Intelligent mold production control method, device, equipment and storage medium

Abstract

The present invention relates to the technical field of knife die production control, and discloses a production control method, device, equipment and storage medium for intelligent molds. The method comprises: performing deep neural network analysis on material property data of the knife die to obtain a first cutting parameter set, and creating a first cutting path scheme for the knife die; performing finite element analysis and real-time temperature monitoring on the cutting process of the knife die based on the first cutting parameter set and the first cutting path scheme to obtain cutting temperature distribution data; predicting the thermal expansion behavior of the knife die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy; performing deep reinforcement learning analysis and multi-objective optimization on the first cutting parameter set and the first cutting path scheme based on the thermal expansion compensation strategy to obtain a second cutting parameter set and a second cutting path scheme. The present invention can adaptively optimize cutting parameters and adjust cutting paths in real time, thereby realizing a high-precision and high-efficiency intelligent mold production process.

Description

Technical Field

[0001] The present invention relates to the field of knife die production control technology, and in particular to a production control method, device, equipment and storage medium for an intelligent mold. Background Art

[0002] Traditional mold production methods have been unable to meet the growing precision and complexity requirements. Especially in the production process of laser cutting molds, issues such as material thermal expansion, cutting parameter optimization, and path planning have always been key factors restricting production efficiency and quality.

[0003] At present, most mold production control methods still rely on empirical parameters and simplified models, which cannot fully consider the influence of material properties, thermal-mechanical coupling effects, and dynamic processing environments. This leads to problems such as insufficient cutting accuracy, excessive heat-affected zones, and low processing efficiency. In addition, existing control methods are often static, single-objective optimization processes, which are difficult to cope with complex multi-objective optimization requirements and dynamically changing processing environments. Summary of the invention

[0004] The present invention provides a production control method, device, equipment and storage medium for an intelligent mold. The present invention can adaptively optimize cutting parameters and adjust cutting paths in real time, thereby realizing a high-precision and high-efficiency intelligent mold production process.

[0005] In a first aspect, the present invention provides a production control method for an intelligent mold, the production control method for an intelligent mold comprising:

[0006] Performing a deep neural network analysis on the material property data of the cutting die to obtain a first cutting parameter set, and creating a first cutting path plan for the cutting die;

[0007] Based on the first cutting parameter set and the first cutting path scheme, a finite element analysis and real-time temperature monitoring are performed on the cutting process of the cutting die to obtain cutting temperature distribution data;

[0008] Predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy;

[0009] Based on the thermal expansion compensation strategy, deep reinforcement learning analysis and multi-objective optimization are performed on the first cutting parameter set and the first cutting path solution to obtain a second cutting parameter set and a second cutting path solution.

[0010] In a second aspect, the present invention provides a production control device for an intelligent mold, the production control device for an intelligent mold comprising:

[0011] A creation module is used to perform a deep neural network analysis on the material characteristic data of the cutting die, obtain a first cutting parameter set, and create a first cutting path plan for the cutting die;

[0012] An analysis module, configured to perform finite element analysis and real-time temperature monitoring on the cutting process of the cutting die based on the first cutting parameter set and the first cutting path scheme, and obtain cutting temperature distribution data;

[0013] A prediction module, used for predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy;

[0014] An optimization module is used to perform deep reinforcement learning analysis and multi-objective optimization on the first cutting parameter set and the first cutting path scheme based on the thermal expansion compensation strategy to obtain a second cutting parameter set and a second cutting path scheme.

[0015] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned intelligent mold production control method.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned intelligent mold production control method.

[0017] In the technical solution provided by the present invention, by performing deep neural network analysis on the characteristic data of the die material, the cutting parameters can be predicted more accurately, the accuracy of the initial parameter setting can be improved, and the number of trial cuts can be reduced. Finite element analysis and real-time temperature monitoring are performed in combination with the first cutting parameter set and the path scheme, so that the temperature distribution during the cutting process can be accurately simulated, and the thermal expansion behavior is predicted based on the cutting temperature distribution data, and a thermal expansion compensation strategy is generated, which can effectively reduce the processing error caused by thermal deformation. Through deep reinforcement learning analysis and multi-objective optimization, multiple performance indicators such as cutting accuracy, efficiency and minimization of thermal effects can be considered at the same time to achieve more comprehensive optimization. The reinforcement learning environment model and the action value estimation network are used to realize the adaptive adjustment of the cutting parameters and improve the robustness of the processing process. Multi-objective optimization and cubic spline interpolation technology are used to generate a continuous and smooth cutting trajectory, reduce vibration and stress concentration during processing, and improve cutting quality. Collision detection and obstacle avoidance optimization are performed on the cutting trajectory to ensure the safety and reliability of the processing process, thereby realizing a high-precision and high-efficiency intelligent mold production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0019] Figure 1 A schematic diagram of the steps of the production control method of the intelligent mold in an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of the structure of the production control device of the intelligent mold in the embodiment of the present invention. DETAILED DESCRIPTION

[0021] Embodiments of the present invention provide a production control method, device, equipment and storage medium for an intelligent mold. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the production control method of the intelligent mold in the embodiment of the present invention includes:

[0023] Step S1, performing deep neural network analysis on the material characteristic data of the cutting die to obtain a first cutting parameter set, and creating a first cutting path plan for the cutting die;

[0024] It is understandable that the execution subject of the present invention may be a production control device of an intelligent mold, or may be a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0025] Specifically, the material characteristic data of the die are feature extracted and normalized, the numerical ranges of various material characteristics are standardized, and a unified standardized material characteristic vector is obtained. The standardized material characteristic vector is input into a preset deep neural network for forward propagation calculation. Through the deep learning algorithm, the neural network can extract important cutting speed parameters and laser power parameters from the input data. The cutting speed parameters and laser power parameters are cross-validated and error analyzed to effectively test and correct the potential errors in the parameters and obtain a more reliable first cutting parameter set. According to the first cutting parameter set, the geometric shape data of the die is discretized to convert the complex geometric shape into a grid model that can be processed by the computer. The grid model can provide a detailed cutting path framework. The grid model is searched for a path, and an initial cutting path is preliminarily determined through a computer algorithm. According to the initial cutting path, the cutting sequence is optimized by a genetic algorithm. By simulating the process of natural selection and genetic variation, the genetic algorithm can find the optimal solution in a large number of possible cutting paths and obtain an optimized cutting sequence. In order to ensure the local optimality of the optimization result, the optimized cutting sequence is locally searched, and the optimization result is further fine-tuned and corrected to finally obtain the first cutting path solution.

[0026] The standardized material feature vector is input to the input layer of the deep neural network, and the standardized material feature vector is mapped to a 256-dimensional feature space to obtain the initial feature representation. The initial feature representation is input to the first convolutional layer, and processed by using 64 3x3 convolution kernels and ReLU activation function to generate a low-level feature map. The low-level feature map retains some primary characteristics of the input features. The low-level feature map is subjected to a maximum pooling operation, and a 2x2 pooling window and a downsampling method with a step size of 2 are used to compress the feature map to obtain a compressed feature map. The compressed feature map is input to the second convolutional layer, in which 128 3x3 convolution kernels and LeakyReLU activation function are used to process the features to obtain a mid-level feature map. The mid-level feature map contains more complex information about the input features. The mid-level feature map is batch normalized, and an exponential moving average method with a momentum of 0.99 is used to obtain a standardized feature map to reduce the deviation and fluctuation of the features. In order to improve the expressiveness of the feature map, the attention mechanism is used to process the standardized feature map, and the self-attention module is applied to calculate the correlation between features to obtain the enhanced feature map. The enhanced feature map is subjected to global average pooling to compress the spatial dimension of each feature channel into a single value to form a target feature vector. The target feature vector is input into the fully connected layer and processed using 512 neurons and the Tanh activation function to generate a high-level feature representation. In order to prevent overfitting, the high-level feature representation is Dropout regularized, and a dropout rate of 0.5 is used to randomly mask a part of the neurons to obtain the regularized features. The regularized features are input into the output layer, and the cutting speed parameters and laser power parameters are output respectively through two parallel fully connected layers.

[0027] Step S2, performing finite element analysis and real-time temperature monitoring on the cutting process of the die cutter based on the first cutting parameter set and the first cutting path scheme to obtain cutting temperature distribution data;

[0028] Specifically, according to the first cutting parameter set and the first cutting path scheme, the three-dimensional geometric modeling of the knife die is performed to obtain an accurate three-dimensional model of the knife die. The accurate three-dimensional model of the knife die is meshed, and the complex geometric shape is divided into a limited number of small units to obtain a finite element mesh with local encryption. The locally encrypted mesh can provide higher calculation accuracy in key areas. According to the laser power parameter in the first cutting parameter set, the laser heat source is mathematically modeled to obtain a time-varying heat source function. The time-varying heat source function can describe the heat input of the laser during the cutting process. Based on the finite element mesh and the time-varying heat source function, a heat conduction partial differential equation is constructed, and it is discretized using the Crank-Nicolson implicit difference format to obtain a group of heat conduction equations. The group of heat conduction equations can describe the transfer and distribution of heat inside the knife die. The group of heat conduction equations is solved to obtain the temperature field distribution of each time step. According to the first cutting path scheme, the temperature field distribution is dynamically updated to obtain the spatiotemporal temperature field evolution data. Thermal stress analysis is performed on the spatiotemporal temperature field evolution data to obtain the stress distribution during the cutting process. The stress distribution data can reflect the stress changes of the cutting die under the action of heat. Based on the stress distribution, the thermal deformation of the cutting die is calculated to obtain the deformation field during the cutting process. The deformation field data can reflect the deformation of the cutting die under the action of heat. The deformation field is monitored in real time to obtain discrete temperature monitoring data. The discrete temperature monitoring data is interpolated and reconstructed to reconstruct the discrete data points into a continuous temperature distribution to obtain the cutting temperature distribution data.

[0029] Step S3, predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy;

[0030] Specifically, the cutting temperature distribution data is decomposed in time and space to analyze the main modes and time coefficients of the temperature field. By decomposing the complex temperature distribution data into several basic modes, the change characteristics of the temperature field can be simplified and abstracted, which is convenient for subsequent processing. According to the main modes and time coefficients, a rapid heat conduction prediction model of the die is constructed. The model can predict the temperature change of the die in a short time, greatly improving the calculation efficiency. The parameter sensitivity analysis of the rapid heat conduction prediction model is carried out. By analyzing the influence of each parameter on the heat conduction result, the key parameter set affecting thermal expansion is obtained. The key parameter set reflects which factors play a major role in the thermal expansion process. Based on the key parameter set, the thermal expansion probability distribution function of the die is constructed. The probability distribution function can describe the possibility of thermal expansion of the die under different conditions. The thermal expansion probability distribution function is simulated and analyzed, and a representative thermal expansion scenario set is obtained through a large number of simulation experiments. The representative scenario set covers the thermal expansion of the die under various possible conditions. According to the representative thermal expansion scenario set, the thermal expansion compensation strategy is optimized. Through the optimization process, the best compensation method under different thermal expansion conditions is found to ensure that the die maintains accuracy and stability during the actual cutting process. Generate thermal expansion compensation strategies, which are applied in the actual production control process of the die cutter to adjust the cutting parameters in real time to compensate for the errors caused by thermal expansion.

[0031] Step S4: Based on the thermal expansion compensation strategy, deep reinforcement learning analysis and multi-objective optimization are performed on the first cutting parameter set and the first cutting path plan to obtain a second cutting parameter set and a second cutting path plan.

[0032] Specifically, the thermal expansion compensation strategy is parameterized and encoded, and the compensation strategy is converted into a state space representation that can be understood and processed by the machine learning algorithm. The state space representation can fully describe the effect and performance of the thermal expansion compensation strategy under different conditions. According to the state space representation and the first cutting parameter set, a reinforcement learning environment model and a state transition function of the die cutting process are constructed. The reinforcement learning environment model is used to simulate the changes of various states in the actual cutting process, while the state transition function describes how the state is transferred from one state to another under different operations. Based on the model and function, an action value estimation network is created and trained. Through a large number of simulations and iterations, the Q function in the network is optimized so that it can accurately evaluate the value of different actions in different states. According to the optimized Q function, a cutting parameter adjustment strategy is generated to obtain the second cutting parameter set. At the same time, multi-objective modeling is performed on the first cutting path scheme, and an optimization target that comprehensively considers multiple factors is established by setting a multi-objective optimization function for minimizing the path length and the heat affected zone. Based on the multi-objective optimization function, multiple candidate cutting path schemes are calculated, and the candidate cutting path schemes are respectively interpolated by cubic spline path to obtain multiple continuous and smooth cutting trajectories. Cubic spline interpolation can make the cutting path smoother and more continuous, reducing mutations and unstable factors in the cutting process. Collision detection and optimization solutions are performed on multiple continuous and smooth cutting trajectories to ensure that the cutting path will not collide with other parts in actual applications and is the optimal cutting path. The second cutting path solution is obtained.

[0033] The multi-objective optimization function is weighted vectorized to obtain a set of uniformly distributed weight vector sets. The weight vector set can evenly distribute the weights of each objective, so that each objective in the multi-objective optimization process can be fully considered. According to the weight vector set, the multi-objective optimization problem is decomposed, and the weighted sum method is used to convert the multi-objective into a single objective to obtain m sub-problems. Each sub-problem combines multiple objectives into one objective through the weighted sum method, making the problem solution simpler and clearer. Particle swarm optimization is performed on each sub-problem to effectively search for the global optimal solution and obtain m groups of optimal solutions, each of which represents a candidate cutting path solution. The m groups of optimal solutions are non-dominated sorted to screen out the target optimal solution set. Each target optimal solution represents a solution that performs well in multi-objective optimization. The path representation conversion is performed on each solution in the target optimal solution set, and the discrete point sequence is converted into a parameterized curve representation to obtain a control point set. The control point set can effectively describe the shape and direction of the cutting path. Based on the control point set, the cubic spline interpolation function is constructed using natural boundary conditions to obtain piecewise polynomial functions, which can smoothly connect various control points to form a continuous curve. The coefficients of the piecewise polynomial function are solved to obtain the coefficient matrix of the spline function. According to the coefficient matrix, each candidate cutting path scheme is sampled with high density to generate a cutting trajectory point set. The curvature analysis of the cutting trajectory point set is performed, and the curvature value of each point is calculated to obtain the curvature distribution function. Curvature analysis can reflect the curvature of the cutting path. Based on the curvature distribution function, the cutting trajectory is adaptively sampled, and more sampling is performed at key curvature change points to ensure the smoothness and accuracy of the cutting trajectory, and multiple continuous and smooth cutting trajectories are obtained.

[0034] In the embodiment of the present invention, by performing deep neural network analysis on the characteristic data of the die material, the cutting parameters can be predicted more accurately, the accuracy of the initial parameter setting can be improved, and the number of trial cuts can be reduced. Finite element analysis and real-time temperature monitoring are performed in combination with the first cutting parameter set and the path scheme, and the temperature distribution during the cutting process can be accurately simulated. Based on the cutting temperature distribution data, the thermal expansion behavior is predicted, and a thermal expansion compensation strategy is generated, which can effectively reduce the processing error caused by thermal deformation. Through deep reinforcement learning analysis and multi-objective optimization, multiple performance indicators such as cutting accuracy, efficiency and minimization of thermal effects can be considered at the same time to achieve more comprehensive optimization. The reinforcement learning environment model and the action value estimation network are used to realize the adaptive adjustment of the cutting parameters and improve the robustness of the processing process. Multi-objective optimization and cubic spline interpolation technology are used to generate a continuous and smooth cutting trajectory, reduce vibration and stress concentration during processing, and improve cutting quality. Collision detection and obstacle avoidance optimization are performed on the cutting trajectory to ensure the safety and reliability of the processing process, thereby realizing a high-precision and high-efficiency intelligent mold production process.

[0035] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0036] Perform feature extraction and normalization on the material characteristic data of the cutting die to obtain a standardized material feature vector;

[0037] The standardized material feature vector is input into the preset deep neural network for forward propagation calculation to obtain the cutting speed parameters and laser power parameters;

[0038] Cross-validation and error analysis are performed on cutting speed parameters and laser power parameters to obtain a first cutting parameter set;

[0039] Discretizing the geometric shape data of the cutting die according to the first cutting parameter set to obtain a grid model of the cutting die, and performing path search on the grid model to obtain an initial cutting path;

[0040] The cutting sequence is optimized by genetic algorithm according to the initial cutting path to obtain an optimized cutting sequence, and a local search is performed on the optimized cutting sequence to obtain a first cutting path solution.

[0041] Specifically, the material property data of the cutting die are collected, including parameters such as the density, hardness, thermal conductivity, and thermal expansion coefficient of the material. Feature extraction is performed on these data to extract key indicators that can reflect the material properties. For example, thermal conductivity and thermal expansion coefficient may have an important influence on heat conduction and deformation during the cutting process. The extracted features are normalized, and data of different dimensions are converted to a unified scale to obtain standardized material feature vectors. These vectors are used as input data and enter the preset deep neural network for forward propagation calculation. In the deep neural network, the standardized material feature vector is input into the input layer of the network, and the cutting speed parameters and laser power parameters are obtained through layer-by-layer calculation of the multi-layer network. The forward propagation process of the deep neural network can be expressed by the following formula:

[0042] ;

[0043] in, Indicates the output value, represents the weight matrix, represents the input vector, represents the bias vector, Represents the activation function. Through the trained deep neural network, accurate cutting speed parameters and laser power parameters are obtained. Cross-validation and error analysis are performed on the cutting speed parameters and laser power parameters. Cross-validation evaluates the stability and generalization ability of the model by splitting the data set and training the model multiple times. Error analysis evaluates the accuracy of the model by calculating the error between the predicted value and the actual value. For example, the formula for using mean square error to measure the error size is as follows:

[0044] ;

[0045] in, represents the number of samples, Indicates the actual value, Represents the predicted value. Through cross validation and error analysis, an accurate first cutting parameter set is obtained. According to the first cutting parameter set, the geometric shape data of the die is discretized, and the continuous geometric shape is converted into a discrete grid model. The grid model divides the geometric shape into multiple small units, each unit consists of nodes and edges, and can describe the local characteristics of the geometric shape. Based on the grid model, a path search is performed to find the initial cutting path in the cutting process. The purpose of the path search is to find a path that can efficiently cover all the areas to be cut. The cutting sequence is optimized. Genetic algorithm is a commonly used optimization algorithm that finds the optimal solution by simulating the process of natural selection and genetic mutation. The basic steps of genetic algorithm include selection, crossover and mutation. Select several paths from the initial cutting path as the parent generation, generate the child path through the crossover operation, and introduce random changes through the mutation operation to increase the diversity of the path. After multiple generations of evolution, the genetic algorithm can find the optimized cutting sequence. Local search is performed on the optimized cutting sequence. On the basis of the global optimal solution, the quality of the path is improved by fine-tuning the local path. For example, some nodes in the cutting path are adjusted to optimize the smoothness and continuity of the path, thereby obtaining a final first cutting path solution.

[0046] In a specific embodiment, the process of executing the step of inputting the standardized material feature vector into a preset deep neural network for forward propagation calculation to obtain the cutting speed parameter and the laser power parameter may specifically include the following steps:

[0047] Input the standardized material feature vector into the input layer of the deep neural network, map the standardized material feature vector to a 256-dimensional feature space, and obtain an initial feature representation;

[0048] The initial feature representation is input into the first convolutional layer, processed using 64 3x3 convolution kernels and ReLU activation function to obtain a low-level feature map, the low-level feature map is subjected to maximum pooling operation, and a 2x2 pooling window and a step size of 2 are used for downsampling to obtain a compressed feature map;

[0049] The compressed feature map is input into the second convolutional layer and processed using 128 3x3 convolution kernels and the LeakyReLU activation function to obtain an intermediate feature map. The intermediate feature map is batch normalized and an exponential moving average with a momentum of 0.99 is used to obtain a standardized feature map.

[0050] The standardized feature map is processed by the attention mechanism, and the correlation between features is calculated using the self-attention module to obtain the enhanced feature map. The enhanced feature map is globally averaged pooled to compress the spatial dimension of each feature channel into a single value to obtain the target feature vector.

[0051] The target feature vector is input into the fully connected layer and processed using 512 neurons and the Tanh activation function to obtain a high-level feature representation;

[0052] Dropout regularization is performed on the high-level feature representation, and neurons are randomly masked using a dropout rate of 0.5 to obtain regularized features. The regularized features are input into the output layer, and two parallel fully connected layers are used to output the cutting speed parameters and laser power parameters respectively.

[0053] Specifically, the preprocessed standardized material feature vector is input into the input layer of the neural network, and the vector is mapped to a 256-dimensional feature space through a linear transformation. This process can be expressed as:

[0054] ;

[0055] in, represents the initial feature representation, is the weight matrix, is the normalized material eigenvector, is the bias vector. Through the transformation, the material feature data is mapped to a high-dimensional feature space, allowing the network to capture richer information. The initial feature representation is input into the first convolutional layer. In this layer, 64 3x3 convolution kernels are used for convolution operation, and nonlinear transformation is performed through the ReLU activation function to obtain a low-level feature map. The role of the ReLU activation function is to introduce nonlinear characteristics. The formula is expressed as:

[0056] ;

[0057] in, is the input value, is the output value. Through the ReLU activation function, the network can learn the nonlinear features in the input data. The low-level feature map is subjected to the maximum pooling operation, and a 2x2 pooling window and a step size of 2 are used for downsampling to obtain a compressed feature map. The role of maximum pooling is to reduce the size of the feature map while retaining important feature information. The formula is expressed as:

[0058] ;

[0059] in, is the value after pooling, is the value in the pooling window. The compressed feature map is input into the second convolutional layer and processed using 128 3x3 convolution kernels and the LeakyReLU activation function to obtain an intermediate feature map. The role of the LeakyReLU activation function is to introduce a small negative slope to solve the "death" problem of the ReLU activation function. The formula is expressed as:

[0060] ;

[0061] in, is the input value, is the output value. Perform batch normalization on the intermediate feature map, use the exponential moving average with a momentum of 0.99 to standardize the feature map, and obtain the standardized feature map. The role of batch normalization is to speed up the training process and improve the stability of the model. The formula is expressed as:

[0062] ;

[0063] in, is the normalized feature, is the mean, is the variance, is a small constant used to prevent the denominator from being zero. The attention mechanism is applied to the standardized feature map. The self-attention module is used to calculate the correlation between features and obtain an enhanced feature map. The role of the self-attention mechanism is to enable the model to focus on the important parts of the input features. The formula is expressed as:

[0064] ;

[0065] in, is the query matrix, is the key matrix, is the value matrix, is the dimension of the key vector. Through the attention mechanism, the model can dynamically adjust the focus on different features. Global average pooling is performed on the enhanced feature map to compress the spatial dimension of each feature channel into a single value to obtain the target feature vector. The role of global average pooling is to summarize the information of each feature channel into one value, and the formula is expressed as:

[0066] ;

[0067] in, is the value after pooling, is the value in the feature channel, is the length of the feature channel. The target feature vector is input into the fully connected layer and processed using 512 neurons and the Tanh activation function to obtain a high-level feature representation. The function of the Tanh activation function is to limit the output value to the range of [-1, 1], and the formula is expressed as:

[0068] ;

[0069] in, is the input value, is the output value. In order to prevent overfitting, Dropout regularization is performed on the high-level feature representation, and neurons are randomly shielded with a dropout rate of 0.5 to obtain regularized features. The role of Dropout is to prevent the model from overfitting and make the model more robust by randomly discarding some neurons. The regularized features are input into the output layer, and two parallel fully connected layers are used to output the cutting speed parameters and laser power parameters respectively. These two parameters are key parameters in the cutting process and directly affect the cutting effect and quality.

[0070] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0071] According to the first cutting parameter set and the first cutting path scheme, three-dimensional geometric modeling is performed on the cutting die to obtain an accurate three-dimensional model of the cutting die, and the accurate three-dimensional model of the cutting die is meshed to obtain a finite element mesh with local encryption;

[0072] According to the laser power parameter in the first cutting parameter set, the laser heat source is mathematically modeled to obtain a time-varying heat source function, and based on the finite element mesh and the time-varying heat source function, a heat conduction partial differential equation is constructed, which is discretized using the Crank-Nicolson implicit difference format to obtain a group of heat conduction equations;

[0073] Solve the heat conduction equations to obtain the temperature field distribution of each time step, and dynamically update the temperature field distribution according to the first cutting path scheme to obtain the spatiotemporal temperature field evolution data;

[0074] Conduct thermal stress analysis on the temporal and spatial temperature field evolution data to obtain the stress distribution during the cutting process. Based on the stress distribution, calculate the thermal deformation of the die to obtain the deformation field during the cutting process.

[0075] The deformation field is monitored in real time to obtain discrete temperature monitoring data, and the discrete temperature monitoring data is interpolated and reconstructed to obtain cutting temperature distribution data.

[0076] Specifically, by using CAD software or other three-dimensional modeling tools, the three-dimensional geometric modeling of the cutting die is performed to reflect its shape and structure. The three-dimensional model is meshed, and the complex geometric model is decomposed into smaller finite units to facilitate finite element analysis. In order to improve the calculation accuracy, the key areas are locally encrypted. For example, for the heat-affected zone or the stress concentration zone, a smaller unit grid is used, and a larger grid unit is used in other areas, which not only ensures the calculation accuracy but also optimizes the use of computing resources. According to the laser power parameter in the first cutting parameter set, the laser heat source is mathematically modeled to obtain a time-varying heat source function. The time-varying heat source function can accurately describe the heat input of the laser that changes with time during the cutting process. The common heat source model can be described by a Gaussian distribution, and its formula is:

[0077] ;

[0078] in, Indicates the distance from the laser center Place, time The heat source intensity at is the laser power, is the laser spot radius. Based on the finite element grid and the time-varying heat source function, the heat conduction partial differential equation is constructed. The heat conduction equation describes the heat transfer process inside the die, and is discretized using the Crank-Nicolson implicit difference format. The Crank-Nicolson method is a stable method for numerically solving partial differential equations, and its discretization formula is:

[0079] ;

[0080] in, and Represents the time step and The temperature distribution at is the time step, is the thermal diffusion coefficient. By solving the heat conduction equations, the temperature field distribution of each time step is obtained. According to the first cutting path scheme, the temperature field distribution is dynamically updated to obtain the spatiotemporal temperature field evolution data. The spatiotemporal temperature field evolution data can fully describe the temperature changes during the cutting process, which is of great significance for predicting and controlling the thermal deformation of the die. Thermal stress analysis is performed on the spatiotemporal temperature field evolution data to obtain the stress distribution during the cutting process. The purpose of thermal stress analysis is to determine the stress distribution caused by the temperature gradient. The formula is as follows:

[0081] ;

[0082] in, For thermal stress, is the elastic modulus, is the coefficient of thermal expansion, is the temperature change. Based on the stress distribution, the thermal deformation of the die is calculated to obtain the deformation field during the cutting process. The purpose of the thermal deformation calculation is to predict the deformation of the die under the action of heat, so as to guide the actual cutting operation and ensure the accuracy of the final product. In order to ensure the real-time and accuracy of the data, the deformation field is monitored in real time to obtain discrete temperature monitoring data. Real-time monitoring can use sensing equipment such as thermocouples and infrared thermal imagers to obtain temperature data of each key point in the cutting process. The discrete temperature monitoring data is interpolated and reconstructed to obtain continuous cutting temperature distribution data. The interpolation and reconstruction method can use linear interpolation, spline interpolation and other technologies to connect discrete data points into a continuous temperature distribution curve through the interpolation function to provide a comprehensive temperature field description.

[0083] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0084] The cutting temperature distribution data is decomposed in time and space to obtain the main modes and time coefficients of the temperature field, and a rapid heat conduction prediction model of the cutting die is constructed based on the main modes and time coefficients;

[0085] The parameter sensitivity analysis of the rapid heat conduction prediction model was carried out to obtain the key parameter set that affects thermal expansion, and based on the key parameter set, the thermal expansion probability distribution function of the die was constructed;

[0086] The thermal expansion probability distribution function is simulated and analyzed to obtain a representative thermal expansion scenario set, and the thermal expansion compensation strategy is optimized based on the representative thermal expansion scenario set to generate a thermal expansion compensation strategy.

[0087] Specifically, the cutting temperature distribution data is decomposed in time and space to extract the main modes and time coefficients of the temperature field. The complex time and space temperature field data is decomposed into several main modes, which can capture the main characteristics of the temperature field change, and the time coefficient is used to describe the evolution of these modes over time. Commonly used methods include singular value decomposition and principal component analysis. These methods can effectively reduce the dimension and convert high-dimensional data into low-dimensional representation while retaining the main information. For example, for a temperature field distribution matrix , through singular value decomposition, we get the following form:

[0088] ;

[0089] in, are singular values, is the spatial mode, is the time coefficient, is the number of main modes. According to the main modes and time coefficients, a rapid heat conduction prediction model for the die is constructed. The model uses a few main modes and time coefficients to quickly predict the temperature field distribution at different time points. In this way, the amount of calculation is reduced while maintaining a high prediction accuracy. A parameter sensitivity analysis is performed on the rapid heat conduction prediction model to identify the key parameter sets that affect thermal expansion. The purpose of parameter sensitivity analysis is to determine which parameters have the greatest impact on heat conduction and thermal expansion, so that these key parameters can be optimized. Commonly used methods include local sensitivity analysis and global sensitivity analysis. Local sensitivity analysis measures the impact of each parameter by calculating its partial derivative, while global sensitivity analysis uses statistical methods to evaluate the overall impact of parameter changes on the output results. For example, consider a parameter The influence on the temperature field, its sensitivity coefficient can be expressed as:

[0090] ;

[0091] By calculating the sensitivity coefficients of all parameters, the key parameter set that affects thermal expansion is obtained. Based on the key parameter set, the thermal expansion probability distribution function of the die is constructed to describe the thermal expansion behavior of the die under different parameter conditions. The thermal expansion probability distribution function can be established by statistical analysis methods, such as using Monte Carlo simulation. Monte Carlo simulation generates a series of possible thermal expansion scenarios through a large number of random samplings to construct the probability distribution function of thermal expansion. The thermal expansion probability distribution function is simulated and analyzed to obtain a representative thermal expansion scenario set. The representative thermal expansion scenario set covers the possible thermal expansion of the die under different conditions. For example, using the thermal expansion probability distribution function, several typical thermal expansion scenarios are generated, each of which corresponds to a set of specific parameter conditions and thermal expansion results. The thermal expansion compensation strategy is optimized based on these representative thermal expansion scenario sets to generate a thermal expansion compensation strategy. The purpose of the compensation strategy is to compensate for the error caused by thermal expansion by adjusting the cutting parameters during the actual cutting process. For example, by adjusting parameters such as laser power and cutting speed, the distribution of the temperature field is controlled to reduce the impact of thermal expansion. The optimization of the compensation strategy can be achieved through a multi-objective optimization method, the goal of which is to minimize the thermal expansion error while ensuring cutting efficiency.

[0092] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0093] Parameterizing and encoding the thermal expansion compensation strategy to obtain a state space representation of the thermal expansion compensation strategy;

[0094] According to the state space representation and the first cutting parameter set, a reinforcement learning environment model and a state transfer function of the die cutting process are constructed;

[0095] An action value estimation network is created based on the state transfer function, and the action value estimation network is trained to obtain an optimized Q function;

[0096] According to the optimized Q function, a cutting parameter adjustment strategy is generated to obtain a second cutting parameter set;

[0097] Perform multi-objective modeling on the first cutting path scheme, and set a multi-objective optimization function for minimizing the path length and the heat-affected zone;

[0098] Calculate multiple candidate cutting path solutions based on a multi-objective optimization function, and perform cubic spline interpolation on the multiple candidate cutting path solutions to obtain multiple continuous and smooth cutting trajectories;

[0099] Collision detection and optimization are performed on multiple continuous and smooth cutting trajectories to obtain a second cutting path solution.

[0100] Specifically, each adjustment parameter in the compensation strategy is encoded to form a multi-dimensional state vector. The state vector can fully describe all aspects of the thermal expansion compensation strategy. Assuming that the thermal expansion compensation strategy includes multiple parameters such as laser power adjustment, cutting speed adjustment, and path correction, these parameters are normalized and formed into a state vector ,in represents the laser power adjustment parameter, Indicates the cutting speed adjustment parameter, Represents the path correction parameter. According to the state space representation and the first cutting parameter set, the reinforcement learning environment model and the state transfer function of the die cutting process are constructed. In the reinforcement learning environment model, the state transfer function describes how the state of the die is transferred from one state to another under specific cutting parameters and compensation strategies. This state transfer can be modeled using a Markov decision process, as follows:

[0101] ;

[0102] in, Indicates the next state, Indicates the current state. Indicates the current action (i.e. the current cutting parameter adjustment strategy), is the state transition function. Based on the state transition function, an action value estimation network (Q network) is created and trained. The purpose of the action value estimation network is to learn the value of each state-action pair, which represents the long-term benefit of taking a specific action in a specific state. The training of the Q network usually uses a deep Q learning algorithm, and the goal is to minimize the following loss function

[0103] ;

[0104] in, Indicates immediate reward, is the discount factor, and Respectively represent the parameters of the current Q network and the target Q network. After training, the optimized Q function is obtained. According to the optimized Q function, a cutting parameter adjustment strategy is generated to obtain a second cutting parameter set. This strategy adjusts the cutting parameters by selecting the action with the highest Q value in each state. The formula is:

[0105] ;

[0106] At the same time, multi-objective modeling is performed on the first cutting path scheme, and a multi-objective optimization function is set to minimize the path length and the heat-affected zone. The multi-objective optimization function aims to optimize the length of the cutting path and the size of the heat-affected zone at the same time, and the formula is as follows:

[0107] ;

[0108] in, Indicates The length of the segment path, Indicates The heat affected zone corresponding to the segment path. Based on the multi-objective optimization function, multiple candidate cutting path solutions are calculated. The candidate cutting path solutions are interpolated with cubic spline to obtain multiple continuous and smooth cutting trajectories. Cubic spline interpolation generates a smooth curve by constructing a cubic polynomial between every two data points. The formula is as follows:

[0109] ;

[0110] in, is the spline function, is the coefficient to be determined, The node is a node. The continuous and smooth cutting path is collided and optimized. The purpose of collision detection is to ensure that the cutting path will not collide with the die or other equipment in actual operation. The optimization solution makes the cutting process more efficient and safe by adjusting the path and parameters.

[0111] In a specific embodiment, the execution step calculates multiple candidate cutting path solutions based on the multi-objective optimization function, and performs path cubic spline interpolation on the multiple candidate cutting path solutions respectively to obtain multiple continuous and smooth cutting trajectories, which may specifically include the following steps:

[0112] Perform weight vectorization on the multi-objective optimization function to obtain a set of uniformly distributed weight vectors;

[0113] According to the weight vector set, the multi-objective optimization problem is decomposed, and the weighted sum method is used to transform the multi-objective into a single objective, and m sub-problems are obtained;

[0114] Particle swarm optimization is performed on each sub-problem to obtain m groups of optimal solutions, each group of optimal solutions represents a candidate cutting path solution;

[0115] Perform non-dominated sorting on the m groups of optimal solutions to obtain the target optimal solution set, and perform path representation conversion on each solution in the target optimal solution set, convert the discrete point sequence into parameterized curve representation, and obtain the control point set;

[0116] Based on the set of control points, a cubic spline interpolation function is constructed using natural boundary conditions to obtain a piecewise polynomial function;

[0117] Solve the coefficients of the piecewise polynomial function to obtain the coefficient matrix of the spline function, and perform high-density sampling on each candidate cutting path scheme according to the coefficient matrix to obtain a cutting trajectory point set;

[0118] The curvature analysis is performed on the cutting trajectory point set, and the curvature value of each point is calculated to obtain the curvature distribution function. The cutting trajectory is adaptively sampled based on the curvature distribution function to obtain multiple continuous smooth cutting trajectories.

[0119] Specifically, multiple optimization objectives are weighted to obtain a set of uniformly distributed weight vectors. Different weights are assigned to each objective to reflect its importance. The formula is as follows:

[0120] ;

[0121] in, Indicates The weight value of the weight vector, is the total number of weight vectors. In this way, a set of weight vectors is uniformly generated. According to the weight vector set, the multi-objective optimization problem is decomposed. Each weight vector corresponds to a sub-problem, and the weighted sum method is used to transform the multi-objective optimization problem into a single-objective optimization problem. The formula is as follows:

[0122] ;

[0123] in, represents the weighted single objective function, Indicates The objective function, are the corresponding weights. In this way, the multi-objective optimization problem is decomposed into Sub-problems. Particle swarm optimization is performed on each sub-problem to obtain Particle swarm optimization is a bionic optimization algorithm that simulates the foraging behavior of bird flocks to find the optimal solution to the problem. Each particle represents a possible solution, and by continuously updating the speed and position of the particle, it gradually approaches the optimal solution. The particle position update formula is as follows:

[0124] ;

[0125] in, Indicates Particles in The speed of time, Indicates Particles in The location at the moment, is the inertia weight, and is the learning factor, and is a random number, For the The historical optimal position of a particle, is the global optimal position. Through particle swarm optimization, we get The optimal solutions of each group are grouped, and each group of optimal solutions represents a candidate cutting path solution. The optimal solutions are non-dominated sorted, and the solutions that perform well on all objectives are screened out to obtain the target optimal solution set. Non-dominated sorting divides all solutions into different levels by comparing the pros and cons of solutions, and the optimal solution set is at the highest level. Each solution in the target optimal solution set is converted into a path representation, and the discrete point sequence is converted into a parameterized curve representation to obtain a control point set. Based on the control point set, the cubic spline interpolation function is constructed using natural boundary conditions. The formula is as follows:

[0126] ;

[0127] in, is the spline function, is the coefficient to be determined, is a node. By constructing a cubic spline interpolation function, a piecewise polynomial function is obtained. The coefficients of the piecewise polynomial function are solved to obtain the coefficient matrix of the spline function. According to the coefficient matrix, each candidate cutting path scheme is sampled with high density to generate a cutting trajectory point set. The high-density sampling points can describe every detail of the cutting path. The curvature analysis is performed on the cutting trajectory point set, and the curvature value of each point is calculated to obtain the curvature distribution function. Curvature analysis can reflect the degree of curvature of the cutting path. Based on the curvature distribution function, the cutting trajectory is adaptively sampled, and more sampling is performed at key curvature change points to ensure the smoothness and accuracy of the cutting trajectory, and multiple continuous and smooth cutting trajectories are obtained.

[0128] The above describes the production control method of the intelligent mold in the embodiment of the present invention. The following describes the production control device of the intelligent mold in the embodiment of the present invention. Figure 2 , an embodiment of the production control device of the intelligent mold in the embodiment of the present invention includes:

[0129] A creation module is used to perform a deep neural network analysis on the material characteristic data of the cutting die, obtain a first cutting parameter set, and create a first cutting path plan for the cutting die;

[0130] An analysis module, used for performing finite element analysis and real-time temperature monitoring on the cutting process of the cutting die based on the first cutting parameter set and the first cutting path scheme, and obtaining cutting temperature distribution data;

[0131] A prediction module is used to predict the thermal expansion behavior of the cutting die based on the cutting temperature distribution data and generate a thermal expansion compensation strategy;

[0132] The optimization module is used to perform deep reinforcement learning analysis and multi-objective optimization on the first cutting parameter set and the first cutting path plan based on the thermal expansion compensation strategy to obtain the second cutting parameter set and the second cutting path plan.

[0133] Through the collaboration of the above components, by performing deep neural network analysis on the material characteristic data of the die, the cutting parameters can be predicted more accurately, the accuracy of the initial parameter setting can be improved, and the number of trial cuts can be reduced. Finite element analysis and real-time temperature monitoring combined with the first cutting parameter set and path scheme can accurately simulate the temperature distribution during the cutting process, predict the thermal expansion behavior based on the cutting temperature distribution data, and generate a thermal expansion compensation strategy, which can effectively reduce the processing error caused by thermal deformation. Through deep reinforcement learning analysis and multi-objective optimization, multiple performance indicators such as cutting accuracy, efficiency and minimization of thermal effects can be considered at the same time to achieve more comprehensive optimization. The reinforcement learning environment model and action value estimation network are used to achieve adaptive adjustment of cutting parameters and improve the robustness of the processing process. Multi-objective optimization and cubic spline interpolation technology are used to generate a continuous and smooth cutting trajectory, reduce vibration and stress concentration during processing, and improve cutting quality. Collision detection and obstacle avoidance optimization are performed on the cutting trajectory to ensure the safety and reliability of the processing process, thereby realizing a high-precision and high-efficiency intelligent mold production process.

[0134] The present invention also provides a computer device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the production control method of the intelligent mold in the above-mentioned embodiments.

[0135] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the production control method of the intelligent mold.

[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0137] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0138] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A production control method for an intelligent mold, characterized in that: The production control method of the intelligent mold includes: A deep neural network analysis is performed on the material property data of the knife die to obtain a first cutting parameter set, and a first cutting path plan for the knife die is created; specifically, the method comprises: extracting features and normalizing the material property data of the knife die to obtain a standardized material feature vector; inputting the standardized material feature vector into the input layer of the deep neural network, mapping the standardized material feature vector into a 256-dimensional feature space, and obtaining an initial feature representation; inputting the initial feature representation into the first convolution layer, processing it with 64 3x3 convolution kernels and a ReLU activation function, obtaining a low-level feature map, performing a maximum pooling operation on the low-level feature map, and downsampling with a 2x2 pooling window and a step size of 2 to obtain a compressed feature map; inputting the compressed feature map into the second convolution layer, processing it with 128 3x3 convolution kernels and a LeakyReLU activation function, obtaining an intermediate feature map, performing a batch normalization operation on the intermediate feature map, and using an exponential moving average with a momentum of 0.99 to obtain a standardized feature map; performing an attention mechanism on the standardized feature map, and calculating the feature using a self-attention module The correlation between the two features is obtained to obtain an enhanced feature map, and the enhanced feature map is globally averaged pooled to compress the spatial dimension of each feature channel into a single value to obtain a target feature vector; the target feature vector is input into a fully connected layer, and is processed using 512 neurons and a Tanh activation function to obtain a high-level feature representation; Dropout regularization is performed on the high-level feature representation, and neurons are randomly shielded using a dropout rate of 0.5 to obtain regularized features, and the regularized features are input into an output layer, and two parallel fully connected layers are used to output cutting speed parameters and laser power parameters respectively; cross-validation and error analysis are performed on the cutting speed parameters and the laser power parameters to obtain a first cutting parameter set; according to the first cutting parameter set, the geometric shape data of the knife die is discretized to obtain a grid model of the knife die, and a path search is performed on the grid model to obtain an initial cutting path; according to the initial cutting path, the cutting sequence is optimized by a genetic algorithm to obtain an optimized cutting sequence, and a local search is performed on the optimized cutting sequence to obtain a first cutting path solution; Based on the first cutting parameter set and the first cutting path scheme, a finite element analysis and real-time temperature monitoring are performed on the cutting process of the cutting die to obtain cutting temperature distribution data; Predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy; Based on the thermal expansion compensation strategy, deep reinforcement learning analysis and multi-objective optimization are performed on the first cutting parameter set and the first cutting path solution to obtain a second cutting parameter set and a second cutting path solution.

2. The production control method of the intelligent mold according to claim 1, characterized in that: The step of performing finite element analysis and real-time temperature monitoring on the cutting process of the cutting die based on the first cutting parameter set and the first cutting path scheme to obtain cutting temperature distribution data includes: According to the first cutting parameter set and the first cutting path scheme, three-dimensional geometric modeling is performed on the cutting die to obtain an accurate three-dimensional model of the cutting die, and the accurate three-dimensional model of the cutting die is meshed to obtain a finite element mesh with local encryption; According to the laser power parameter in the first cutting parameter set, a laser heat source is mathematically modeled to obtain a time-varying heat source function, and a heat conduction partial differential equation is constructed based on the finite element mesh and the time-varying heat source function, and a Crank-Nicolson implicit difference format is used for discretization to obtain a heat conduction equation group; Solving the heat conduction equations to obtain the temperature field distribution at each time step, and dynamically updating the temperature field distribution according to the first cutting path scheme to obtain spatiotemporal temperature field evolution data; Performing thermal stress analysis on the spatiotemporal temperature field evolution data to obtain stress distribution during the cutting process, and performing thermal deformation calculation on the cutting die based on the stress distribution to obtain a deformation field during the cutting process; The deformation field is monitored in real time to obtain discrete temperature monitoring data, and the discrete temperature monitoring data is interpolated and reconstructed to obtain cutting temperature distribution data.

3. The production control method of the intelligent mold according to claim 1, characterized in that: The method of predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data and generating a thermal expansion compensation strategy comprises: Decomposing the cutting temperature distribution data in time and space to obtain the main modes and time coefficients of the temperature field, and constructing a rapid heat conduction prediction model of the cutting die according to the main modes and the time coefficients; Performing parameter sensitivity analysis on the rapid heat conduction prediction model to obtain a key parameter set that affects thermal expansion, and constructing a thermal expansion probability distribution function of the cutting die based on the key parameter set; The thermal expansion probability distribution function is simulated and analyzed to obtain a representative thermal expansion scenario set, and a thermal expansion compensation strategy is optimized based on the representative thermal expansion scenario set to generate a thermal expansion compensation strategy.

4. The production control method of the intelligent mold according to claim 1, characterized in that: The method of performing deep reinforcement learning analysis and multi-objective optimization on the first cutting parameter set and the first cutting path scheme based on the thermal expansion compensation strategy to obtain a second cutting parameter set and a second cutting path scheme includes: Parameterizing and encoding the thermal expansion compensation strategy to obtain a state space representation of the thermal expansion compensation strategy; Constructing a reinforcement learning environment model and a state transfer function of a die cutting process according to the state space representation and the first cutting parameter set; Creating an action value estimation network based on the state transfer function, and training the action value estimation network to obtain an optimized Q function; Generate a cutting parameter adjustment strategy according to the optimized Q function to obtain a second cutting parameter set; Performing multi-objective modeling on the first cutting path scheme, and setting a multi-objective optimization function for minimizing the path length and the heat-affected zone; Calculating a plurality of candidate cutting path solutions based on the multi-objective optimization function, and performing path cubic spline interpolation on the plurality of candidate cutting path solutions respectively to obtain a plurality of continuous and smooth cutting trajectories; The plurality of continuous and smooth cutting trajectories are subjected to collision detection and optimization solution to obtain a second cutting path solution.

5. The production control method of the intelligent mold according to claim 4 is characterized in that: The calculating of a plurality of candidate cutting path solutions based on the multi-objective optimization function, and performing path cubic spline interpolation on the plurality of candidate cutting path solutions respectively to obtain a plurality of continuous and smooth cutting trajectories, comprises: Performing weight vectorization processing on the multi-objective optimization function to obtain a set of uniformly distributed weight vector sets; Decomposing the multi-objective optimization problem according to the weight vector set, and converting the multi-objective into a single objective using a weighted sum method to obtain m sub-problems; Particle swarm optimization is performed on each sub-problem to obtain m groups of optimal solutions, each group of optimal solutions represents a candidate cutting path solution; Performing non-dominated sorting on the m groups of optimal solutions to obtain a target optimal solution set, and performing path representation conversion on each solution in the target optimal solution set to convert the discrete point sequence into a parameterized curve representation to obtain a control point set; Based on the control point set, a cubic spline interpolation function is constructed using natural boundary conditions to obtain a piecewise polynomial function; Solving the coefficients of the piecewise polynomial function to obtain a coefficient matrix of the spline function, and performing high-density sampling on each candidate cutting path scheme according to the coefficient matrix to obtain a cutting trajectory point set; The curvature analysis is performed on the cutting trajectory point set, and the curvature value of each point is calculated to obtain a curvature distribution function, and the cutting trajectory is adaptively sampled based on the curvature distribution function to obtain multiple continuous smooth cutting trajectories.

6. A production control device for an intelligent mold, characterized in that: Used to execute the production control method of the intelligent mold according to any one of claims 1 to 5, the production control device of the intelligent mold comprises: A creation module is used to perform a deep neural network analysis on the material characteristic data of the cutting die, obtain a first cutting parameter set, and create a first cutting path plan for the cutting die; An analysis module, configured to perform finite element analysis and real-time temperature monitoring on the cutting process of the cutting die based on the first cutting parameter set and the first cutting path scheme, and obtain cutting temperature distribution data; A prediction module, used for predicting the thermal expansion behavior of the cutting die based on the cutting temperature distribution data, and generating a thermal expansion compensation strategy; An optimization module is used to perform deep reinforcement learning analysis and multi-objective optimization on the first cutting parameter set and the first cutting path scheme based on the thermal expansion compensation strategy to obtain a second cutting parameter set and a second cutting path scheme.

7. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the production control method of the intelligent mold described in any one of claims 1 to 5 is implemented. 8 . A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the production control method of the intelligent mold according to any one of claims 1 to 5.

Images