An intelligent stamping control method, system, medium and computer device of a high-precision numerical control punch press

Abstract

The present application relates to the technical field of numerical control punch press, and particularly relates to an intelligent stamping control method, system, medium and computer equipment of a high-precision numerical control punch press.The present application quantitatively analyzes the complex coupling relationship between parameters by acquiring initial parameters and constructing a coupling model, accurately captures the interaction between each parameter, and dynamically adjusts the weight value to realize multi-parameter collaborative optimization, ensuring the globality and stability of the optimization process, and effectively improves the optimization efficiency and accuracy by solving the collaborative optimization objective function using an improved particle swarm optimization algorithm, and verifies the optimization scheme through virtual stamping test and real-time data acquisition, ensuring that the optimization scheme can maintain a low parameter fluctuation index in the actual stamping process, and the present application effectively solves the problem that single parameter adjustment in the prior art cannot fully consider the mutual influence between parameters, resulting in difficulty in achieving a global optimal solution through the construction and dynamic optimization of a multi-parameter coupling model.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of numerical control punch presses, in particular to an intelligent stamping control method, system, medium and computer equipment of a high-precision numerical control punch press. BACKGROUND

[0002] The stamping of the high-precision numerical control punch press refers to a punch press controlled by a numerical control system, which is used for punching, cutting, forming and other machining operations on materials such as metals and plastics, and has high precision and automation level, and can accurately control the speed, pressure, position and other parameters in the stamping process, so as to realize high-precision and high-efficiency production.

[0003] At present, the intelligent stamping control method of the high-precision numerical control punch press in the prior art mainly adjusts the stamping parameters dynamically according to the real-time machining state through the self-adaptive parameter adjustment control method to avoid machining defects, but in the stamping process, multiple parameters (punch speed, pressure, stroke depth, plate positioning accuracy, etc.) interact with each other, and the adjustment of a single parameter may trigger a chain reaction, which makes it difficult for the system to find a global optimal solution, and thus reduces the machining stability. SUMMARY

[0004] The main purpose of the application is to provide an intelligent stamping control method of a high-precision numerical control punch press, which aims to solve the technical problems in the prior art.

[0005] The application provides an intelligent stamping control method of a high-precision numerical control punch press, which comprises the following steps:

[0006] Obtaining multiple initial parameter values of the high-precision numerical control punch press under a preset stamping working condition, and constructing a multi-parameter coupling model according to each initial parameter value;

[0007] Quantitative analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model to obtain a parameter coupling coefficient matrix, and setting a dynamic weight value for each initial parameter value according to the parameter coupling coefficient matrix;

[0008] Constructing a multi-parameter collaborative optimization objective function according to multiple dynamic weight values, and solving the multi-parameter collaborative optimization objective function by using an improved particle swarm optimization algorithm to obtain an initial parameter optimization scheme;

[0009] Inputting the initial parameter optimization scheme into a numerical control punch press simulation system to perform a virtual stamping test and collecting multiple parameter real-time data in the virtual stamping process in real time;

[0010] Obtaining a parameter fluctuation index according to multiple parameter real-time data, and judging whether the parameter fluctuation index is less than a preset fluctuation threshold;

[0011] if the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is taken as a candidate optimization scheme;

[0012] if the parameter fluctuation index is not less than the preset fluctuation threshold, the multi-parameter coupling model is corrected according to the parameter real-time data, and the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model is returned to until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is taken as a candidate optimization scheme.

[0013] Preferably, the step of constructing a multi-parameter coupling model according to each initial parameter value comprises:

[0014] a parameter sample set is constructed according to a plurality of initial parameter values, wherein the parameter sample set comprises a plurality of sample points;

[0015] principal component analysis is used to reduce the dimension of the parameter sample set and extract a plurality of main components, and a multi-parameter coupling initial model is constructed according to the plurality of main components;

[0016] the parameter sample set is divided into training parameter samples and test parameter samples;

[0017] the multi-parameter coupling initial model is trained and tested according to the training parameter samples and the test parameter samples until the mean square error of the multi-parameter coupling initial model is less than a preset error threshold, and the multi-parameter coupling initial model is obtained.

[0018] Preferably, the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model to obtain a parameter coupling coefficient matrix comprises:

[0019] a deep belief network is constructed according to the initial parameter values of the multi-parameter coupling model, and the normalized initial parameter values in the multi-parameter coupling model are input into the visible layer of the deep belief network;

[0020] the hidden layer of the deep belief network is trained layer by layer through a greedy algorithm, and the activation probability of the hidden layer is obtained according to the data of the visible layer;

[0021] the reconstructed visible layer data is obtained according to the activation probability of the hidden layer, and the network weight and bias are updated according to the difference between the reconstructed visible layer data and the original visible layer data;

[0022] the weight matrix of each layer of the deep belief network after training is extracted, and the parameter coupling coefficient is obtained according to each weight matrix;

[0023] the plurality of parameter coupling coefficients are arranged in parameter order to construct a parameter coupling coefficient matrix.

[0024] Preferably, the step of solving the multi-parameter collaborative optimization objective function to obtain an initial parameter optimization solution includes:

[0025] Inputting each of the initial parameter values ​​into the multi-parameter collaborative optimization objective function to obtain multiple fitness values, and selecting the initial parameter value corresponding to the minimum fitness value from the multiple fitness values ​​as the individual optimal position;

[0026] Selecting an initial parameter value corresponding to the minimum fitness value from the multiple individual optimal positions as the global optimal position, and obtaining a global optimal position deviation based on the global optimal position and the individual optimal position;

[0027] Obtaining a historical optimal position of the initial parameter value corresponding to each of the individual optimal positions, and obtaining an individual optimal position deviation based on the historical optimal position and the individual optimal position;

[0028] Obtaining a current speed of the initial parameter value corresponding to each of the individual optimal positions, and obtaining an update speed of the corresponding initial parameter value based on the current speed, the individual optimal position deviation, and the global optimal position deviation;

[0029] An updated position is obtained according to the update speed and the individual optimal position, and an initial parameter optimization solution is obtained according to the update position and the update speed.

[0030] Preferably, the step of obtaining a parameter fluctuation index based on the real-time data of the plurality of parameters comprises:

[0031] Preprocessing the real-time data of each parameter to obtain corresponding standard parameter data;

[0032] Obtaining multiple initial parameter values ​​in the initial parameter optimization scheme, and obtaining corresponding parameter deviation values ​​according to each of the initial parameter values ​​and standard parameter data;

[0033] A parameter deviation square sum is obtained according to the plurality of parameter deviation values, and a parameter fluctuation index is obtained according to the parameter deviation square sum.

[0034] Preferably, the step of correcting the multi-parameter coupling model according to the real-time parameter data includes:

[0035] Inputting each of the initial parameter values ​​into a multi-parameter coupling model to obtain first predicted parameter data, and obtaining corresponding parameter residuals based on the first predicted parameter data and parameter real-time data;

[0036] The parameter residuals are fitted using the least squares method to obtain a residual model, and the residual model is added to the multi-parameter coupling initial model to obtain a revised multi-parameter coupling model;

[0037] inputting each of the initial parameter values into the corrected multi-parameter coupling model to obtain second predicted parameter data;

[0038] obtaining actual parameter values of each of the initial parameter values under a preset stamping working condition, and obtaining corresponding absolute deviations according to each of the actual parameter values and the second predicted parameter data;

[0039] obtaining a root mean square error according to a plurality of the absolute deviations, and determining whether the root mean square error is greater than a preset error threshold;

[0040] if the root mean square error is greater than the preset error threshold, determining that the corrected multi-parameter coupling model fails the verification, and returning to the step of fitting the parameter residuals by the least square method to obtain the residual model until the root mean square error is not greater than the preset error threshold;

[0041] if the root mean square error is not greater than the preset error threshold, determining that the corrected multi-parameter coupling model passes the verification.

[0042] The application also provides an intelligent stamping control system of a high-precision numerical control punch press, comprising:

[0043] a construction module configured to obtain a plurality of initial parameter values of the high-precision numerical control punch press under a preset stamping working condition, and construct a multi-parameter coupling model according to each of the initial parameter values;

[0044] a quantitative analysis module configured to quantitatively analyze a coupling relationship between the initial parameters according to the multi-parameter coupling model, obtain a parameter coupling coefficient matrix, and set a dynamic weight value for each of the initial parameter values according to the parameter coupling coefficient matrix;

[0045] a solving module configured to construct a multi-parameter collaborative optimization objective function according to a plurality of the dynamic weight values, and solve the multi-parameter collaborative optimization objective function by an improved particle swarm optimization algorithm to obtain an initial parameter optimization scheme;

[0046] a test module configured to input the initial parameter optimization scheme into a numerical control punch press simulation system to perform a virtual stamping test and collect a plurality of parameter real-time data in a virtual stamping process in real time;

[0047] a determination module configured to obtain a parameter fluctuation index according to a plurality of the parameter real-time data, and determine whether the parameter fluctuation index is less than a preset fluctuation threshold;

[0048] if the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is taken as a candidate optimization scheme;

[0049] If the parameter fluctuation index is not less than the preset fluctuation threshold, the multi-parameter coupling model is corrected according to the parameter real-time data, and the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model is returned to until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is taken as a candidate optimization scheme.

[0050] As preferred, the construction module comprises:

[0051] The first construction unit is configured to construct a parameter sample set according to the plurality of initial parameter values, wherein the parameter sample set comprises a plurality of sample points.

[0052] The second construction unit is configured to perform dimension reduction processing on the parameter sample set by using a principal component analysis method and extract a plurality of principal components, and construct a multi-parameter coupling initial model according to the plurality of principal components.

[0053] The division unit is configured to divide the parameter sample set into training parameter samples and test parameter samples.

[0054] The training unit is configured to train and test the multi-parameter coupling initial model according to the training parameter samples and test parameter samples until the mean square error of the multi-parameter coupling initial model is less than a preset error threshold, so as to obtain the multi-parameter coupling initial model.

[0055] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the intelligent stamping control method of the high-precision numerical control punch when executing the computer program.

[0056] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the intelligent stamping control method of the high-precision numerical control punch when executed by a processor.

[0057] The beneficial effects of the present application are: the present application quantitatively analyzes the complex coupling relationship between parameters by acquiring initial parameters and constructing a coupling model, accurately captures the interaction between each parameter, and dynamically adjusts the weight value to realize multi-parameter collaborative optimization, avoids the chain reaction that may be caused by relying only on single parameter adjustment in the traditional technology, ensures the globality and stability of the optimization process, effectively improves the optimization efficiency and accuracy by solving the collaborative optimization objective function by using the improved particle swarm optimization algorithm, and verifies the optimization scheme through virtual punching test and real-time data acquisition, ensures that the optimization scheme can maintain a low parameter fluctuation index in the actual stamping process, significantly improves the stability and accuracy of the stamping process, and the present application solves the problem that the mutual influence between parameters cannot be fully considered by single parameter adjustment in the prior art, and the global optimal solution is difficult to achieve. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 The method flowchart of an embodiment of the present application.

[0059] Figure 2 The system structure schematic diagram of an embodiment of the present application.

[0060] Figure 3 The internal structure schematic diagram of a computer device of an embodiment of the present application.

[0061] The implementation of the present application, functional characteristics and advantages will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0062] It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0063] As shown in the figure, the present application provides an intelligent stamping control method of a high-precision numerical control punch press, comprising: Figures 1-3

[0064] S1, acquiring a plurality of initial parameter values of a high-precision numerical control punch press under a preset stamping working condition, and constructing a multi-parameter coupling model according to each initial parameter value;

[0065] S2, quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model, obtaining a parameter coupling coefficient matrix, and setting a dynamic weight value for each initial parameter value according to the parameter coupling coefficient matrix;

[0066] S3, constructing a multi-parameter collaborative optimization objective function according to a plurality of dynamic weight values, and solving the multi-parameter collaborative optimization objective function by using an improved particle swarm optimization algorithm to obtain an initial parameter optimization scheme;

[0067] ​S4, inputting the initial parameter optimization scheme into a numerical control punch simulation system to perform a virtual punching test and collecting multiple parameter real-time data in a virtual punching process in real time;

[0068] S5, obtaining a parameter fluctuation index according to the multiple parameter real-time data, and judging whether the parameter fluctuation index is less than a preset fluctuation threshold;

[0069] If the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is taken as a candidate optimization scheme.

[0070] If the parameter fluctuation index is not less than the preset fluctuation threshold, the multiple-parameter coupling model is corrected according to the parameter real-time data, and the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multiple-parameter coupling model is returned to until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is taken as the candidate optimization scheme.

[0071] As described in the above steps S1-S5, the multiple initial parameter values refer to a collective group of multiple different parameter values, each of which includes punch initial speed, initial pressure, initial stroke depth, and initial positioning accuracy of the sheet metal, and the like, which are punching core parameters of a high-precision numerical control punch. The setting of dynamic weight values for each initial parameter value according to the parameter coupling coefficient matrix is mainly based on the elements (representing the degree of mutual influence between two parameters) in the parameter coupling coefficient matrix, which quantifies the influence weight of each initial parameter in the overall coupling relationship. The greater the absolute value of the coupling coefficient, the more significant the influence of the parameter on other parameters, and the higher the initial weight value. The adaptive weight distribution algorithm dynamically updates the weight according to the real-time change of the parameter coupling coefficient. When the coupling coefficient sum of a certain parameter and other parameters increases, its weight value automatically increases (such as from 0.3 to 0.5). When the coupling coefficient sum decreases, the weight value decreases accordingly, ensuring that the weight distribution always reflects the dominant position of the parameter in the current coupling relationship.

[0072] The step of constructing a multi-parameter collaborative optimization objective function according to the multiple dynamic weight values includes determining weight coefficients of punching precision, punching efficiency, and energy consumption, and the sum of the weight coefficients is 1. The weight coefficients of each index are determined by the analytic hierarchy process. The steps of the analytic hierarchy process are to construct a judgment matrix, the elements in the judgment matrix represent the relative importance between indexes, and the 1-9 scale method is used to assign values to the relative importance. The maximum eigenvalue of the judgment matrix and the corresponding eigenvector are calculated, the eigenvector is normalized to obtain the weight coefficients of each index, and consistency test is performed. When the consistency ratio is less than 0.1, the judgment matrix has consistency, otherwise the judgment matrix is reconstructed. ; wherein, represents the number of parameters affecting the punching precision, represents the a dynamic weight value of the parameter, representing an actual value of the parameter, representing a theoretical value of the parameter, representing a parameter serial number affecting the stamping precision, a stamping efficiency sub-objective function (a negative number of a ratio of the number of stamping completed in a unit time to a unit time, the negative number indicating that the index is a maximization target, the higher the efficiency, the smaller the function value, consistent with the overall optimization target) and an energy consumption sub-objective function (a ratio of the power consumption in the stamping process in a unit time to a unit time), and a multi-parameter collaborative optimization objective function is constructed according to a weighted sum of the stamping precision sub-objective function, the stamping efficiency sub-objective function and the energy consumption sub-objective function, so as to realize collaborative optimization of the stamping precision, the efficiency and the energy consumption, avoid a chain reaction caused by single parameter adjustment, and guide to a global optimal solution.

[0073] The present application can truly reflect a plurality of physical parameters in the stamping process under actual working conditions by obtaining a plurality of initial parameter values of the high-precision numerical control punch press under the preset stamping working condition and constructing a multi-parameter coupled model according to each initial parameter value, which not only improves the representativeness of the data, but also ensures that the data source of the model training is close to the actual operation, avoids the error of pure theory or hypothetical data, and the prior art often relies on a too simplified model or a single parameter to adjust the equipment, while the present application comprehensively collects multi-dimensional data, covers the complex relationship of a plurality of parameter changes, provides a more accurate and comprehensive basis for the subsequent optimization model, and fully considers the interaction between various stamping parameters by establishing a multi-parameter coupled model, avoids the problem that the traditional method only focuses on single parameter optimization, and comprehensively analyzes the mutual influence between different parameters through the coupled model, so that the optimization process is closer to the actual situation, the prior art usually adopts a single optimization parameter, and ignores the complex coupling relationship between the multi-parameters, while the present application can realize more efficient optimization through the multi-parameter coupled model, and avoid performance degradation or instability problems caused by the coupling effect between parameters.

[0074] The parameter coupling coefficient matrix is obtained by quantitatively analyzing the coupling relationship between the initial parameters through a multi-parameter coupling model, and a dynamic weight value is set for each initial parameter value according to the parameter coupling coefficient matrix. By quantifying the coupling relationship between the parameters, the contribution of each parameter to the overall performance can be clearly understood, thereby providing a quantitative basis for subsequent optimization schemes, so as to help identify which parameters are most critical to the stamping precision and stability, and focus on adjusting these parameters during optimization. The prior art often simply adjusts a single parameter or adjusts based on experience, lacking systematic analysis of the coupling relationship between parameters. However, the present application can more accurately identify which parameters need to be optimized in coordination by constructing a coupling coefficient matrix, avoiding the "local optimum" problem that may occur in traditional optimization methods. By setting a dynamic weight value for each initial parameter, it means that each parameter has different importance in different stamping conditions. Dynamic weight adjustment can adjust the optimization focus according to the real-time changes of the stamping process, thereby achieving more accurate optimization results. Traditional optimization methods usually use static weights or fixed parameter importance evaluations, which often cannot be flexibly adjusted in different conditions. However, the present application can perform personalized optimization according to different production environments and conditions through dynamic weight values, better adapting to changing production demands.

[0075] A multi-parameter collaborative optimization objective function is constructed through multiple dynamic weight values, and an improved particle swarm optimization algorithm is used to solve the multi-parameter collaborative optimization objective function to obtain an initial parameter optimization scheme. In the multi-parameter collaborative optimization objective function, the introduction of dynamic weights allows the optimization process of different parameters to be coordinated with each other, and the overall performance will not be unstable due to extreme changes in a single parameter. This objective function can consider the influence of multiple parameters to ensure that the system stability is improved while meeting the process requirements during optimization. Traditional optimization techniques often focus on improving a single parameter, ignoring the influence of other related parameters on the overall system. However, the present application optimizes each parameter and ensures the coordination between parameters through multi-parameter collaborative optimization, effectively avoiding system instability or performance degradation. Particle swarm optimization is a highly efficient global optimization method. Through the improved particle swarm optimization algorithm, some limitations of traditional particle swarm optimization can be overcome, such as slow convergence speed and easy to fall into local optimum. The improved algorithm improves the accuracy and speed of optimization, and can obtain a relatively ideal initial parameter optimization scheme in a relatively short time. Existing technologies may use traditional optimization algorithms such as genetic algorithm or gradient descent method, but these methods may not perform well in handling multiple parameters and complex constraints. However, the present application can better handle complex multi-objective optimization problems through the improved particle swarm optimization algorithm, and has strong global search ability, which can avoid falling into local optimal solution.

[0076] The initial parameter optimization scheme is input into the numerical control punch simulation system to perform a virtual punching test and collect multiple parameter real-time data in the virtual punching process, the parameter fluctuation index is obtained through the multiple parameter real-time data, and the effectiveness of the initial parameter optimization scheme can be verified without actually performing the punching operation through the virtual punching test in the numerical control punch simulation system, which not only saves production cost, but also avoids risks in actual operation, identifies potential problems in advance, and the traditional method usually relies on actual processing test to verify the optimization scheme, which may cause resource waste and production interruption, and the virtual simulation test of the application can be verified multiple times in a safe and low-cost environment to ensure the feasibility of the final optimization scheme, and the real-time data in the virtual punching process is collected to monitor the parameter fluctuation in real time, so that the optimization scheme in the actual punching process will not cause excessive fluctuation or instability, this method not only improves the control accuracy, but also can adjust and correct the optimization scheme in real time to avoid large errors in the production process, and the prior art may rely on periodic inspection or batch control, and the application can dynamically adjust at any time through real-time data feedback to improve the response speed and stability of the system.

[0077] The initial parameter optimization scheme is input into the numerical control punch simulation system to perform a virtual punching test and collect multiple parameter real-time data in the virtual punching process, the parameter fluctuation index is obtained through the multiple parameter real-time data, and the effectiveness of the initial parameter optimization scheme can be verified without actually performing the punching operation through the virtual punching test in the numerical control punch simulation system, which not only saves production cost, but also avoids risks in actual operation, identifies potential problems in advance, and the traditional method usually relies on actual processing test to verify the optimization scheme, which may cause resource waste and production interruption, and the virtual simulation test of the application can be verified multiple times in a safe and low-cost environment to ensure the feasibility of the final optimization scheme, and the real-time data in the virtual punching process is collected to monitor the parameter fluctuation in real time, so that the optimization scheme in the actual punching process will not cause excessive fluctuation or instability, this method not only improves the control accuracy, but also can adjust and correct the optimization scheme in real time to avoid large errors in the production process, and the prior art may rely on periodic inspection or batch control, and the application can dynamically adjust at any time through real-time data feedback to improve the response speed and stability of the system.

[0078] In one embodiment, the step S1 of constructing a multi-parameter coupling model according to each of the initial parameter values comprises:

[0079] S11, constructing a parameter sample set according to a plurality of the initial parameter values, wherein the parameter sample set comprises a plurality of sample points, and each sample point is a normalized initial parameter value;

[0080] S12, performing dimension reduction processing on the parameter sample set by using a principal component analysis method and extracting a plurality of principal components, and constructing a multi-parameter coupling initial model according to the plurality of principal components;

[0081] S13, dividing the parameter sample set into training parameter samples and test parameter samples;

[0082] S14, training and testing the multi-parameter coupling initial model according to the training parameter samples and the test parameter samples until the mean square error of the multi-parameter coupling initial model is less than a preset error threshold, to obtain the multi-parameter coupling initial model.

[0083] As described in the steps S11-S14, before constructing the parameter sample set according to the plurality of initial parameter values, the initial parameters need to be normalized to map the parameter values to the interval [0, 1]. The input layer and the hidden layer of the multi-parameter coupling initial model are constructed according to the extracted principal components. The number of input layer neurons is determined by the number of extracted principal components (i.e. the number of principal components obtained by the principal component analysis method). The number of layers and the number of neurons in each layer of the hidden layer are empirically designed based on the size of the input layer. The number of input layer neurons is the number of principal components. The hidden layer includes 2-3 layers, and the number of neurons in each layer is 1.5-2 times the number of input layer neurons. The training samples are 70%-80% of the sample set, and the test samples are 20%-30% of the sample set. The learning rate is set to 0.01-0.1, and the number of training iterations is 1000-5000 times. The principal components are extracted by dimension reduction, solving the problem of high original parameter dimension and strong correlation. The model input is provided with simplified and complete information features. The input layer directly uses the number of principal components as the dimension. The hidden layer is designed based on the size of the input layer. The coupling relationship between the principal components is learned through a multi-layer network, and finally an initial model reflecting the correlation of multiple parameters is constructed.

[0084] The application constructs a parameter sample set through multiple initial parameter values, wherein the parameter sample set includes multiple sample points, each sample point being a normalized initial parameter value, the first step of constructing the parameter sample set is to form a sample set through multiple initial parameter values, the key of this process lies in collecting enough initial parameter data to cover the diversity of the system, avoiding relying on a single working condition or setting parameter, thereby improving the adaptability of the system to different processing conditions, which is different from the single setting value in the prior art, and can comprehensively capture the variable factors in the stamping process, this method helps to construct a more comprehensive model, enhances the generalization ability and robustness of the subsequent model in actual operation, the normalization step aims to unify parameters of different dimensions or different ranges to a standard scale, eliminating numerical deviation caused by dimension difference between different parameters, since multiple parameters (such as punch speed, pressure, stroke depth, plate positioning accuracy, etc.) are involved in the stamping process, the value range of these parameters may differ greatly, if not normalized, some parameters with large numerical range may dominate in subsequent calculations, thereby affecting the stability and accuracy of the model, after normalization, the contribution weight of all parameters tends to be consistent, making the influence of each parameter on dimension reduction analysis more balanced, improving the overall performance and accuracy of the model;

[0085] By adopting principal component analysis to perform dimension reduction processing on the parameter sample set and extracting multiple principal components, the principal component analysis can map the high-dimensional data set to a low-dimensional space through linear transformation, thereby reducing the redundant information in the data, improving the calculation efficiency, and retaining the main features of the original data, in the stamping control system, multiple parameters involved may have strong correlation, and these correlations may cause multicollinearity problem, making it difficult for traditional methods to effectively process, through principal component analysis, a set of uncorrelated principal components can be extracted, representing the direction of maximum variance in the data, thereby simplifying the complexity of the model and improving the calculation efficiency of the model in processing high-dimensional data, compared with the traditional single parameter adjustment method in the prior art, the principal component analysis not only extracts the most important features, but also reduces the calculation amount, making the multi-parameter coupling model more efficient, avoiding the problem of too high calculation complexity, and constructing a multi-parameter coupling initial model according to multiple principal components, the multi-parameter coupling initial model constructed based on multiple principal components can effectively reduce the dimension and complexity of the model while retaining most of the information, by taking multiple principal components as input, a model integrating the influence of multiple parameters is constructed, which can capture the interaction between parameters in the stamping process, compared with the single parameter adjustment in the prior art, this method can consider the comprehensive influence of multiple parameters on the stamping process at the same time, thereby avoiding the chain reaction caused by single parameter adjustment, this multi-parameter coupling model can find a more accurate solution in a higher-dimensional control space, improving the stability and processing precision of the system;

[0086] By dividing the parameter sample set into training parameter samples and test parameter samples, the multi-parameter coupling initial model is trained and tested by the training parameter samples and the test parameter samples until the mean square error of the multi-parameter coupling initial model is less than the preset error threshold, and the multi-parameter coupling initial model is obtained. Dividing the data set into training samples and test samples is a standard step in the machine learning and modeling process. By dividing the data into training sets and test sets, overfitting can be effectively avoided to ensure that the model does not only remember the noise of the data during training, but can learn the potential laws of the data. The training samples are used to adjust the parameters of the model, and the test samples are used to evaluate the generalization ability of the model. This step ensures the consistency of the model's performance on different data sets and improves the reliability and applicability of the model. In the traditional single parameter adjustment method, due to the lack of an effective test verification process, it may lead to over-reliance on specific parameter values ​​and inability to fully adapt to different working conditions. The performance of the multi-parameter coupling model is continuously optimized through the training and testing process until the mean square error of the model is lower than the preset error threshold. The mean square error is commonly used in regression analysis. The measurement index can effectively evaluate the prediction accuracy of the model in practical applications. In this step, by continuously optimizing and adjusting the model parameters, it can ensure that the model maintains a high accuracy under the interaction of multiple parameters, thereby improving the stability and quality of the stamping control system. Unlike the adjustment method in the existing technology that relies only on intuition and experience, this step adopts a scientific training and verification mechanism to ensure that the output of the model is more reliable and avoids system fluctuations or failures caused by improper parameter adjustment. By continuously iterating training and testing until the model error is lower than the preset threshold, it can ensure that the model can give more accurate predictions and adjustment suggestions under all possible working conditions. This iterative optimization process can gradually approach the global optimal solution of the system, thereby avoiding the error accumulation and system instability caused by local parameter optimization in traditional methods. In the control of high-precision CNC punching machines, it is crucial to ensure the prediction accuracy of each step and the stability of the adjustment strategy. Through multiple rounds of training, the model can not only adapt to a single working condition, but also cope with complex multi-parameter coupling scenarios, thereby greatly improving the processing capability of the system.

[0087] In one embodiment, the step S2 of performing quantitative analysis on the coupling relationship between the initial parameters according to the multi-parameter coupling model to obtain a parameter coupling coefficient matrix includes:

[0088] S21, constructing a deep belief network according to the initial parameter values ​​of the multi-parameter coupling model, and inputting the normalized initial parameter values ​​in the multi-parameter coupling model into the visible layer of the deep belief network;

[0089] S22. Train the hidden layer of the deep belief network layer by layer through a greedy algorithm, and obtain the activation probability of the hidden layer based on the data of the visible layer;

[0090] S23. Obtain reconstructed visible layer data according to the activation probability of the hidden layer, and update the network weights and bias according to the difference between the reconstructed visible layer data and the original visible layer data;

[0091] S24, extracting the weight matrix of each layer of the deep belief network after training is completed, and calculating the parameter coupling coefficient according to each weight matrix, wherein the calculation formula is: ; =1,2,,,, , k≠j;

[0092] in, Representation parameters The parameter coupling coefficient with parameter j, Representation parameters and the weight value of parameter j in the weight matrix, Representation parameters With parameters (All other parameters) The maximum value among the weight values ​​in the weight matrix is ​​used for normalization to eliminate the dimension effect. i, j, and k all represent the parameter numbers;

[0093] S25. Arrange the multiple parameter coupling coefficients in parameter order to construct a parameter coupling coefficient matrix.

[0094] As described in steps S21-S25, the structure of the deep belief network is initialized to include 1 visible layer and 3-5 hidden layers, the number of visible layer neurons is the total number of initial parameters, the number of hidden layer neurons decreases layer by layer, and the number of neurons in each layer is 0.5-0.8 times that of the previous layer to achieve layer-by-layer extraction and compression of parameter characteristics; the training parameters of the deep belief network are set, including a learning rate of 0.001-0.01 for controlling the step size of network parameter updating to avoid convergence oscillation or slow convergence, a training batch size of 32-64 for dividing the input data into multiple batches for training to balance training efficiency and model stability, and a training iteration number of 500-1000 times to ensure that the network fully learns the coupling characteristics between parameters until the model converges; the connection parameters between the current layer and the previous layer are trained each time by a greedy algorithm, ignoring the influence of other layers, to achieve hierarchical optimization, and only the connection parameters between the current layer and the previous layer are trained each time, ignoring the influence of other layers, to achieve hierarchical optimization. In the training process, the network parameters are updated using the contrastive divergence algorithm, the steps of which are: first, calculate the activation probability of the hidden layer according to the visible layer data (achieved by the Sigmoid activation function, which is a function used to calculate the activation probability of the hidden layer in the training process of the deep belief network, and its role is to map the input value to the [0, 1] interval to quantify the correlation strength between parameters), then reconstruct the visible layer data according to the activation probability of the hidden layer to simulate the reverse transmission of parameter characteristics, and finally calculate the difference (error value) between the original visible layer data and the reconstructed visible layer data, and update the network weights and biases based on the difference to minimize the reconstruction error. In the weight matrix of each layer of the deep belief network, the weight matrix reflects the connection strength between different parameters, and each element represents the connection strength between the corresponding two parameters (or characteristics), and the larger the value, the more significant the coupling relationship between the two. Each element in the parameter coupling coefficient matrix represents the degree of mutual influence between the corresponding two parameters, wherein the coupling coefficient between punch speed and pressure is calculated by the ratio of punch speed change and pressure change, the coupling coefficient between pressure and stroke depth is determined by the product of pressure fluctuation value and stroke depth deviation value, and the coupling coefficient between stroke depth and plate positioning accuracy is calculated by the difference between the actual value and the theoretical value of the stroke depth and the mean square error of the plate positioning accuracy error. Each element in the parameter coupling coefficient matrix directly reflects the mutual influence of the corresponding two parameters, so that the coupling relationship between the initial parameters can be quantitatively analyzed;

[0095] The present invention constructs a deep belief network through the initial parameter values ​​of the multi-parameter coupling model, and inputs the normalized initial parameter values ​​in the multi-parameter coupling model into the visible layer of the deep belief network. In the existing CNC punching machine intelligent control system, it usually relies on a single control parameter or a combination of certain parameters to adjust the system performance. However, there is a complex coupling relationship between multiple parameters (such as punch speed, pressure, stroke depth, etc.) in the stamping process. It is difficult for traditional control methods to integrate the multi-dimensional information of these parameters and optimize the combination of each parameter to ensure the acquisition of the global optimal solution. The present invention uses the initial parameter values ​​of the multi-parameter coupling model as input to construct a deep belief network, which can be trained The network is trained to simultaneously consider the mutual influence of multiple parameters, thereby avoiding the chain reaction that may be caused by the adjustment of a single parameter, thereby improving the system's overall optimization ability for the coupling of multiple parameters and providing higher stability and reliability. Through normalization, the initial parameter values ​​will be converted to a unified scale, which can avoid the unnecessary impact of dimensional differences of different parameters on network training. The normalized data helps to accelerate the training process and improve the convergence speed and accuracy of the network. Since the parameters involved in the punching operation vary greatly, this process ensures that the deep belief network can effectively capture the complex relationship between parameters, thereby reducing the error caused by data inconsistency;

[0096] The hidden layers of a deep belief network are trained layer by layer using a greedy algorithm. This is a common strategy in deep belief network training. In each layer, the greedy algorithm first trains the connection from the visible layer to the hidden layer, thus avoiding the gradient vanishing problem in complex backpropagation training. For the intelligent control of high-precision CNC punching machines, greedy algorithm layer-by-layer training can better capture the hierarchical structure between each layer and gradually extract high-order features from the data. Compared with traditional end-to-end training methods, this strategy reduces training time and improves model stability and accuracy. It can effectively avoid model overfitting, especially when dealing with complex parameter coupling problems.

[0097] According to the data acquisition hidden layer of visible layer, the activation probability of hidden layer is obtained by the activation probability of hidden layer, the activation probability of hidden layer is obtained by the data of visible layer, which means that the deep belief network not only carries out prediction or reconstruction, but also can reflect the 'confidence' or 'importance' of each layer network when processing data, in the application scene of high precision numerical control punch, the activation probability of hidden layer provides key data support for subsequent parameter optimization, for example, the influence of some stamping parameters may not be significant, while some parameters may have important influence on the final stamping result, through the activation probability system, the parameters which have greater influence on stamping precision can be focused on more intelligently, so as to improve the intelligent adjustment ability in the machining process, by obtaining the reconstructed visible layer data, the system can correct itself and optimize the adjustment of parameters, through the reconstruction error (i.e. the difference between the original visible layer data and the reconstructed data), the error source and potential problems in parameter adjustment can be identified, in the prior art, due to the lack of effective error feedback mechanism, small machining defects may be produced in the stamping process, and these defects cannot be found in real time, but the deep belief network can optimize the stamping parameters in real time in each machining through the reconstruction error, so as to reduce the accumulation of errors and maintain the machining precision;

[0098] The network weight and bias are updated by the difference between the reconstructed visible layer data and the original visible layer data, and the network weight and bias are updated by error, so that the network learns more accurate parameter coupling relationship. In high-precision punch press control, small errors in parameter adjustment may cause unstable processing or defects. The existing control method often ignores the comprehensive influence of multiple parameters. This parameter updating method based on error feedback enables the system to adjust each parameter in real time to obtain better stamping performance. Unlike the existing technology of adjusting each parameter separately, the feedback updating method based on deep belief network can more accurately optimize the interaction between multiple parameters to achieve a global optimal solution. By extracting the weight matrix of each layer of the deep belief network after training and calculating the parameter coupling coefficients according to each weight matrix, the control method in the prior art relies on artificial rules or empirical formulas and lacks dynamic adjustment capability. The present application can automatically obtain the coupling coefficients between different parameters through the training of the deep belief network and the extraction of the weight matrix. These coefficients reflect the dependency between parameters. Compared with the traditional fixed rule method, the deep belief network can learn and dynamically adjust the parameter coupling coefficients in real time, so that the punch press system can achieve optimal performance under different working conditions. By arranging the multiple parameter coupling coefficients in parameter order to construct a parameter coupling coefficient matrix, the interaction between multiple parameters in a high-precision numerical control punch press is difficult to describe and optimize by traditional single control parameter. The present application can directly use these coefficients to adjust the parameter settings when facing multiple complex stamping conditions, thereby improving the stability and precision of processing. Compared with the potential chain reaction caused by single parameter adjustment in the prior art, the present application can optimize the global performance of the system based on considering the mutual influence of multiple parameters, reduce processing defects, and improve stamping stability.

[0099] In one embodiment, the step S3 of solving the multi-parameter collaborative optimization objective function to obtain an initial parameter optimization scheme comprises:

[0100] S31, input each initial parameter value into the multi-parameter collaborative optimization objective function to obtain multiple fitness values, and select the initial parameter value corresponding to the minimum fitness value from the multiple fitness values as the individual optimal position;

[0101] S32, select the initial parameter value corresponding to the minimum fitness value from the multiple individual optimal positions as the global optimal position, and obtain the global optimal position deviation according to the global optimal position and the individual optimal position;

[0102] S33, obtain the historical optimal position of the initial parameter value corresponding to each individual optimal position, and obtain the individual optimal position deviation according to the historical optimal position and the individual optimal position;

[0103] S34, obtaining a current speed of each initial parameter value corresponding to the individual optimal position, and obtaining an update speed of the corresponding initial parameter value according to the current speed, the individual optimal position deviation and the global optimal position deviation;

[0104] S35, obtaining an update position according to the update speed and the individual optimal position, and obtaining an initial parameter optimization scheme according to the update position and the update speed.

[0105] As described above in steps S31-S35, before each initial parameter value is input into the multi-parameter collaborative optimization objective function, the position and speed of the initialized particle swarm need to be determined. The position vector of the particle (initial parameter value) represents a set of parameter combinations (collective group) to be optimized, that is, the stamping core parameters of the high-precision numerical control punch press including the initial speed of the punch, the initial pressure, the initial stroke depth and the initial positioning accuracy of the plate, and the speed vector of the particle is used to represent the rate of parameter adjustment. Since the initial parameter value is a collective group including the stamping core parameters of the high-precision numerical control punch press such as the initial speed of the punch, the initial pressure, the initial stroke depth and the initial positioning accuracy of the plate, when input into the multi-parameter collaborative optimization objective function, the fitness value corresponding to each stamping core parameter is obtained. The smaller the fitness value, the closer the parameter combination corresponding to the particle to the global optimal solution. For each particle, compare the current fitness value with the historical fitness value, and determine the position corresponding to the minimum fitness value as the individual optimal position of the particle. In all individual optimal positions of the particles, the position with the smallest fitness value is selected as the global optimal position. Based on the output fitness value result, the preliminary screening of the local and global optimum is completed, and the direction for subsequent parameter iteration is provided.

[0106] The formula of the update speed is: ; wherein, represents the speed of the i-th particle in the j-th parameter dimension at the k-th iteration, represents the inertia weight, represents the current iteration number, represents the current position of the i-th particle in the j-th parameter dimension at the k-th iteration, represents the first learning factor, represents the second learning factor, represents the first random number in the interval [0, 1], represents the second random number in the interval [0, 1], which respectively guides the particle to move towards the individual optimal and global optimal directions, ​​​​​​Indicates the The particle in The optimal position of an individual in the parameter dimension, Indicates that all particles in The global optimal position in the parameter dimension;

[0107] The formula for updating the position is: ; Adjust the particle position based on the updated speed to achieve iterative optimization of the parameter combination. The updated position is the superposition of the current position of the particle and the new speed. The essence is to convert the parameter adjustment trend represented by the speed into an actual parameter value change, so that the particle moves to a better position in the parameter space. Before the step of obtaining the updated position according to the updated speed and the individual optimal position, it is also necessary to judge the iteration termination condition. Specifically, if the current number of iterations reaches the preset maximum value, the iteration is stopped, and the parameter combination corresponding to the global optimal position is output as the initial parameter optimization solution. If the termination condition is not met, return to recalculate the fitness value and repeat the iterative steps until the termination condition is met.

[0108] The present invention obtains multiple fitness values ​​by inputting each initial parameter value into the multi-parameter collaborative optimization objective function. Compared with the traditional single parameter adjustment method, the present invention can more comprehensively consider the relationship and influence between each parameter by inputting multiple parameters into the collaborative optimization objective function, avoiding system instability or local optimal solution that may be caused by single parameter adjustment. Traditional technology often takes a certain parameter as the main optimization target and ignores the linkage effect of other parameters. The present invention solves the chain reaction problem caused by single parameter adjustment by considering multiple parameters at the same time, so that the system can more comprehensively and accurately find the optimal parameter combination, and select the initial parameter value corresponding to the minimum fitness value from multiple fitness values ​​as the individual optimal position. By evaluating multiple fitness values, the initial parameter value corresponding to the minimum fitness value is selected as the individual optimal position, which ensures that the system can always move towards the minimization target during the optimization process, thereby improving optimization efficiency and accuracy. Traditional optimization methods may fall into local optimality, but the present invention avoids this situation by selecting the minimum fitness value, which helps to ultimately find the global optimal solution;

[0109] The initial parameter value corresponding to the minimum fitness value is selected from the multiple individual optimal positions as the global optimal position, and the global optimal position deviation is obtained according to the global optimal position and the individual optimal position, the global optimal solution is ensured by selecting the parameter with the minimum fitness value from the multiple individual optimal positions, compared with the traditional method which may only consider the local optimum of the individual, the information of all individuals is integrated, the stability and performance of the overall system are improved, the global optimal position is selected, the system can realize collaborative optimization in multiple dimensions, the limitation of single dimension is avoided, the deviation between the global optimal position and the individual optimal position is calculated, the difference between the global and local optimal is compared in the optimization process, and the optimization path is further refined, the deviation calculation can guide the adjustment pace in the optimization process, so that the parameter adjustment is more fine and efficient, through the calculation and feedback of the deviation, the pace can be dynamically adjusted in the optimization process, over-adjustment or slow adjustment is avoided, and the convergence speed of the optimization algorithm is improved;

[0110] The historical optimal position of the initial parameter value corresponding to each individual optimal position is obtained, and the individual optimal position deviation is obtained according to the historical optimal position and the individual optimal position, by introducing the data of the historical optimal position, the past optimization experience can be used, the current optimization process is more intelligent, and the possibility of blind adjustment is reduced, which is more efficient than the real-time data in the prior art, by considering the historical optimal position, the optimization algorithm not only depends on the current optimization state, but also balances the past successful experience, so that overfitting or local optimal phenomenon can be avoided in some special states, the current speed of the initial parameter value corresponding to each individual optimal position is obtained, and the update speed of the corresponding initial parameter value is obtained according to the current speed, the individual optimal position deviation and the global optimal position deviation, the update speed is calculated by combining the current speed, the individual optimal deviation and the global optimal deviation in the application, the flexibility and adaptability of parameter adjustment are ensured, compared with the traditional fixed update step method, the optimization speed can be automatically adjusted according to the real-time situation, so that the parameter update is more accurate, and the instability caused by too large or too small step is reduced, by adjusting the update speed, the problems of system oscillation or low optimization efficiency caused by improper update step are avoided, so that the overall optimization stability is improved;

[0111] The updating position is obtained by updating the speed and the individual optimal position, and the initial parameter optimization scheme is obtained according to the updating position and the updating speed, the position is updated by combining the updating speed and the individual optimal position, the final value of each parameter can be further optimized, the optimization process depends not only on the global optimum, but also considers the influence of each local optimal position, thereby avoiding the adverse effects of local optimum, the application ensures that the dynamic characteristics of speed adjustment are considered when the updating position is obtained, thereby avoiding the limitations of a single static optimization method, the final optimization scheme can remain efficient and stable during long-time operation, compared with the traditional adaptive parameter adjustment control method, the multi-parameter collaborative optimization method of the application considers the mutual influence between multiple parameters and the dynamic adjustment mechanism, avoids the chain reaction problem caused by single parameter adjustment, enables the system to find a global optimal solution, significantly improves the processing stability, by combining historical optimal position, current speed, local and global optimal deviation and other factors, the application can effectively improve the optimization efficiency, avoid overfitting or local optimum, and finally realize more efficient and stable intelligent stamping control.

[0112] In one embodiment, the step S5 of obtaining the parameter fluctuation index according to the plurality of parameter real-time data comprises:

[0113] S51, pre-processing each parameter real-time data to obtain corresponding standard parameter data;

[0114] S52, obtaining a plurality of initial parameter values in the initial parameter optimization scheme, and obtaining a corresponding parameter deviation value according to each initial parameter value and the standard parameter data;

[0115] S53, obtaining the parameter deviation sum of squares according to a plurality of parameter deviation values, and obtaining the parameter fluctuation index according to the parameter deviation sum of squares.

[0116] As described in steps S51-S53, wherein the fluctuation index is the square root of the sum of squares of parameter deviations, the application obtains corresponding standard parameter data by preprocessing real-time data of each parameter. The preprocessing of real-time data can effectively filter out interference factors caused by changes in external environment or device noise, thereby ensuring the accuracy and reliability of the data. By converting real-time data into standard parameter data, a stable benchmark can be provided for the subsequent optimization process. Compared with the prior art, the use of unprocessed real-time data directly can lead to fluctuations or deviations in the calculation results, so preprocessing can improve the accuracy of the system, reduce errors caused by unstable data during the stamping process, and effectively improve the machining precision. By obtaining multiple initial parameter values in the initial parameter optimization scheme and obtaining corresponding parameter deviation values according to each initial parameter value and standard parameter data, multiple initial parameter values are obtained to ensure that all possible parameter ranges are covered during the optimization process, rather than relying solely on a single initial value. By introducing multiple initial parameter values, the application provides diversified choices for parameter adjustment during the optimization process, thereby avoiding local optimization problems and increasing the probability of finding a global optimal solution, thereby improving the overall performance of stamping control. The deviation value between each initial parameter value and the standard parameter data is obtained by quantifying the deviation to measure the gap between the current system state and the target state, so that the system can accurately track the influence of each parameter on the final machining result. Compared with traditional adaptive control methods, single parameter adjustment often ignores the mutual relationship between parameters, which can easily lead to inaccurate adjustment and non-optimal control effect. By obtaining the deviation value of each parameter in real time, the application can more accurately analyze and diagnose the influence of each parameter on the stamping process, thereby providing a fine basis for subsequent adjustment.The parameter deviation square sum is obtained through the plurality of parameter deviation values, and the parameter fluctuation index is obtained according to the parameter deviation square sum, the parameter deviation square sum is calculated through the plurality of parameter deviation values, the square sum is adopted, the overall deviation between parameters can be more comprehensively reflected, the excessive influence of a single parameter on the overall performance is avoided, not only the deviation of an individual parameter is considered, but also the influence of a plurality of parameters is comprehensively considered, the overall state of the system can be optimized, in the prior art, many methods simply depend on the adjustment of a key parameter, the influence of other parameters can be ignored, so that the imbalance of the optimization result is caused, through the square sum method, the present application provides a way of comprehensively considering the interaction of all parameters, the plurality of variables in the stamping process can be more effectively balanced, the system disorder caused by local adjustment is reduced, the introduction of the parameter fluctuation index can quantify the fluctuation of the whole system, as a key indicator of stamping stability, the value of the fluctuation index can reflect the stability of the system in real time, through monitoring the change trend, the abnormal fluctuation in the stamping process can be responded in time, the traditional adaptive control system often excessively depends on the fluctuation of a single parameter, and the overall fluctuation of a plurality of parameters is ignored, the fluctuation of a plurality of parameters is combined into a whole indicator through the fluctuation index, the stability of the stamping process can be more comprehensively and accurately monitored and adjusted, so that the machining defects caused by parameter fluctuation are effectively avoided, the machining quality is improved, through comprehensively considering the deviation and fluctuation of all parameters, the overall optimization of the system level is realized, the coordination between a plurality of parameters is ensured, and the machining stability of the high-precision numerical control punch is effectively improved.

[0117] In one embodiment, the step S5 of correcting the multi-parameter coupling model according to the parameter real-time data comprises:

[0118] S54, inputting each initial parameter value into the multi-parameter coupling model to obtain first predicted parameter data, and obtaining a corresponding parameter residual according to the first predicted parameter data and the parameter real-time data;

[0119] S55, fitting the parameter residual by using the least square method to obtain a residual model, and adding the residual model to the multi-parameter coupling initial model to obtain a corrected multi-parameter coupling model;

[0120] S56, inputting each initial parameter value into the corrected multi-parameter coupling model to obtain second predicted parameter data;

[0121] S57, obtaining an actual parameter value of each initial parameter value under a preset stamping working condition, and obtaining a corresponding absolute deviation according to the difference between each actual parameter value and the second predicted parameter data;

[0122] S58, obtaining a root mean square error according to a plurality of absolute deviations, and determining whether the root mean square error is greater than a preset error threshold.

[0123] If the root mean square error is greater than the preset error threshold, it is determined that the modified multi-parameter coupling model is not verified, and the step of fitting the parameter residual by the least square method is returned to until the root mean square error is not greater than the preset error threshold;

[0124] If the root mean square error is not greater than the preset error threshold, it is determined that the modified multi-parameter coupling model is verified.

[0125] As described in steps S54-S58, the present application inputs each initial parameter value into the multi-parameter coupling model to obtain first predicted parameter data, and obtains corresponding parameter residual according to the first predicted parameter data and the real-time parameter data. The present application comprehensively considers multiple interacting parameters (such as punch speed, pressure, stroke depth, plate positioning accuracy, etc.) through the multi-parameter coupling model, more accurately reflects the multi-dimensional dynamic characteristics of the entire stamping process, avoids the chain reaction that may be caused by single parameter adjustment, provides more comprehensive preliminary prediction results, breaks through the bottleneck of traditional methods limited to single parameter optimization, generates residual by comparing the first predicted parameter data with the real-time data, and the residual reflects the deviation between the prediction and the actual result, can directly reflect the deviation size of the current model, and provides a basis for subsequent model adjustment by quantifying the error. The traditional adaptive control method may only focus on real-time correction of a single parameter, without fully considering the complex interaction between multiple parameters. The residual calculation of the present application enables us to capture these complex deviations and provides a basis for more accurate correction.

[0126] The residual model is obtained by fitting the parameter residual by the least square method, and the residual model is added to the multi-parameter coupling initial model to obtain the modified multi-parameter coupling model. The least square method fitting residual can accurately obtain the optimal parameters of the modified model, minimize the residual, and further improve the prediction accuracy. Unlike the simple linear or empirical model that may be used in traditional methods, the least square method can adapt to the multivariate error of complex systems through a strict mathematical optimization process, ensuring the efficiency and accuracy of model correction. By combining the difference between the preliminary prediction and the actual observation value, a modified model closer to the actual working condition is generated. This correction method has stronger global adaptability than the local correction of traditional methods, can dynamically adjust the overall model according to the mutual influence of multiple parameters, thereby effectively improving the control accuracy and avoiding the instability caused by single parameter adjustment.

[0127] The second predicted parameter data is obtained by inputting each initial parameter value into the corrected multi-parameter coupling model, the actual parameter value of each initial parameter value under the preset stamping working condition is obtained, the corresponding absolute deviation is obtained according to the difference between each actual parameter value and the second predicted parameter data, new prediction is input into the corrected model, the effect of the adjusted model can be verified, the corrected model can adapt to the changes in actual production through continuous correction and prediction verification, compared with the traditional method of 'parameter adjustment and use', the model is predicted again through the corrected model, the variability of the system and the complexity between the multiple parameters are fully considered, so that the prediction result is more accurate and can better reflect the real working condition, by comparing the predicted value with the actual value, the deviation of the model prediction can be accurately quantified, unlike the traditional method of adjusting the control strategy only according to a single error threshold, this step provides specific adjustment basis for each parameter by calculating the actual deviation, so that each parameter can be independently adjusted and monitored in real time, thereby realizing fine control and enhancing the adaptability of the model, so that the system can cope with different working condition changes and avoid system instability caused by inaccurate control of each parameter in the traditional method.

[0128] The root mean square error is obtained through the plurality of absolute deviations, and it is judged whether the root mean square error is greater than a preset error threshold, if the root mean square error is greater than the preset error threshold, it is determined that the modified multi-parameter coupling model fails to pass the verification, and the step of fitting the parameter residual through the least square method to obtain the residual model is returned to until the root mean square error is not greater than the preset error threshold, if the root mean square error is not greater than the preset error threshold, it is determined that the modified multi-parameter coupling model passes the verification, the root mean square error is obtained through the plurality of absolute deviations, which can more comprehensively evaluate the precision of the model, compared with the single deviation method, the root mean square error can better reflect the overall deviation of the model on multiple parameters, avoiding the false judgment caused by the excessive single deviation, the root mean square error as a comprehensive evaluation index can more accurately reflect the actual error range of the model, thereby providing a scientific standard to judge whether the model is accurate enough, through the judgment and control of the error, it is ensured that the model can reach the preset precision requirement after each iteration, unlike the traditional method, the effectiveness of the modified model is verified dynamically, the correction effect is evaluated in real time by using the root mean square error, the potential risks that may exist when the model is directly put into use after being modified are avoided, it is ensured that the model can be strictly verified after each adjustment, the precision and stability of the process are improved, the defect that the model is no longer verified after being adjusted in the traditional method is avoided, the model is adjusted through continuous iteration until the root mean square error reaches the preset error threshold, and finally it is ensured that the modified model can fully adapt to the actual working condition, unlike the fixed parameter adjustment of the traditional method, the present application provides a dynamic correction process, so that the model can be continuously optimized, and high precision stability under different working conditions is ensured, each adjustment is accurately controlled within the error range, which can avoid excessive correction while ensuring that each correction minimizes the error, thereby improving the stability and precision of the stamping process.

[0129] The application also provides an intelligent stamping control system of a high-precision numerical control punch press, comprising:

[0130] A construction module is configured to obtain a plurality of initial parameter values of the high-precision numerical control punch press under a preset stamping working condition, and construct a multi-parameter coupling model according to each initial parameter value.

[0131] A quantitative analysis module is configured to quantitatively analyze the coupling relationship between the initial parameters according to the multi-parameter coupling model, obtain a parameter coupling coefficient matrix, and set a dynamic weight value for each initial parameter value according to the parameter coupling coefficient matrix.

[0132] A solving module is configured to construct a multi-parameter collaborative optimization objective function according to a plurality of dynamic weight values, and solve the multi-parameter collaborative optimization objective function by using an improved particle swarm optimization algorithm to obtain an initial parameter optimization scheme.

[0133] a test module, configured to input the initial parameter optimization scheme into a numerical control punch simulation system to perform a virtual punching test and collect real-time data of a plurality of parameters in the virtual punching process;

[0134] a judgment module, configured to acquire a parameter fluctuation index according to the real-time data of the plurality of parameters, and determine whether the parameter fluctuation index is less than a preset fluctuation threshold;

[0135] if the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is taken as a candidate optimization scheme;

[0136] if the parameter fluctuation index is not less than the preset fluctuation threshold, the multi-parameter coupling model is corrected according to the real-time data of the plurality of parameters, and the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model is returned to until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is taken as the candidate optimization scheme.

[0137] In one embodiment, the construction module comprises:

[0138] a first construction unit, configured to construct a parameter sample set according to the plurality of initial parameter values, wherein the parameter sample set comprises a plurality of sample points;

[0139] a second construction unit, configured to perform dimension reduction processing on the parameter sample set by using a principal component analysis method and extract a plurality of principal components, and construct a multi-parameter coupling initial model according to the plurality of principal components;

[0140] a division unit, configured to divide the parameter sample set into training parameter samples and test parameter samples;

[0141] a training unit, configured to train and test the multi-parameter coupling initial model according to the training parameter samples and the test parameter samples until a mean square error of the multi-parameter coupling initial model is less than a preset error threshold, to obtain the multi-parameter coupling initial model.

[0142] It should be noted that each module and unit in the intelligent punching control system of the high-precision numerical control punch corresponds to one step in the intelligent punching control method of the high-precision numerical control punch.

[0143] As shown in Figure 3 The present application also provides a computer device, which can be a server, and the internal structure thereof can be as shown in Figure 3The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store all data required by the process of the intelligent stamping control method of the high-precision numerical control punch. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the intelligent stamping control method of the high-precision numerical control punch.

[0144] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied.

[0145] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. The computer program is executed by the processor to implement the intelligent stamping control method of the high-precision numerical control punch.

[0146] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to the memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM) and the like.

[0147] It is to be understood that the terminology "including", "comprising", or any other variation thereof, is intended to cover a non-exclusive inclusion such that process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0148] The above description is merely the preferred embodiments of the present application, and is not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made according to the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An intelligent stamping control method for a high-precision CNC punch press, characterized in that: include: Acquire multiple initial parameter values ​​of a high-precision CNC punch press under a preset punching working condition, and construct a multi-parameter coupling model based on each of the initial parameter values; Quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model to obtain a parameter coupling coefficient matrix, and setting a dynamic weight value for each initial parameter value according to the parameter coupling coefficient matrix; Constructing a multi-parameter collaborative optimization objective function according to the plurality of dynamic weight values, and solving the multi-parameter collaborative optimization objective function using an improved particle swarm optimization algorithm to obtain an initial parameter optimization solution; Inputting the initial parameter optimization scheme into a CNC punch press simulation system to perform a virtual punching test and collecting real-time data of multiple parameters during the virtual punching process; Obtaining a parameter fluctuation index based on the real-time data of the plurality of parameters, and determining whether the parameter fluctuation index is less than a preset fluctuation threshold; If the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is used as a candidate optimization scheme; If the parameter fluctuation index is not less than the preset fluctuation threshold, the multi-parameter coupling model is corrected according to the real-time parameter data, and the process returns to the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is used as a candidate optimization scheme.

2. The intelligent stamping control method for a high-precision CNC punch press according to claim 1 is characterized in that: The step of constructing a multi-parameter coupling model according to each of the initial parameter values ​​comprises: Constructing a parameter sample set according to the multiple initial parameter values, wherein the parameter sample set includes multiple sample points; Using principal component analysis to perform dimensionality reduction processing on the parameter sample set and extract multiple main components, and constructing a multi-parameter coupling initial model based on the multiple main components; Dividing the parameter sample set into training parameter samples and testing parameter samples; The multi-parameter coupling initial model is trained and tested according to the training parameter samples and the test parameter samples until the mean square error of the multi-parameter coupling initial model is less than a preset error threshold, thereby obtaining the multi-parameter coupling initial model.

3. The intelligent stamping control method for a high-precision CNC punch press according to claim 1, characterized in that: The step of performing quantitative analysis on the coupling relationship between the initial parameters according to the multi-parameter coupling model to obtain a parameter coupling coefficient matrix includes: Constructing a deep belief network according to the initial parameter values ​​of the multi-parameter coupling model, and inputting the normalized initial parameter values ​​in the multi-parameter coupling model into a visible layer of the deep belief network; The hidden layer of the deep belief network is trained layer by layer through a greedy algorithm, and the activation probability of the hidden layer is obtained based on the data of the visible layer; Obtain the reconstructed visible layer data based on the activation probability of the hidden layer, and update the network weights and biases based on the difference between the reconstructed visible layer data and the original visible layer data; Extracting the weight matrix of each layer of the deep belief network after training is completed, and obtaining the parameter coupling coefficient according to each weight matrix; Arrange multiple parameter coupling coefficients in parameter order to construct a parameter coupling coefficient matrix.

4. The intelligent stamping control method for a high-precision CNC punch press according to claim 1, characterized in that: The step of solving the multi-parameter collaborative optimization objective function to obtain an initial parameter optimization solution includes: Inputting each of the initial parameter values ​​into the multi-parameter collaborative optimization objective function to obtain multiple fitness values, and selecting the initial parameter value corresponding to the minimum fitness value from the multiple fitness values ​​as the individual optimal position; Selecting an initial parameter value corresponding to the minimum fitness value from the multiple individual optimal positions as the global optimal position, and obtaining a global optimal position deviation based on the global optimal position and the individual optimal position; Obtaining a historical optimal position of the initial parameter value corresponding to each of the individual optimal positions, and obtaining an individual optimal position deviation based on the historical optimal position and the individual optimal position; Obtaining a current speed of the initial parameter value corresponding to each of the individual optimal positions, and obtaining an update speed of the corresponding initial parameter value based on the current speed, the individual optimal position deviation, and the global optimal position deviation; An updated position is obtained according to the update speed and the individual optimal position, and an initial parameter optimization solution is obtained according to the update position and the update speed.

5. The intelligent stamping control method for a high-precision CNC punch press according to claim 1, characterized in that: The step of obtaining a parameter fluctuation index based on the real-time data of the plurality of parameters comprises: Preprocessing the real-time data of each parameter to obtain corresponding standard parameter data; Obtaining multiple initial parameter values ​​in the initial parameter optimization scheme, and obtaining corresponding parameter deviation values ​​according to each of the initial parameter values ​​and standard parameter data; A parameter deviation square sum is obtained according to the plurality of parameter deviation values, and a parameter fluctuation index is obtained according to the parameter deviation square sum.

6. The intelligent stamping control method for a high-precision CNC punch press according to claim 1, characterized in that: The step of correcting the multi-parameter coupling model according to the real-time parameter data includes: Inputting each of the initial parameter values ​​into a multi-parameter coupling model to obtain first predicted parameter data, and obtaining corresponding parameter residuals based on the first predicted parameter data and parameter real-time data; The parameter residuals are fitted using the least squares method to obtain a residual model, and the residual model is added to the multi-parameter coupling initial model to obtain a revised multi-parameter coupling model; Inputting each of the initial parameter values ​​into the modified multi-parameter coupling model to obtain second prediction parameter data; Obtaining an actual parameter value of each of the initial parameter values ​​under a preset stamping working condition, and obtaining a corresponding absolute deviation based on each of the actual parameter values ​​and the second predicted parameter data; Obtaining a root mean square error according to the multiple absolute deviations, and determining whether the root mean square error is greater than a preset error threshold; If the root mean square error is greater than the preset error threshold, it is determined that the modified multi-parameter coupling model verification fails, and the process returns to the step of fitting the parameter residuals using the least squares method to obtain a residual model until the root mean square error is no greater than the preset error threshold; If the root mean square error is not greater than the preset error threshold, it is determined that the modified multi-parameter coupling model has passed the verification.

7. An intelligent stamping control system for a high-precision CNC punch press, characterized in that: include: A construction module is used to obtain multiple initial parameter values ​​of the high-precision CNC punch press under a preset punching working condition, and to construct a multi-parameter coupling model according to each of the initial parameter values; a quantitative analysis module, configured to perform a quantitative analysis on the coupling relationship between the initial parameters according to the multi-parameter coupling model, obtain a parameter coupling coefficient matrix, and set a dynamic weight value for each initial parameter value according to the parameter coupling coefficient matrix; A solution module is used to construct a multi-parameter collaborative optimization objective function according to the multiple dynamic weight values, and use an improved particle swarm optimization algorithm to solve the multi-parameter collaborative optimization objective function to obtain an initial parameter optimization solution; A test module, used for inputting the initial parameter optimization scheme into a CNC punch press simulation system to perform a virtual punching test and collect real-time data of multiple parameters during the virtual punching process; a judgment module, configured to obtain a parameter fluctuation index based on the real-time data of the plurality of parameters, and to judge whether the parameter fluctuation index is less than a preset fluctuation threshold; If the parameter fluctuation index is less than the preset fluctuation threshold, the initial parameter optimization scheme is used as a candidate optimization scheme; If the parameter fluctuation index is not less than the preset fluctuation threshold, the multi-parameter coupling model is corrected according to the real-time parameter data, and the process returns to the step of quantitatively analyzing the coupling relationship between the initial parameters according to the multi-parameter coupling model until the parameter fluctuation index is less than the preset fluctuation threshold, and the initial parameter optimization scheme at this time is used as a candidate optimization scheme.

8. The intelligent stamping control system for a high-precision CNC punch press according to claim 7, characterized in that: The building blocks include: A first constructing unit is configured to construct a parameter sample set according to the multiple initial parameter values, wherein the parameter sample set includes multiple sample points; The second construction unit is used to perform dimensionality reduction processing on the parameter sample set by using a principal component analysis method and extract a plurality of main components, and construct a multi-parameter coupling initial model based on the plurality of main components; A division unit, configured to divide the parameter sample set into training parameter samples and testing parameter samples; The training unit is used to train and test the multi-parameter coupling initial model according to the training parameter samples and the test parameter samples until the mean square error of the multi-parameter coupling initial model is less than a preset error threshold, thereby obtaining the multi-parameter coupling initial model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images