use fixed process parameters. This method has certain applicability under standardized production conditions. However, in actual production, due to the dynamic changes in the environment and working conditions, such as differences in material properties, external vibration interference, and fluctuations in operating conditions, fixed parameters are difficult to meet actual needs and have poor adaptability.

[0003] To address this issue, some solutions pre-set process parameter combinations for different scenarios to meet production needs. However, this approach is difficult to flexibly respond to real-time changes in the production line. In multi-tool collaboration or complex assembly tasks, the lack of a unified communication and collaboration mechanism between tools leads to parameter mismatches between devices, affecting operational consistency and ultimately negatively impacting assembly quality. Furthermore, traditional tightening tools also have significant shortcomings in their collaboration with other manufacturing systems.

[0004] With the popularization of Industrial Internet of Things (IIoT), multi-device collaboration and intelligence have become industry trends. However, traditional tightening tools cannot achieve seamless integration with the Industrial Internet of Things platform and lack unified data collection and analysis methods. This limits the further improvement of tool performance and makes it difficult to meet the needs of intelligent production. Therefore, a tightening tool parameter optimization system and method based on deep learning is urgently needed to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a tightening tool parameter optimization system and method based on deep learning to solve the problem that the core logic of traditional solutions is still mainly based on the optimization of a single tool, which is difficult to meet the needs of complex production environments and has obvious deficiencies in real-time performance, multi-device collaboration, dynamic adjustment and intelligent application.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a tightening tool parameter optimization system based on deep learning, which includes:

[0009] Data acquisition module, real-time collection of tightening tool operation data and environmental parameters;

[0010] The edge computing module receives real-time data transmitted by the data acquisition module, performs preliminary data processing, uses the deep learning model RNN to quickly analyze the real-time data, predicts the optimal tightening parameters of a single tool in real time, and passes the results to the process parameter optimization module;

[0011] The analysis module uses the deep learning Transformer algorithm to predict the collaborative operation parameters of multiple tools based on the data transmitted by the edge computing module, outputs the optimization strategy, and updates the model parameters of the process parameter optimization module and the edge computing module;

[0012] The process parameter optimization module receives the optimization parameters output by the analysis module and the edge computing module, and adjusts the tightening parameters based on the real-time working conditions, including the current tool status and environmental changes;

[0013] The feedback module collects the operation performance and adjusted operating parameters in real time after the tightening operation is completed, uploads the operating parameters to the cloud, and transmits them to the analysis module through the Industrial Internet of Things (IIoT) platform;

[0014] The Industrial Internet of Things (IIoT) platform, as the hub connecting various modules, is responsible for real-time transmission and sharing of data, receiving data uploaded by the feedback module and passing it to the analysis module.

[0015] As a preferred solution of the deep learning-based tightening tool parameter optimization system and method described in the present invention, the operating data includes: tightening torque, tightening speed and tightening times; the environmental parameters include: material hardness and roughness, as well as temperature, humidity and vibration in the production environment.

[0016] As an optimal solution of the deep learning-based tightening tool parameter optimization system and method described in the present invention, the operating parameters include operating results and execution parameters; the operating results include whether the tightening quality meets the standards, and the execution parameters include torque value and tightening angle deviation.

[0017] As a preferred solution of the deep learning-based tightening tool parameter optimization system and method described in the present invention, the Industrial Internet of Things (IIoT) platform is used to optimize parameter sharing among multiple tools; adjust operation sequence, load balancing and task allocation.

[0018] In a second aspect, the present invention provides a method for optimizing tightening tool parameters based on deep learning, comprising:

[0019] Step S1: The data acquisition module collects the operating parameters of the tightening tool, including torque, speed, tightening material hardness, and temperature and humidity. The operating parameters are cleaned, standardized, and feature extracted in the edge computing module to generate real-time data;

[0020] In step S2, the edge computing module sends the real-time data to the analysis module via a network connection. The analysis module uses a deep learning RNN model to predict the optimal tightening parameters, including torque and speed. The process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and the real-time working conditions.

[0021] Step S3: The tightening tool performs a tightening operation according to the adjusted tightening parameters. The data acquisition module collects the adjusted operating parameters and operating performance in real time and uploads them to the cloud via the Industrial Internet of Things (IIoT) platform.

[0022] In step S4, the cloud receives the adjusted operating parameters and work performance through the Industrial Internet of Things (IIoT) platform, and uses the deep learning Transformer model to analyze the load balancing and work order between tightening tools; based on the analysis results, the optimization algorithm is updated and the optimized model is sent to the edge computing module.

[0023] As a preferred solution of the system and method for optimizing tightening tool parameters based on deep learning described in the present invention, the steps of performing cleaning, standardization and feature extraction in the edge computing module to generate real-time data are as follows:

[0024] Perform outlier detection based on distribution analysis to detect the collected data points x i Whether it is within a reasonable range, the monitoring formula is: i ∈[Q1-1.5·IQR,Q3+1.5·IQR], where Q1 is the first quartile of the data, Q3 is the third quartile of the data, and IQR=Q3-Q1 is the interquartile range;

[0025] For the detected outliers, linear interpolation method is used to process the outliers. The processing formula is:

[0026]

[0027] Among them, x i is an outlier, x i-1 ,x i+1 are the normal data points before and after the outlier;

[0028] For each data point, ZScore normalization is performed, and the normalization formula is:

[0029]

[0030] Among them, z i is the standardized data, x i is the original data after cleaning, μ x is the mean of the data, σ x is the standard deviation of the data;

[0031] Feature extraction is performed on the standardized data, and the following features are extracted for the torque T(t) and velocity v(t), including:

[0032] average value

[0033]

[0034] Maximum value t max :

[0035] T max =max(T i ),

[0036] RMS value T rms :

[0037]

[0038] Among them, T i is the torque value at the i-th time point, and N is the total number of sampling points;

[0039] For calculating the dynamic change rate of temperature and humidity, the calculation formula is:

[0040]

[0041] Among them, h t ,T t is the humidity and temperature at the current time point, R h ,R T is the rate of change of humidity and temperature, Δt is the sampling time interval;

[0042] Calculate the Pearson correlation coefficient between torque and speed using the following formula:

[0043]

[0044] Among them, T i ,v i are the torque and velocity at the i-th time point, is the average value of torque and velocity,

[0045] r T,v is a linear correlation.

[0046] As a preferred solution of the system and method for optimizing tightening tool parameters based on deep learning described in the present invention, the steps of: the data acquisition module collecting the adjusted operating parameters and operating performance in real time, and uploading them to the cloud through the Industrial Internet of Things (IIoT) platform are as follows:

[0047] The adjusted operating parameters collected in real time include: torque value T act,tand tightening angle θ act,t , let the collected operating parameter data set be P run , P run ={(T act,t ,θ act,t )|t=1,2,…,N}, where N is the total number of sampling time points and t is the current sampling time point;

[0048] The collected operational performance includes:

[0049] Operation effect Q tight,t : Whether the tightening quality meets the standards,

[0050] If T act,t ∈[T min ,T max ] and θ act,t ∈[θ min ,θ max ], then Q tight,t =1, otherwise Q tight,t =0; where T min ,T max is the allowable torque range, θ min ,θ max is the allowable tightening angle range,

[0051] Execution parameters, including actual torque value deviation ΔT t and angular deviation Δθ t :

[0052] ΔT t =T act,t -T * , Δθ t =θ act,t -θ * , where T * ,θ * are the target torque value and target angle;

[0053] The collected job performance data set is P perf :

[0054] P perf ={(Q tight,t ,ΔT t ,Δθ t )|t=1,2,…,N};

[0055] Integrate the operating parameters and job performance data into a data set D, D = {(P run ,P perf )}, where D is the complete data packet after a single job is completed;

[0056] The integrated data is uploaded to the cloud through the Industrial Internet of Things (IIoT) platform, and the upload frequency is controlled by the time window method. The data is uploaded to the cloud through the IIoT platform.

[0057] As a preferred solution of the system and method for optimizing tightening tool parameters based on deep learning described in the present invention, the process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and in combination with the real-time working conditions.

[0058] Assume that the real-time dataset after cleaning, standardization and feature extraction is X={x1,x2,…,x N}, where x i Represents the data vector of the i-th sampling point: x i =[T i ,v i ,h i ,R T ,R h ,r T,v ], where T i is the torque value, v i is the speed value, h i is the material hardness, R T ,R h is the rate of change of temperature and humidity, r T,v is the correlation between torque and speed;

[0059] Convert X to batch format X over a network connection batch :X batch ={X1,X2,…,X M}, where M is the number of batches, and each batch X m Contains a fixed number of time steps of data;

[0060] The deep learning RNN model analyzes the input time series data and predicts the optimal tightening parameters. The RNN receives the input batch data X batch , the characteristics of the time series are calculated by recursive propagation of the hidden state, and the calculation formula is: h t =f(W h ·h t-1 +W x ·x t +b h ), where h t is the hidden state vector at time step t, W h is the weight matrix of the hidden state, W x is the weight matrix of the input data, x t is the input data vector of the current time step, b h is the bias term, f(·) is the activation function;

[0061] The final output of RNN predicts the optimal tightening parameters through the fully connected layer. The prediction formula is:

[0062] in, is the predicted tightening parameter at time step t, W y is the weight matrix of the output layer, b y is the bias term of the output layer;

[0063] According to the tightening parameters output by the analysis module, the tightening parameters of the tool are dynamically adjusted in combination with the real-time working conditions. The adjustment parameters are corrected according to the working condition function. The correction formula is:

[0064] Among them, T * ,v * are the adjusted torque and speed, The torque and speed predicted by the analysis module, ΔT env is the current temperature deviation, Δh mat is the deviation of the current material hardness, α and β are the sensitivity coefficients of the environment and material working conditions to the torque and speed adjustment; the final adjusted optimal tightening parameters are output and transmitted to the tool execution module.

[0065] As a preferred embodiment of the system and method for optimizing tightening tool parameters based on deep learning described in the present invention, the step of using the deep learning Transformer model to analyze the load balance and operation sequence between tightening tools is as follows:

[0066] Receive the integrated operating parameters and operation performance data through the Industrial Internet of Things (IIoT) platform, and receive the uploaded data of all tightening tools from the IIoT platform. cloud :

[0067]

[0068] Among them, D cloud is all the tool data integrated in the cloud, K is the number of tightening tools, and Decode(·) is the decoding function;

[0069] Extract the operating parameters and performance data of each tool using the following formula:

[0070] X k ={(T act,t ,θ act,t ,Q tight,t ,ΔT t ,Δθ t )|t=1,2,…,N k},

[0071] Among them, X kis the data set of the kth tool, N k is the total number of sampling points of the kth tool;

[0072] Organize tool data into time series form: S k =Sequence(X k ), where S k is the time series data of the kth tool, Sequence(·) means formatting the multidimensional data by time step;

[0073] Use the Transformer model to analyze multi-tool data, explore the rules of load imbalance and job sequence optimization, and transform time series data S k Converted to embedded representation, the conversion formula is:

[0074] in, is the initial embedding matrix, Embed(·) is the feature embedding function, which maps the multi-dimensional features into a fixed-dimensional vector, and PosEnc(·) is the position encoding;

[0075] Transformer captures the correlation between tools through a multi-head attention mechanism. The capture process is expressed as:

[0076]

[0077] Where Q = Z k W Q ,K=Z k W K ,V=z k W V are query, key and value matrices, respectively, through the weight matrix W Q ,W k ,W V Projection, d k is the dimension of the key, Attention(·) is the attention score;

[0078] After L layers of Transformer encoding, the final output is obtained

[0079]

[0080] Combining the data from all tools, it is expressed as:

[0081]

[0082] Use the output representation Z to predict the load balancing status B between tools k and recommended task sequence Q k , the prediction formula is:

[0083]

[0084] Among them, B k Score the load of the kth tool, O k is the recommended order of the kth tool, MLP B ,MLP O It is a multi-layer perceptron.

[0085] As a preferred solution of the system and method for optimizing tightening tool parameters based on deep learning described in the present invention, the steps of updating the optimization algorithm based on the analysis results and sending the optimized model to the edge computing module are as follows:

[0086] According to the results of load balancing and job order analysis, the parameters and strategies of the edge computing model are updated. The optimized tightening task allocation strategy is expressed as:

[0087]

[0088] Among them, W k is the task weight of the kth tool, and the task is assigned according to W k Adjustment, B k Score the load of the kth tool;

[0089] Rearrange the order of tool operations. The sorting formula is:

[0090] O=Sort(O1,O2,…,O K ),

[0091] Among them, O is the sorted tool operation sequence,

[0092] Combined with the load optimization and order adjustment results, the model parameters of the edge computing module are updated. The update formula is:

[0093]

[0094] Among them, θ new is the updated model parameter, θ old is the old model parameter, η is the learning rate, The updated model is sent to the edge computing module as the loss function.

[0095] The beneficial effects of the present invention are:

[0096] The present invention combines deep learning technology with the Industrial Internet of Things (IIoT) platform, overcoming the problems of poor real-time performance, insufficient collaboration, and lack of data closed loop in traditional methods, and significantly improving production efficiency and tightening quality. First, the edge computing module cleans, standardizes and extracts features of real-time data, and uses the RNN model to predict the optimal tightening parameters. The process parameter optimization module dynamically adjusts parameters based on real-time working conditions, and can quickly adapt to complex environmental changes. Secondly, the cloud uses the Transformer model to analyze the load balancing and operation sequence between multiple tools and optimize the task allocation strategy. At the same time, the IIoT platform realizes the sharing of optimization parameters between multiple tools, significantly improving the collaborative efficiency of the production line. In addition, the feedback module collects tightening operation performance and operating parameters, and uploads the data to the cloud. The deep learning model is used to continuously analyze the collaboration between tools and historical data, and the optimization algorithm is dynamically updated to improve the system's adaptability to dynamic working conditions and continuous improvement capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] 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.

[0098] Figure 1 Schematic diagram of the framework of the tightening tool parameter optimization system in Example 1.

[0099] Figure 2 This is an optimization flow chart of the adaptive tightening tool process parameter optimization system in Example 1;

[0100] Figure 3 This is a workflow diagram of the process parameter optimization module in Example 1. DETAILED DESCRIPTION

[0101] 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.

[0102] 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.

[0103] 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.

[0104] Example 1, with reference to Figure 1 、 Figure 2 and Figure 3 , this embodiment provides a tightening tool parameter optimization system based on deep learning, including:

[0105] Data acquisition module, real-time collection of tightening tool operation data and environmental parameters;

[0106] Operation data include: tightening torque, tightening speed and tightening times,

[0107] Environmental parameters include: material hardness and roughness, as well as temperature, humidity and vibration in the production environment.

[0108] The edge computing module receives real-time data transmitted by the data acquisition module, performs preliminary data processing, uses the deep learning model RNN to quickly analyze the real-time data, predicts the optimal tightening parameters of a single tool in real time, and passes the results to the process parameter optimization module;

[0109] The analysis module uses the deep learning Transformer algorithm to predict the collaborative operation parameters of multiple tools based on the data transmitted by the edge computing module, outputs the optimization strategy, and updates the model parameters of the process parameter optimization module and the edge computing module;

[0110] The process parameter optimization module receives the optimization parameters output by the analysis module and the edge computing module, and adjusts the tightening parameters based on the real-time working conditions, including the current tool status and environmental changes;

[0111] The feedback module collects the operation performance and adjusted operating parameters in real time after the tightening operation is completed, uploads the operating parameters to the cloud, and transmits them to the analysis module through the Industrial Internet of Things (IIoT) platform;

[0112] Operation parameters include operation effects and execution parameters;

[0113] The operation effect includes whether the tightening quality meets the standards, and the execution parameters include torque value and tightening angle deviation;

[0114] The Industrial Internet of Things (IIoT) platform, as the hub connecting various modules, is responsible for real-time data transmission and sharing, receiving data uploaded by the feedback module and passing it to the analysis module.

[0115] The Industrial Internet of Things (IIoT) platform is used to optimize parameter sharing among multiple tools; adjust job sequences, load balancing, and task allocation.

[0116] This embodiment also provides a method for optimizing tightening tool parameters based on deep learning, including:

[0117] Step S1: The data acquisition module collects the operating parameters of the tightening tool, including torque, speed, tightening material hardness, and temperature and humidity. The operating parameters are cleaned, standardized, and feature extracted in the edge computing module to generate real-time data;

[0118] The steps for cleaning, standardizing and feature extraction in the edge computing module to generate real-time data are as follows:

[0119] Perform outlier detection based on distribution analysis to detect the collected data points x i Whether it is within a reasonable range, the monitoring formula is: i ∈[Q1-1.5·IQR,Q3+1.5·IQR], where Q1 is the first quartile of the data, Q3 is the third quartile of the data, and IQR=Q3-Q1 is the interquartile range;

[0120] For the detected outliers, linear interpolation method is used to process the outliers. The processing formula is:

[0121]

[0122] Among them, x i is an outlier, x i-1 ,x i+1 are the normal data points before and after the outlier;

[0123] For each data point, ZScore normalization is performed, and the normalization formula is:

[0124]

[0125] Among them, z i is the standardized data, x i is the original data after cleaning, μ x is the mean of the data, σ x is the standard deviation of the data;

[0126] Feature extraction is performed on the standardized data, and the following features are extracted for the torque T(t) and velocity v(t), including:

[0127] average value

[0128]

[0129] Maximum value t max:

[0130] T max =max(T i ),

[0131] RMS value T rms :

[0132]

[0133] Among them, T i is the torque value at the i-th time point, and N is the total number of sampling points;

[0134] For calculating the dynamic change rate of temperature and humidity, the calculation formula is:

[0135]

[0136] Among them, h t ,T t is the humidity and temperature at the current time point, R h ,R T is the rate of change of humidity and temperature, Δt is the sampling time interval;

[0137] Calculate the Pearson correlation coefficient between torque and speed using the following formula:

[0138]

[0139] Among them, T i ,v i are the torque and velocity at the i-th time point, is the average value of torque and velocity, r T,v is a linear correlation,

[0140] Specifically, the cleaned data were standardized by ZScore to eliminate unit effects, and time series characteristics, dynamic environmental change rates, and correlations between variables were extracted.

[0141] In step S2, the edge computing module sends the real-time data to the analysis module via a network connection. The analysis module uses a deep learning RNN model to predict the optimal tightening parameters, including torque and speed. The process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and the real-time working conditions.

[0142] The process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and the real-time working conditions.

[0143] Assume that the real-time dataset after cleaning, standardization and feature extraction is X={x1,x2,…,x N}, where x i Represents the data vector of the i-th sampling point: xi =[T i ,v i ,h i ,R T ,R h ,r T,v ], where T i is the torque value, v i is the speed value, h i is the material hardness, R T ,R h is the rate of change of temperature and humidity, r T,v is the correlation between torque and speed;

[0144] Convert X to batch format X over a network connection batch :X batch ={X1,X2,…,X M}, where M is the number of batches, and each batch X m Contains a fixed number of time steps of data;

[0145] The deep learning RNN model analyzes the input time series data and predicts the optimal tightening parameters. The RNN receives the input batch data X batch , the characteristics of the time series are calculated by recursive propagation of the hidden state, and the calculation formula is: h t =f(W h ·h t-1 +W x ·x t +b h ), where h t is the hidden state vector at time step t, W h is the weight matrix of the hidden state, W x is the weight matrix of the input data, x t is the input data vector of the current time step, b h is the bias term, f(·) is the activation function;

[0146] The final output of RNN predicts the optimal tightening parameters through the fully connected layer. The prediction formula is:

[0147] in, is the predicted tightening parameter at time step t, W y is the weight matrix of the output layer, b y is the bias term of the output layer;

[0148] According to the tightening parameters output by the analysis module, the tightening parameters of the tool are dynamically adjusted in combination with the real-time working conditions. The adjustment parameters are corrected according to the working condition function. The correction formula is:

[0149] Among them, T* ,v * are the adjusted torque and speed, The torque and speed predicted by the analysis module, ΔT env is the current temperature deviation, Δh mat is the deviation of the current material hardness, α and β are the sensitivity coefficients of the environment and material working conditions to the torque and speed adjustment; the final adjusted optimal tightening parameters are output and transmitted to the tool execution module;

[0150] Specifically, data is sent to the analysis module through the edge computing module, and the time series feature extraction capability of the RNN model is used to predict the optimal tightening parameters. The output tightening parameters are dynamically adjusted in combination with the real-time working conditions, which can not only meet the real-time requirements, but also dynamically adapt to complex working conditions.

[0151] Step S3: The tightening tool performs a tightening operation according to the adjusted tightening parameters. The data acquisition module collects the adjusted operating parameters and operating performance in real time and uploads them to the cloud via the Industrial Internet of Things (IIoT) platform.

[0152] The data acquisition module collects the adjusted operating parameters and operating performance in real time and uploads them to the cloud through the Industrial Internet of Things (IIoT) platform.

[0153] The adjusted operating parameters collected in real time include: torque value T act,t and tightening angle θ act,t , let the collected operating parameter data set be P run , P run ={(T act,t ,θ act,t )|t=1,2,…,N}, where N is the total number of sampling time points and t is the current sampling time point;

[0154] The collected operational performance includes:

[0155] Operation effect Q tight,t : Whether the tightening quality meets the standards,

[0156] If T act,t ∈[T min ,T max ] and θ act,t ∈[θ min ,θ max ], then Q tight,t =1, otherwise Q tight,t =0; where T min ,T max is the allowable torque range, θ min ,θ max is the allowable tightening angle range,

[0157] Execution parameters, including actual torque value deviation ΔT t and angular deviation Δθ t :

[0158] ΔT t =T act,t -T * , Δθ t =θ act,t -θ * , where T * ,θ * are the target torque value and target angle;

[0159] The collected job performance data set is P perf :

[0160] P perf ={(Q tight,t ,ΔT t ,Δθ t )|t=1,2,…,N};

[0161] Integrate the operating parameters and job performance data into a data set D, D = {(P run ,P perf )}, where D is the complete data packet after a single job is completed;

[0162] Upload the integrated data to the cloud through the Industrial Internet of Things (IIoT) platform, use the time window method to control the upload frequency, and upload the data to the cloud through the IIoT platform;

[0163] Specifically, the real-time operating parameters and operation performance data obtained through the data acquisition module are integrated and encoded, and then uploaded to the cloud according to the time window using the Industrial Internet of Things (IIoT) platform to ensure data integrity and timeliness.

[0164] In step S4, the cloud receives the adjusted operating parameters and working performance through the Industrial Internet of Things (IIoT) platform, uses the deep learning Transformer model to analyze the load balancing and operation sequence between tightening tools, updates the optimization algorithm based on the analysis results, and sends the optimized model to the edge computing module;

[0165] The steps for using the deep learning Transformer model to analyze the load balance and operation sequence between tightening tools are as follows:

[0166] Receive the integrated operating parameters and operation performance data through the Industrial Internet of Things (IIoT) platform, and receive the uploaded data of all tightening tools from the IIoT platform. cloud :

[0167]

[0168] Among them, D cloud is all the tool data integrated in the cloud, K is the number of tightening tools, and Decode(·) is the decoding function;

[0169] Extract the operating parameters and performance data of each tool using the following formula:

[0170] X k ={(T act,t ,θ act,t ,Q tight,t ,ΔT t ,Δθ t )|t=1,2,…,N k},

[0171] Among them, X k is the data set of the kth tool, N k is the total number of sampling points of the kth tool;

[0172] Organize tool data into time series form: S k =Sequence(X k ), where S k is the time series data of the kth tool, Sequence(·) means formatting the multidimensional data by time step;

[0173] Use the Transformer model to analyze multi-tool data, explore the rules of load imbalance and job sequence optimization, and transform time series data S k Converted to embedded representation, the conversion formula is:

[0174] in, is the initial embedding matrix, Embed(·) is the feature embedding function, which maps the multi-dimensional features into a fixed-dimensional vector, and PosEnc(·) is the position encoding;

[0175] Transformer captures the correlation between tools through a multi-head attention mechanism. The capture process is expressed as:

[0176]

[0177] Where Q = Z k W Q ,K=Z k W k ,V=Z k W V are query, key and value matrices, respectively, through the weight matrix W Q ,W K ,W V Projection, d kis the dimension of the key, Attention(·) is the attention score;

[0178] After L layers of Transformer encoding, the final output is obtained

[0179]

[0180] Combining the data from all tools, it is expressed as:

[0181]

[0182] Use the output representation Z to predict the load balancing status B between tools k and recommended order of operations k , the prediction formula is:

[0183]

[0184] Among them, B k Score the load of the kth tool, O k is the recommended order of the kth tool, MLP B ,MLP O is a multi-layer perceptron;

[0185] Based on the analysis results, the optimization algorithm is updated and the optimized model is sent to the edge computing module.

[0186] According to the results of load balancing and job order analysis, the parameters and strategies of the edge computing model are updated. The optimized tightening task allocation strategy is expressed as:

[0187]

[0188] Among them, W k is the task weight of the kth tool, and the task is assigned according to W k Adjustment, B k Score the load of the kth tool;

[0189] Rearrange the order of tool operations. The sorting formula is:

[0190] O=Sort(O1,O2,…,O K ),

[0191] Among them, O is the sorted tool operation sequence,

[0192] Combined with the load optimization and order adjustment results, the model parameters of the edge computing module are updated. The update formula is:

[0193]

[0194] Among them, θ new is the updated model parameter, θ old is the old model parameter, η is the learning rate, The updated model is sent to the edge computing module as the loss function;

[0195] Specifically, by using the Transformer model to analyze the load and operation sequence between tools, the task allocation and model parameters are dynamically adjusted based on the optimization results, and then sent to the edge computing module through the IIoT platform to ensure the efficiency and coordination of the entire production line.

[0196] 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 tool parameter optimization system based on deep learning, characterized by: include, Data acquisition module, real-time collection of tightening tool operation data and environmental parameters; The edge computing module receives real-time data transmitted by the data acquisition module, performs preliminary data processing, uses the deep learning model RNN to quickly analyze the real-time data, predicts the optimal tightening parameters of a single tool in real time, and passes the results to the process parameter optimization module; The analysis module uses the deep learning Transformer algorithm to predict the collaborative operation parameters of multiple tools based on the data transmitted by the edge computing module, outputs the optimization strategy, and updates the model parameters of the process parameter optimization module and the edge computing module; The process parameter optimization module receives the optimization parameters output by the analysis module and the edge computing module, and adjusts the tightening parameters based on the real-time working conditions, including the current tool status and environmental changes; The feedback module collects the operation performance and adjusted operating parameters in real time after the tightening operation is completed, uploads the operating parameters to the cloud, and transmits them to the analysis module through the Industrial Internet of Things (IIoT) platform; The Industrial Internet of Things (IIoT) platform, as the hub connecting various modules, is responsible for real-time data transmission and sharing, receiving data uploaded by the feedback module and passing it to the analysis module; The Industrial Internet of Things (IIoT) platform is used to optimize parameter sharing among multiple tools; adjust operation sequence, load balancing and task allocation.

2. The deep learning-based tightening tool parameter optimization system according to claim 1, characterized in that: The operating data includes: tightening torque, tightening speed and tightening times; the environmental parameters include: material hardness and roughness, as well as temperature, humidity and vibration in the production environment.

3. The deep learning-based tightening tool parameter optimization system according to claim 2, characterized in that: The operating parameters include operating results and execution parameters; the operating results include whether the tightening quality meets the standards, and the execution parameters include torque value and tightening angle deviation.

4. A method for optimizing tightening tool parameters based on deep learning, based on a system for optimizing tightening tool parameters based on deep learning according to any one of claims 1 to 3, characterized in that: include: Step S1: The data acquisition module collects the operating parameters of the tightening tool, including torque, speed, tightening material hardness, and temperature and humidity. The operating parameters are cleaned, standardized, and feature extracted in the edge computing module to generate real-time data; In step S2, the edge computing module sends the real-time data to the analysis module via a network connection. The analysis module uses a deep learning RNN model to predict the optimal tightening parameters, including torque and speed. The process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and the real-time working conditions. Step S3: The tightening tool performs a tightening operation according to the adjusted tightening parameters. The data acquisition module collects the adjusted operating parameters and operating performance in real time and uploads them to the cloud via the Industrial Internet of Things (IIoT) platform. In step S4, the cloud receives the adjusted operating parameters and work performance through the Industrial Internet of Things (IIoT) platform, and uses the deep learning Transformer model to analyze the load balancing and work order between tightening tools; based on the analysis results, the optimization algorithm is updated and the optimized model is sent to the edge computing module.

5. The method for optimizing tightening tool parameters based on deep learning according to claim 4, wherein: The steps of performing cleaning, standardization and feature extraction in the edge computing module to generate real-time data are as follows: Perform outlier detection based on distribution analysis to detect the collected data points x i Whether it is within a reasonable range, the monitoring formula is: i ∈[Q1-1.5·IQR,Q3+1.5IQR], where Q1 is the first quartile of the data, Q3 is the third quartile of the data, and IQR=Q3-Q1 is the interquartile range; For the detected outliers, linear interpolation method is used to process the outliers. The processing formula is: Among them, x i is an outlier, x i-1 ,x i+1 are the normal data points before and after the outlier; For each data point, ZScore normalization is performed, and the normalization formula is: Among them, z i is the standardized data, x i is the original data after cleaning, μ x is the mean of the data, σ x is the standard deviation of the data; Feature extraction is performed on the standardized data, and the following features are extracted for the torque T(t) and velocity v(t), including: average value Maximum value T max : T max =max(T i ), RMS value T rms : Among them, T i is the torque value at the i-th time point, and N is the total number of sampling points; For calculating the dynamic change rate of temperature and humidity, the calculation formula is: Among them, h t ,T t is the humidity and temperature at the current time point, R h ,R T is the rate of change of humidity and temperature, Δt is the sampling time interval; Calculate the Pearson correlation coefficient between torque and speed using the following formula: Among them, T i ,v i are the torque and velocity at the i-th time point, is the average value of torque and velocity, r T,v is a linear correlation.

6. The method for optimizing tightening tool parameters based on deep learning according to claim 5, wherein: The data acquisition module collects the adjusted operating parameters and operating performance in real time and uploads them to the cloud through the Industrial Internet of Things (IIoT) platform. The adjusted operating parameters collected in real time include: torque value T act,t and tightening angle θ act,t , let the collected operating parameter data set be P run , P run ={(T act,t ,θ act,t )|t=1,2,…,N}, where N is the total number of sampling time points and t is the current sampling time point; The collected operational performance includes: Operation effect Q tight,t : Whether the tightening quality meets the standards, If T act,t ∈[T min ,T max ] and θ act,t ∈[θ min ,θ max ], then Q tight,t =1, otherwise Q tight,t =0; where T min ,T max is the allowable torque range, θ min ,θ max is the allowable tightening angle range, Execution parameters, including actual torque value deviation ΔT t and angular deviation Δθ t : ΔT t =T act,t -T * , Δθ t =θ act,t -θ * , where T * ,θ * are the target torque value and target angle; The collected job performance data set is P perf : P perf ={(Q tight,t ,ΔT t ,Δθ t )∣t=1,2,…,N}; Integrate the operating parameters and job performance data into a data set D, D = {(P run ,P perf )}, where D is the complete data packet after a single job is completed; The integrated data is uploaded to the cloud through the Industrial Internet of Things (IIoT) platform, and the upload frequency is controlled by the time window method. The data is uploaded to the cloud through the IIoT platform.

7. The method for optimizing tightening tool parameters based on deep learning according to claim 6, wherein: The process parameter optimization module automatically adjusts the tightening parameters based on the optimal tightening parameters and in combination with the real-time working conditions. Assume that the real-time dataset after cleaning, standardization and feature extraction is X={x1,x2,…,x N }, where x i Represents the data vector of the i-th sampling point: x i =[T i ,v i ,h i ,R T ,R h ,r T,v ], where T i is the torque value, v i is the speed value, h i is the material hardness, R T ,R h is the rate of change of temperature and humidity, r T,v is the correlation between torque and speed; Convert X to batch format X over a network connection batch :X batch ={X1,X2,…,X M }, where M is the number of batches, and each batch X m Contains a fixed number of time steps of data; The deep learning RNN model analyzes the input time series data and predicts the optimal tightening parameters. The RNN receives the input batch data X batch , the characteristics of the time series are calculated by recursive propagation of the hidden state, and the calculation formula is: h t =f(W h ·h t-1 +W x ·x t +b h ), where h t is the hidden state vector at time step t, W h is the weight matrix of the hidden state, W x is the weight matrix of the input data, x t is the input data vector of the current time step, b h is the bias term, f(·) is the activation function; The final output of RNN predicts the optimal tightening parameters through the fully connected layer. The prediction formula is: in, is the predicted tightening parameter at time step t, W y is the weight matrix of the output layer, b y is the bias term of the output layer; According to the tightening parameters output by the analysis module, the tightening parameters of the tool are dynamically adjusted in combination with the real-time working conditions. The adjustment parameters are corrected according to the working condition function. The correction formula is: Among them, T * ,v * are the adjusted torque and speed, The torque and speed predicted by the analysis module, ΔT env is the current temperature deviation, Δh mat is the deviation of the current material hardness, α and β are the sensitivity coefficients of the environment and material working conditions to the torque and speed adjustment; the final adjusted optimal tightening parameters are output and transmitted to the tool execution module.

8. The method for optimizing tightening tool parameters based on deep learning according to claim 7, wherein: The steps of using the deep learning Transformer model to analyze the load balance and operation sequence between tightening tools are as follows: Receive the integrated operating parameters and operation performance data through the Industrial Internet of Things (IIoT) platform, and receive the uploaded data of all tightening tools from the IIoT platform. cloud : Among them, D cloud is all the tool data integrated in the cloud, K is the number of tightening tools, and Decode(·) is the decoding function; Extract the operating parameters and performance data of each tool using the following formula: X k ={(T act,t ,i act,t ,Q tight,t ,ΔT t ,Dth t )∣t=1,2,…,N k }, Among them, X k is the data set of the kth tool, N k is the total number of sampling points of the kth tool; Organize tool data into time series form: S k =Sequence(X k ), where S k is the time series data of the kth tool, Sequence(·) means formatting the multidimensional data by time step; Use the Transformer model to analyze multi-tool data, explore the rules of load imbalance and job sequence optimization, and transform time series data S k Converted to embedded representation, the conversion formula is: in, is the initial embedding matrix, Embed(·) is the feature embedding function, which maps the multi-dimensional features into a fixed-dimensional vector, and PosEnc(·) is the position encoding; Transformer captures the correlation between tools through a multi-head attention mechanism. The capture process is expressed as: Where Q = Z k W Q ,K=Z k W K ,V=Z k W V are query, key and value matrices, respectively, through the weight matrix W Q ,W K ,W V Projection, d k is the dimension of the key, Attention(·) is the attention score; After L layers of Transformer encoding, the final output is obtained Combining the data from all tools, it is expressed as: Use the output representation Z to predict the load balancing status B between tools k and recommended order of operations k , the prediction formula is: Among them, B k Score the load of the kth tool, O k is the recommended order of the kth tool, MLP B ,MLP O It is a multi-layer perceptron.

9. The method for optimizing tightening tool parameters based on deep learning according to claim 8, wherein: The steps of updating the optimization algorithm based on the analysis results and sending the optimized model to the edge computing module are: According to the results of load balancing and job order analysis, the parameters and strategies of the edge computing model are updated. The optimized tightening task allocation strategy is expressed as: Among them, W k is the task weight of the kth tool, and the task is assigned according to W k Adjustment, B k Score the load of the kth tool; Rearrange the order of tool operations. The sorting formula is: O=Sort(O1,O2,…,O K ), Among them, O is the sorted tool operation sequence, Combined with the load optimization and order adjustment results, the model parameters of the edge computing module are updated. The update formula is: Among them, θ new is the updated model parameter, θ old is the old model parameter, η is the learning rate, The updated model is sent to the edge computing module as the loss function.

Images