Process parameter data determination method and device, electronic equipment and storage medium

Abstract

Embodiments of the invention disclose a process parameter data determination method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining preset parameter data of preset process parameters and parameter boundary data of to-be-predicted process parameters; performing initialization processing on the population of the genetic algorithm according to the preset parameter data and the parameter boundary data, each individual in the obtained initial population being used for indicating a group of process parameter data; inputting the process parameter data corresponding to the initial population into a pre-trained proxy model to obtain a profile tolerance deviation predicted value; calculating a fitness value according to the profile tolerance deviation predicted value; and performing optimization iteration on the initial population according to the fitness value, taking the maximization of the fitness value as an iterative optimization target until the number of iterations reaches a preset number of times, and outputting process parameter data of the target individual.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and in particular relates to a method, apparatus, electronic device and storage medium for determining process parameter data. Background Technology

[0002] In traditional industrial manufacturing, the optimization of process parameters often relies on empirical formulas or experimental data. While this experience-based optimization method can achieve certain results under specific conditions, its efficiency and accuracy often fall short when faced with complex, multi-dimensional parameter spaces. Empirical formulas are limited by specific process conditions, have a narrow scope of application, are difficult to flexibly adapt to dynamically changing processing environments, and cannot be easily extended to other processing conditions or scenarios.

[0003] In recent years, with the rapid development of intelligent manufacturing technology, machine learning-based process parameter optimization methods have gradually emerged. Machine learning models such as neural networks, through in-depth training on the relationships between a large number of process parameters and target indicators, can more accurately fit nonlinear and multidimensional data, significantly improving the accuracy of process prediction. However, most existing machine learning methods focus on prediction across the entire parameter space, rarely considering the need for dynamic optimization when some parameters are known.

[0004] Furthermore, acquiring high-quality data in industrial settings is costly, especially when large amounts of experimental data are required. The time and economic costs of data collection often become limiting factors. Therefore, effectively utilizing limited data and achieving dynamic optimization with some known parameters has become a significant challenge in current process parameter optimization. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for determining process parameter data, which can effectively optimize process parameter data.

[0006] In a first aspect, embodiments of this application provide a method for determining process parameter data, the method comprising:

[0007] Obtain the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted;

[0008] Based on the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized, and each individual in the initial population is used to indicate a set of process parameter data.

[0009] The process parameter data corresponding to the initial population is input into the pre-trained surrogate model to obtain the contour deviation prediction value; the contour deviation prediction value is used to describe the difference between the contour data of the processed workpiece obtained based on the process parameter data corresponding to the initial population and the preset contour data.

[0010] The fitness value is calculated based on the predicted profile deviation value; the fitness value is used to describe the process performance of the process parameter data indicated by the individual.

[0011] The initial population is optimized iteratively based on the fitness value, with the goal of maximizing the fitness value, until the number of iterations reaches the preset number. The process parameter data of the target individual is then output. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual.

[0012] Secondly, embodiments of this application provide a device for determining process parameter data, the device comprising:

[0013] The acquisition module is used to acquire the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted;

[0014] The initialization module is used to initialize the population of the genetic algorithm based on preset parameter data and parameter boundary data. Each individual in the initial population is used to indicate a set of process parameter data.

[0015] The input module is used to input the process parameter data corresponding to the initial population into the pre-trained surrogate model to obtain the contour deviation prediction value; the contour deviation prediction value is used to describe the difference between the contour data of the processed workpiece obtained based on the process parameter data corresponding to the initial population and the preset contour data.

[0016] The calculation module is used to calculate the fitness value based on the profile deviation prediction value; the fitness value is used to describe the process performance of the process parameter data indicated by the individual.

[0017] The iteration module is used to optimize the initial population based on the fitness value, with the goal of maximizing the fitness value, until the number of iterations reaches a preset number. It outputs the process parameter data of the target individual, where the fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual.

[0018] Thirdly, embodiments of this application provide an electronic device, the device including: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the method as described in the first aspect or any possible implementation of the first aspect.

[0019] Fourthly, embodiments of this application provide a readable storage medium storing computer program instructions that, when executed by a processor, implement the method as described in the first aspect or any possible implementation thereof.

[0020] In this embodiment, by acquiring the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted, the starting point and range of optimization can be clearly defined, avoiding blind search and thus improving optimization efficiency. Based on the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized. Each individual in the initial population is used to indicate a set of process parameter data. The initial individuals cover the key areas of the parameter space, which helps improve the initial accuracy of the model. The process parameter data corresponding to the initial population is input into a pre-trained surrogate model to obtain the contour deviation prediction value. The surrogate model can quickly predict the impact of process parameters on processing quality, avoiding complex physical simulations or experiments. The fitness value is calculated based on the contour deviation prediction value. The fitness value is used to describe the process performance of the process parameter data indicated by the individual. The smaller the fitness value, the better the process parameter data and the better the processing effect. The initial population is iteratively optimized based on the fitness value, with the maximization of the fitness value as the iterative optimization objective, until the number of iterations reaches a preset number, and the process parameter data of the target individual is output. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual. The introduction of fitness values ​​makes the optimization results more comprehensive, taking into account both the average level and fluctuation of the deviation. The optimized process parameters can improve processing efficiency and reduce processing time, thereby improving production efficiency. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating a method for determining process parameter data provided in an embodiment of this application;

[0023] Figure 2 This is a flowchart illustrating a method for determining process parameter data provided in an embodiment of this application;

[0024] Figure 3 This is a schematic diagram of the structure of a neural network model provided in an embodiment of this application;

[0025] Figure 4 This is a flowchart of a population iteration provided in an embodiment of this application;

[0026] Figure 5 This is a schematic diagram illustrating a method for determining process parameter data provided in an embodiment of this application;

[0027] Figure 6 This is a schematic diagram of a loss function value provided in an embodiment of this application;

[0028] Figure 7 This is a schematic diagram of a fitness curve provided in an embodiment of this application;

[0029] Figure 8 This is a schematic diagram of a device for determining process parameter data provided in an embodiment of this application;

[0030] Figure 9 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0031] The features and exemplary embodiments of various aspects of this application will now be described in detail. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only configured to explain this application and are not configured to limit this application. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples of this application.

[0032] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0033] Figure 1 This is a flowchart of a method for determining process parameter data provided in an embodiment of this application.

[0034] like Figure 1 As shown, the method for determining the process parameter data may include steps 110-150. This method is applied to the device for determining process parameter data, as detailed below:

[0035] Step 110: Obtain the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted;

[0036] Preset process parameters: These are parameters that are set in advance during the manufacturing process to control the processing, such as temperature, pressure, and speed. These parameters are calculated based on experience and theory to achieve the best processing results.

[0037] Parameter boundary data: During the optimization process, each process parameter has an allowable range of variation, which is called the parameter boundary. Parameter boundary data defines the minimum and maximum values ​​of each parameter, ensuring that the optimization process is carried out within a reasonable range.

[0038] Before starting optimization, it is necessary to define the preset process parameter values ​​and the allowable range of variation for each parameter. Preset parameter data provides initial reference values, while parameter boundary data ensures that the optimization process proceeds within a reasonable range. Preset process parameter values ​​are obtained from process design documents or historical data. Simultaneously, based on process requirements and equipment capabilities, the minimum and maximum values ​​for each parameter are determined, forming parameter boundary data. Obtaining both preset parameter data and parameter boundary data provides a clear starting point and range for initializing the genetic algorithm population, ensuring that the optimization process proceeds within a reasonable range.

[0039] Step 120: Initialize the population of the genetic algorithm according to the preset parameter data and parameter boundary data. Each individual in the initial population is used to indicate a set of process parameter data.

[0040] Population: In optimization algorithms, a population refers to a set of candidate solutions. Each candidate solution (individual) represents a set of process parameter data. Individuals in the population evolve through methods such as genetic algorithms to find the optimal solution.

[0041] The initial population is the starting point of the optimization process, and each individual in the population represents a set of process parameter data. The initial population is usually generated based on preset parameter data and parameter boundary data to ensure that each individual is within a reasonable range.

[0042] A random generation method can be used to randomly generate a set of process parameter data within the boundary range of each parameter, forming an initial population. Alternatively, a heuristic method can be used to generate an initial population based on preset parameter data. Generating an initial population based on preset parameter data and parameter boundary data provides initial candidate solutions for subsequent optimization.

[0043] Step 130: Input the process parameter data corresponding to the initial population into the pre-trained surrogate model to obtain the contour deviation prediction value; the contour deviation prediction value is used to describe the difference between the contour data of the processed workpiece obtained based on the process parameter data corresponding to the initial population and the preset contour data.

[0044] Proxy models: Proxy models are simplified models used to approximate complex real-world models. They are typically built using a training dataset to quickly predict the output corresponding to input parameters.

[0045] Profile deviation prediction: Profile deviation refers to the difference between the actual profile of a machined workpiece and the designed profile. The profile deviation prediction value is obtained through a surrogate model and is used to evaluate the impact of process parameters on machining quality.

[0046] The surrogate model is used to quickly predict the impact of process parameters on processing quality. By inputting the process parameter data of each individual in the population into the surrogate model, the corresponding profile deviation prediction value can be obtained.

[0047] The process parameter data of each individual in the population is used as input, and a surrogate model is used to predict the profile deviation of each individual. The surrogate model can be a model built based on machine learning algorithms. By inputting the process parameter data corresponding to the initial population into the surrogate model, the profile deviation can be predicted quickly, avoiding complex physical simulations or experiments.

[0048] Step 140: Calculate the fitness value based on the profile deviation prediction value; the fitness value is used to describe the process performance of the process parameter data indicated by the individual.

[0049] Fitness value: Fitness value is an indicator that measures the quality of an individual (i.e., a set of process parameter data). The smaller the fitness value, the better the process parameter data corresponding to the individual, and the better the processing effect.

[0050] Fitness value is an indicator of the quality of an individual process parameter, and it is usually inversely proportional to the predicted profile deviation. The smaller the profile deviation, the larger the fitness value, indicating better process parameter data.

[0051] Based on the predicted profile deviation values, the fitness value of each individual is calculated. The fitness function can be a simple linear function or a complex nonlinear function, depending on the optimization objective and constraints. Calculating the fitness value based on the predicted profile deviation values ​​quantifies the quality of the process parameter data, providing clear guidance for optimization.

[0052] Step 150: Optimize the initial population based on the fitness value, with the goal of maximizing the fitness value, until the number of iterations reaches the preset number. Output the process parameter data of the target individual. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual.

[0053] Through multiple iterations, the individuals in the population are gradually improved to ultimately find the optimal process parameters, achieving dynamic optimization. In each iteration, superior individuals are selected based on their fitness values, and new individuals are generated through crossover and mutation operations in the genetic algorithm, forming a new population. This process is repeated until the preset number of iterations is reached. Finally, the process parameters corresponding to the individual with the lowest fitness value (i.e., the target individual) are output.

[0054] By optimizing process parameters, the contour deviation of machined workpieces can be significantly reduced, improving machining accuracy. Using surrogate models for prediction avoids complex physical simulations or experiments, greatly shortening optimization time. Optimizing process parameters can reduce scrap rates, lower material and energy consumption, and thus reduce production costs. Optimized process parameters can improve machining efficiency and reduce machining time, thereby increasing production efficiency. Optimizing process parameters can reduce fluctuations during machining, enhancing process stability and consistency. In short, optimizing process parameters significantly improves machining quality and production efficiency while reducing costs, demonstrating significant economic benefits and application value.

[0055] like Figure 2 As shown, discrete processing data was generated through designed processing experiments for initial model training. This experimental data provided rich training samples for the neural network model, helping to improve the initial accuracy of the model. The improved accuracy of the initial model laid a solid foundation for the subsequent optimization process and reduced the number of iterations required for model training.

[0056] By combining the high-precision predictive capabilities of neural network models with the global search advantages of genetic algorithms, intelligent recommendation and optimization of process parameters can be achieved. Neural network models can accurately predict the impact of process parameters on processing results, while genetic algorithms can perform global searches under complex, multivariate conditions to find the optimal solution. This combination not only improves optimization efficiency but also ensures that each parameter reaches its optimal configuration under complex, multivariate conditions, significantly enhancing the optimization effect of process parameters.

[0057] By feeding optimized process parameters and actual processing results back to the database, the system can continuously accumulate new data and enrich the dataset. This new data is used for automatic model iteration and optimization, improving the accuracy of subsequent predictions and optimizations. The system's automatic iteration and optimization capabilities enable the model to continuously adapt to new processing conditions and environments, improving the model's generalization ability and prediction accuracy.

[0058] Therefore, discrete processing data is generated through designed processing experiments for initial model training. This experimental data provides rich training samples for the neural network model, helping to improve the model's initial accuracy. By combining the high-precision predictive capability of the neural network model with the global search advantage of the genetic algorithm, intelligent recommendation and optimization of process parameters are achieved. The neural network model can accurately predict the impact of process parameters on processing results, while the genetic algorithm can perform a global search under complex multivariate conditions to find the optimal solution. By feeding the optimized process parameters and actual processing results back to the database, the system can continuously accumulate new data, constantly enriching the dataset. This new data is used for automatic model iteration and optimization, improving the accuracy of subsequent predictions and optimizations.

[0059] In one possible embodiment, step 130 includes:

[0060] The process parameter data corresponding to the initial population is input into the pre-trained surrogate model to obtain the mean and standard deviation of the deviation; the deviation is calculated based on the contour data of the processed workpiece obtained from the process parameter data corresponding to the initial population and the preset contour data; the preset contour data is the contour data of the preset workpiece.

[0061] The calculation of the fitness value based on the predicted profile deviation includes:

[0062] The fitness value is calculated based on the mean and standard deviation of the deviation.

[0063] During the optimization process, the surrogate model is not only used to predict profile deviation, but also to further refine the prediction results, providing the mean and standard deviation of the deviation. The deviation refers to the difference between the actual profile of the machined workpiece and the preset profile. The mean of the deviation reflects the average level of the overall deviation, while the standard deviation reflects the fluctuation of the deviation.

[0064] The process parameter data of each individual in the initial population is input into a pre-trained surrogate model. The surrogate model predicts the contour data of the workpiece based on the input process parameter data. The predicted contour data is compared with preset contour data, and the deviation for each data point is calculated. Statistical analysis is performed on the deviations of all data points to calculate the mean and standard deviation of the deviations.

[0065] Fitness score is an indicator of the quality of an individual process parameter, and it is usually inversely proportional to the mean and standard deviation of the deviation. The smaller the mean and standard deviation of the deviation, the larger the fitness score, indicating that the process parameter data is better.

[0066] Define the fitness function: The fitness function can be a function that takes into account both the mean and standard deviation of the deviation. For example, the fitness value can be defined as:

[0067] Fitness value = -(α mean deviation + β × standard deviation of deviation + γ processing time);

[0068] Among them, α, β, and γ are weighting coefficients used to balance the effects of the mean, standard deviation, and processing time.

[0069] Substitute the mean and standard deviation of each individual into the fitness function to calculate the fitness value of each individual.

[0070] The introduction of the mean and standard deviation of the deviation makes the optimization results more comprehensive, and can simultaneously consider the average level and fluctuation of the deviation, further enhancing the stability of the process, further refining the optimization process, significantly improving processing quality and production efficiency, and reducing costs, thus having significant economic benefits and application value.

[0071] In one possible embodiment, prior to step 130, the following may also be included:

[0072] Step 210: Based on the orthogonal experimental design method, design M experimental schemes according to N process parameters, where N and M are both positive integers; each experimental scheme corresponds to a set of experimental process parameter data; wherein, the experimental process parameter data includes: spindle speed, feed rate, depth of cut and path spacing;

[0073] Step 220: Obtain the processing result data of the workpiece. The processing result data is obtained by conducting processing experiments based on M experimental schemes.

[0074] Step 230: Normalize the experimental process parameter data to obtain normalized experimental process parameter data;

[0075] Step 240: Generate a feature matrix based on the normalized experimental process parameter data. The feature matrix has a dimension of M*N.

[0076] Step 250: Generate a target matrix based on the processing result data of the workpiece. The dimension of the target matrix is ​​M*2.

[0077] Step 260: Train the neural network model based on the feature matrix and the target matrix until the neural network model meets the training stopping condition, and obtain the trained surrogate model.

[0078] Step 210 involves orthogonal experimental design, a method for optimizing experimental design. By rationally selecting experimental points, the number of experiments can be reduced while ensuring the comprehensiveness and representativeness of the experimental results. N process parameters correspond to N dimensions, and M experimental schemes correspond to M experimental points.

[0079] Identify the N process parameters that need optimization, such as temperature, pressure, and speed. Use orthogonal experimental design to design M experimental schemes. Each experimental scheme corresponds to a set of experimental process parameter data.

[0080] Spindle speed refers to the rotational speed of a machine tool spindle, usually measured in revolutions per minute (RPM). Spindle speed directly affects cutting speed and machining efficiency; excessively high or low speeds will negatively impact machining quality and tool life.

[0081] Feed rate refers to the speed at which the cutting tool moves along the cutting direction during machining, usually expressed as feed per minute (mm / min). Feed rate affects cutting force and surface finish; both excessively high and low feed rates will negatively impact machining results.

[0082] Depth of cut refers to the thickness of material cut from the workpiece surface by a cutting tool in a single cutting operation, usually measured in millimeters (mm). The depth of cut directly affects cutting force and machining efficiency; an excessively large depth of cut may lead to premature tool wear, while an excessively small depth of cut may negatively impact machining efficiency.

[0083] Path spacing refers to the distance between adjacent paths of the cutting tool during machining, usually measured in millimeters (mm). Path spacing affects the surface roughness and machining efficiency; excessively large path spacing may result in substandard surface roughness, while excessively small path spacing may affect machining efficiency.

[0084] By dynamically adjusting process parameters, the manufacturing process can be optimized, reducing processing time and resource consumption, thereby improving production efficiency. Whether in the aerospace field or other manufacturing industries, dynamically adjusting process parameters can significantly improve production efficiency.

[0085] In the aerospace field, such as the machining of aircraft engine parts, dynamically adjusting process parameters can shorten the machining cycle and improve production efficiency. In the automotive manufacturing field, such as the machining of key components like engine blocks and crankshafts, dynamically adjusting process parameters can reduce machining time and improve production efficiency.

[0086] Step 220 involves obtaining processing result data for each experimental scheme through actual processing experiments. The processing result data includes the contour data of the processed workpiece, etc.

[0087] Based on the designed M experimental schemes, conduct actual machining experiments. Record the machining results data, such as the contour data of the workpiece corresponding to each experimental scheme.

[0088] Step 230 involves normalizing the experimental process parameter data to obtain normalized experimental process parameter data. Normalization converts process parameter data with different dimensions to the same dimension, facilitating subsequent feature extraction and model training. Normalization can avoid the impact of dimensional differences between different parameters on model training.

[0089] Normalization method selection: Commonly used normalization methods include min-max normalization and Z-score normalization. The experimental process parameter data for each process parameter are normalized to obtain the normalized experimental process parameter data.

[0090] Step 240 involves generating a feature matrix based on the normalized experimental process parameter data. The feature matrix has dimensions M*N. The feature matrix is ​​formed by arranging the normalized experimental process parameter data column-wise into an M*N matrix. Each column corresponds to the process parameter data of one experimental scheme, and each row corresponds to one process parameter.

[0091] Arrange the experimental process parameter data: Arrange the normalized experimental process parameter data column-wise to form a feature matrix. The feature matrix has a dimension of M*N, where N is the number of process parameters and M is the number of experimental schemes.

[0092] Step 250 involves generating a target matrix based on the processing results data of the workpiece. The target matrix has a dimension of M*2. The target matrix is ​​formed by arranging the processing results data column-wise into an M*2 matrix. Each row corresponds to the processing results data of one experimental scheme, typically including the mean and standard deviation of the deviation.

[0093] Arrange the processing result data: Arrange the processing result data of the processed workpieces (such as the mean and standard deviation of the deviations) in columns to form a target matrix. The dimension of the target matrix is ​​M*2, where M is the number of experimental schemes and 2 is the number of processing result data (such as the mean and standard deviation of the deviations).

[0094] Step 260 involves training a neural network model based on the feature matrix and the target matrix until the model meets the training stopping condition, resulting in a trained surrogate model. The neural network model constructs the surrogate model by learning the relationship between the feature matrix and the target matrix. The surrogate model is used to predict the processing results corresponding to the input process parameter data. Selecting a neural network structure: Choose a suitable neural network structure, such as a multilayer perceptron (MLP) or a convolutional neural network (CNN).

[0095] The feature matrix and target matrix are used as training data to train a neural network model. During training, the network parameters are continuously adjusted using the backpropagation algorithm until the model meets the training stopping conditions, such as reaching a preset number of training epochs or loss function convergence. After training, a trained surrogate model is obtained and used for subsequent process parameter optimization.

[0096] By employing orthogonal experimental design, experimental points are rationally selected, reducing the number of experiments while ensuring the comprehensiveness and representativeness of the results. Normalization and feature matrix generation further improve the quality of the model's input data, thereby enhancing the model's prediction accuracy. The orthogonal experimental design method reduces the number of experiments, thus shortening data acquisition time. Normalization and feature matrix generation simplify the processing of the model's input data, further reducing model training time.

[0097] A well-trained surrogate model can quickly predict the processing results corresponding to the input process parameters, thereby accelerating the process parameter optimization process. The introduction of the mean and standard deviation of the deviation makes the optimization process more comprehensive, taking into account both the average level and fluctuation of the deviation, further improving optimization efficiency.

[0098] In summary, by employing orthogonal experimental design, normalization, generation of feature and target matrices, and training of neural network models, the model accuracy and optimization efficiency were significantly improved, while experimental costs were reduced, demonstrating significant economic benefits and application value.

[0099] In one possible embodiment, step 250 includes:

[0100] Based on the processing result data of the workpiece and the preset detection points on the workpiece, determine the contour deviation data of the workpiece;

[0101] For each preset detection point, the deviation is calculated based on the contour deviation data of the processed workpiece;

[0102] For each preset detection point, the deviation in the corresponding direction is calculated based on the predetermined tolerance and deviation;

[0103] For each detection point with out-of-tolerance, the mean and standard deviation of the out-of-tolerance are calculated and used as the target matrix of the neural network model. The dimension of the target matrix is ​​M*2.

[0104] Preset inspection points: Specific points pre-defined on the contour of the workpiece to detect contour deviations. These inspection points are typically located at critical positions on the workpiece and are used to evaluate machining quality.

[0105] Profile deviation data: The difference between the actual profile of the machined workpiece and the preset profile. Profile deviation data reflects the profile accuracy of the machined workpiece.

[0106] Deviation: The difference between the actual contour of the machined workpiece and the preset contour at each preset detection point. Deviation can be positive or negative; a positive value indicates that the actual contour exceeds the preset contour, and a negative value indicates that the actual contour is below the preset contour.

[0107] Preset tolerance: A pre-defined range of permissible profile deviations. Preset tolerance defines the maximum permissible value for the profile deviation of a machined workpiece.

[0108] Out of tolerance: The difference between the actual deviation and the predetermined tolerance at each preset inspection point. Out of tolerance can be positive or negative. A positive value indicates that the actual deviation exceeds the predetermined tolerance, while a negative value indicates that the actual deviation is within the predetermined tolerance range.

[0109] By comparing the actual contour of the machined workpiece with the preset contour, the contour deviation data at each preset detection point is determined. The contour deviation data is used to evaluate the contour accuracy of the machined workpiece.

[0110] Obtain the actual contour data of the workpiece from the machining experiment. Determine the positions of preset detection points on the contour of the workpiece. At each preset detection point, calculate the difference between the actual contour and the preset contour to obtain contour deviation data.

[0111] Deviation is the difference between the actual contour and the preset contour at each preset inspection point. Deviation is used to evaluate the contour accuracy of a machined workpiece at each inspection point. The deviation is obtained by calculating the difference between the actual contour and the preset contour at each preset inspection point.

[0112] Out-of-tolerance (OOT) is the difference between the actual deviation and the predetermined tolerance at each preset inspection point. OOT is used to evaluate whether the profile accuracy of a machined workpiece at each inspection point meets the predetermined tolerance requirements.

[0113] At each preset detection point, the difference between the actual deviation and the predetermined tolerance is calculated to obtain the out-of-tolerance. For the out-of-tolerance at each detection point, the mean and standard deviation of the out-of-tolerance are calculated and used as the target matrix of the neural network model. The target matrix has a dimension of M*2.

[0114] The mean and standard deviation of the out-of-tolerance are used to evaluate the overall contour accuracy of the machined workpiece. The mean of the out-of-tolerance reflects the average level of the overall deviation, while the standard deviation reflects the fluctuation of the deviation. The target matrix is ​​used to train a neural network model, enabling it to predict the machining results corresponding to the input process parameter data.

[0115] Statistical analysis was performed on the deviations at all preset detection points to calculate the mean and standard deviation of the deviations. The mean and standard deviation of the deviations were then arranged column-wise to form a target matrix. The target matrix has a dimension of M*2, where M is the number of experimental schemes and 2 represents the mean and standard deviation of the deviations.

[0116] By calculating the deviation and out-of-tolerance at each preset detection point, and further calculating the mean and standard deviation of the out-of-tolerance, the target matrix becomes more comprehensive and accurate. This helps improve the training accuracy of the neural network model, thereby improving the prediction accuracy of the surrogate model. By calculating the standard deviation of the out-of-tolerance, the fluctuation of the workpiece contour deviation can be assessed. This helps enhance the robustness of the neural network model, enabling it to more stably predict processing results when faced with different process parameter inputs.

[0117] By calculating the mean and standard deviation of the deviations, the contour accuracy of the machined workpiece can be comprehensively evaluated. This helps to optimize process parameters, reduce fluctuations during machining, and improve machining efficiency.

[0118] In summary, by calculating the deviation and out-of-tolerance at each preset detection point, and further calculating the mean and standard deviation of the out-of-tolerance, the model accuracy and optimization efficiency are significantly improved, while reducing experimental costs, demonstrating significant economic benefits and application value.

[0119] In one possible embodiment, the following steps may be included prior to step 260:

[0120] Construct a neural network model with multiple layers, including an input layer, hidden layers, fully connected layers, and an output layer. Each fully connected layer uses L2 regularization and Dropout mechanisms, and each fully connected layer uses the ReLU activation function. The output layer uses a linear activation function.

[0121] Train a neural network model to determine the parameters of the trained neural network layers at each of the multiple layers;

[0122] The trained neural network model is then identified as the surrogate model.

[0123] like Figure 3 As shown, a neural network model consists of multiple layers, including an input layer, hidden layers, fully connected layers, and an output layer. Each layer is responsible for different functions, such as data input, feature extraction, data transformation, and result output. By constructing a multi-layered neural network model, the model's complexity and expressive power can be increased, thereby improving the model's prediction accuracy.

[0124] Input layer: The input layer receives normalized experimental process parameter data as input to the neural network model.

[0125] Hidden layers: Hidden layers consist of multiple fully connected layers, each responsible for feature extraction and data transformation of the input data.

[0126] Fully connected layers: Fully connected layers are the core part of neural network models. Each fully connected layer uses L2 regularization and Dropout mechanisms to prevent overfitting, and uses the ReLU activation function to improve the non-linear expressive power of the model.

[0127] Output layer: The output layer uses a linear activation function to output the mean and standard deviation of the out-of-range as the prediction result of the neural network model.

[0128] The purpose of training a neural network model is to determine the parameters of each layer by learning the relationship between the feature matrix and the target matrix, thereby enabling the model to accurately predict the processing results corresponding to the input process parameter data. The parameters of the neural network model, including weights and biases, are randomly initialized. The feature matrix is ​​input into the neural network model, and the output is calculated through forward propagation. The output is compared with the target matrix, and the loss function is calculated. The gradient of the loss function with respect to each parameter is calculated using the backpropagation algorithm, and the parameters are updated. The process of forward propagation, loss calculation, and backpropagation is repeated until the model meets the training stopping conditions, such as reaching a preset number of training epochs or loss function convergence.

[0129] Once training is complete, the parameters of the neural network model are determined, and the model can accurately predict the processing results corresponding to the input process parameters. At this point, the trained neural network model can be used as a surrogate model for subsequent process parameter optimization.

[0130] The parameters of the trained neural network model are saved for later use. This trained neural network model is then used as a surrogate model to predict the processing results corresponding to the input process parameters.

[0131] By constructing a multi-layered neural network model and using L2 regularization and Dropout mechanisms, overfitting can be prevented, and the model's generalization ability can be improved. Using ReLU activation functions and linear activation functions can enhance the model's non-linear expressiveness and prediction accuracy. Furthermore, the use of L2 regularization and Dropout mechanisms prevents overfitting and enhances the model's robustness. During training, parameters are continuously adjusted using the backpropagation algorithm, enabling the model to better adapt to different input data.

[0132] A well-trained surrogate model can quickly predict the processing results corresponding to the input process parameters, thereby accelerating the process parameter optimization process. Multi-level neural network models can better extract features from the input data, improving prediction accuracy and further enhancing optimization efficiency.

[0133] By employing orthogonal experimental design, the number of experiments was reduced, thereby lowering experimental costs. Normalization and the generation of the characteristic matrix further simplified the data processing, further reducing experimental costs.

[0134] Optimizing process parameters can reduce fluctuations during processing and enhance process stability and consistency. Multi-level neural network models can better extract features from input data, improve prediction accuracy, and further enhance process stability.

[0135] In summary, by constructing a multi-level neural network model and using L2 regularization, Dropout, ReLU activation function, and linear activation function, the model accuracy and optimization efficiency are significantly improved, while the experimental cost is reduced, demonstrating significant economic benefits and application value.

[0136] In one possible embodiment, the initial population is iteratively optimized based on its fitness value, with the goal of maximizing the fitness value, until a preset number of iterations is reached, including:

[0137] The initial population is optimized iteratively based on fitness values, with the goal of maximizing fitness values. The individual with the highest fitness value is selected as the new parent.

[0138] New individuals are generated through crossover and mutation operations;

[0139] A new population is generated based on the new parent generation and the new individuals;

[0140] The process parameter data corresponding to the new population is input into the pre-trained surrogate model to obtain the profile deviation prediction value; and the fitness value is calculated based on the profile deviation prediction value; and the new population is optimized and iterated based on the fitness value until the number of iterations reaches the preset number.

[0141] Fitness value: The fitness value is an indicator that measures the quality of an individual (i.e., a set of process parameter data). The smaller the fitness value, the better the process parameter data corresponding to the individual, and the better the processing effect. The fitness value is usually inversely proportional to the profile deviation prediction value, that is, the smaller the profile deviation, the larger the fitness value (or the smaller, depending on the definition of the fitness function).

[0142] Initial population: In optimization algorithms, the initial population refers to a set of candidate solutions. Each candidate solution (individual) represents a set of process parameter data. The initial population is usually generated randomly or based on some heuristic method and serves as the starting point for the optimization process.

[0143] Parent: In genetic algorithms, the parent refers to the individual selected to generate new individuals. Parents are typically individuals with high fitness values. By selecting individuals with high fitness values ​​as parents, it is ensured that the new individuals have a high genetic advantage, thereby improving optimization efficiency.

[0144] Crossover: Crossover is an operation in genetic algorithms that generates new individuals by exchanging some genes between two parent individuals. Crossover simulates the gene recombination process in biological evolution and helps produce new individuals with higher fitness.

[0145] Mutation: Mutation is an operation in genetic algorithms that randomly alters parts of an individual's genes to generate new individuals. Mutation simulates the gene mutation process in biological evolution, helping to increase population diversity and avoid getting trapped in local optima.

[0146] A new population: A new population refers to a collection of new individuals generated through crossover and mutation operations. A new population contains genetic information from the parent individuals as well as new information resulting from mutations, exhibiting higher diversity and fitness.

[0147] In each iteration, the individual with the highest fitness value is selected as the new parent to ensure that the new individual has a high genetic advantage. The process parameter data of each individual in the initial population are input into a pre-trained surrogate model to obtain the silhouette deviation prediction value, and the fitness value is calculated based on the silhouette deviation prediction value. The individual with the highest fitness value is selected as the new parent.

[0148] Crossover and mutation operations are used to generate new individuals, increasing population diversity and avoiding getting trapped in local optima. Exchanging some genes between two parent individuals generates a new individual. Randomly changing some genes in a new individual generates another new individual.

[0149] The new population contains genetic information from the parent individuals as well as new information generated by mutations, exhibiting higher diversity and fitness. A new population is generated by combining new parents with new individuals created through crossover and mutation operations.

[0150] The surrogate model is used to quickly predict the impact of process parameters on machining quality. By inputting the process parameter data of each individual in the new population into the surrogate model, the corresponding contour deviation prediction value can be obtained. Alternatively, the process parameter data of each individual in the new population can be input into a pre-trained surrogate model, which then predicts the contour deviation of the machined workpiece based on the input process parameter data.

[0151] Fitness value is an indicator of an individual's quality, and it is usually inversely proportional to the predicted profile deviation. The smaller the profile deviation, the larger the fitness value, indicating better process parameter data. Calculating fitness value: Based on the predicted profile deviation value, calculate the fitness value for each individual.

[0152] Through multiple iterations, the individuals in the population are gradually improved to find the optimal process parameters. In each iteration, superior individuals are selected based on their fitness values, and new individuals are generated through crossover and mutation operations in the genetic algorithm, forming a new population. Selecting superior individuals: Superior individuals are selected as new parents based on their fitness values. New individuals are generated through crossover and mutation operations.

[0153] The new parent generation is combined with the new individuals generated through crossover and mutation operations to create a new population. This process is repeated until the preset number of iterations is reached.

[0154] By selecting the individual with the highest fitness value as the new parent and using crossover and mutation operations to generate new individuals, optimal process parameters can be quickly found, improving optimization efficiency. Mutation increases population diversity, preventing entrapment in local optima and improving the robustness of optimization results. Through multiple iterations, gradually improving the individuals in the population enhances the model's prediction accuracy, thereby increasing the accuracy of the optimization results. Using a surrogate model for prediction avoids complex physical simulations or experiments, significantly shortening optimization time.

[0155] like Figure 4 As shown, based on known process parameters and boundary conditions, the genetic algorithm is initialized by inputting the boundary values ​​of the parameters and the known parameter values. A set of possible solutions (population) is then initialized, with each individual representing a parameter combination. The fitness function, based on the predictions of a neural network surrogate model, is used to calculate the merits of each individual. The optimization objective is to minimize the process performance index (the sum of the mean and standard deviation of the deviation). In each generation, the genetic algorithm selects individuals with higher fitness as parents and generates new individuals through crossover and mutation operations, ensuring population diversity and fitness. After multiple generations of iteration, the algorithm gradually evolves, approaching the optimal solution. Finally, the genetic algorithm outputs the best combination of process parameters, which not only meets the known parameter requirements but also ensures optimal performance of the remaining parameters in terms of process performance.

[0156] In summary, by selecting the individual with the highest fitness value as the new parent and using crossover and mutation operations to generate new individuals, the optimization efficiency and model accuracy are significantly improved, while the cost is reduced, demonstrating significant economic benefits and application value.

[0157] In addition, such as Figure 5As shown, by managing data acquired from experiments, data generated by models, and data generated from real-world processing, data can be continuously accumulated and updated, thereby constantly improving system performance. Managing data from different sources allows for the continuous enrichment of the dataset, thus enhancing the model's generalization ability. Generalization ability refers to the model's adaptability and predictive accuracy when faced with new data.

[0158] In actual processing, the user applies the selected process parameters to process the workpiece and supplements the initial dataset with the processing label data obtained from the actual processing (including actual contour deviation and processing time). This process is called dataset iterative update. It is crucial for further iterative training of the surrogate model and can effectively improve the model's prediction accuracy and generalization ability. The following is a detailed description of this process:

[0159] Actual processing data collection: Users perform actual processing based on the process parameters recommended by the agent model and record key data during the processing, such as:

[0160] Actual profile deviation: The deviation between the actual profile of a workpiece and the designed profile, usually expressed numerically or graphically.

[0161] Processing time: Record the time required to complete the processing of the workpiece.

[0162] Data labeling: The collected actual processing data is correlated with the corresponding process parameters to form new training samples.

[0163] Expanding the dataset: Adding new training samples to the initial dataset expands the dataset size.

[0164] Data cleaning and preprocessing: Perform necessary cleaning and preprocessing on new data, such as removing outliers and normalizing, to ensure data quality.

[0165] Model training: Retrain the surrogate model using the updated dataset and update the model parameters.

[0166] Model evaluation: Use the validation set to evaluate the model's performance on the new dataset, such as prediction accuracy and generalization ability.

[0167] Model optimization: Based on the evaluation results, adjust the model structure, hyperparameters, etc., to further optimize model performance.

[0168] Iterative loop: The updated proxy model is applied to actual processing, new processing data is collected, and the above steps are repeated to form a data-driven closed-loop optimization process.

[0169] Continuous model improvement: As the dataset is continuously expanded and the model is trained iteratively, the prediction accuracy and generalization ability of the surrogate model will be continuously improved, ultimately achieving more accurate process parameter recommendations.

[0170] New data can reflect the complexity and uncertainty of actual processing, helping models learn more accurate mapping relationships. Diverse datasets can improve the model's generalization ability, enabling it to make accurate predictions under different operating conditions. New data can provide richer information, accelerate the model training process, and shorten the model convergence time.

[0171] Iterative updates to the dataset are crucial for improving the accuracy and generalization ability of the surrogate model. By continuously collecting and utilizing actual processing data, constantly expanding and updating the dataset, and conducting iterative model training, a more accurate and robust surrogate model can be built, providing more reliable process parameter recommendations for actual processing, ultimately achieving efficient and high-quality workpiece processing.

[0172] By continuously improving system performance and model generalization ability, it can adapt to the design of workpiece process parameters with higher precision. High-precision process parameter design requires models with higher prediction accuracy and stability.

[0173] Therefore, by continuously improving system performance and model generalization ability through data management, the stability and reliability of the system are significantly enhanced, the generalization ability and prediction accuracy of the model are improved, and it is adapted to the design of workpiece process parameters with higher precision, which has significant economic benefits and application value.

[0174] The loss function curve during the training of the surrogate model is shown below. Figure 6 As shown, the fitness curve of the genetic algorithm is as follows: Figure 7 As shown in Table 1, the performance of the proxy model is as follows:

[0175] Table 1

[0176]

[0177] To construct and train a surrogate model capable of predicting workpiece contour deviation and machining time after machining, an initial dataset containing process parameters and corresponding machining experimental results is first required. First, key process parameters affecting workpiece machining are identified, such as cutting speed, feed rate, and depth of cut. An experimental scheme is designed: using orthogonal experimental design and other methods, a series of experiments are designed to ensure that different combinations of process parameter levels cover the expected range.

[0178] According to the experimental plan, workpieces were machined using different combinations of process parameters, and the process parameters and machining results of each experiment were recorded. The machining experiment data were collected, specifically requiring the measurement of the deviation between the actual contour and the designed contour of each machined workpiece, which can be done using equipment such as a coordinate measuring machine. The time required to complete the machining of each workpiece was also recorded.

[0179] Building the initial dataset:

[0180] Process parameters: The combination of process parameters corresponding to each experiment, such as cutting speed, feed rate, depth of cut, etc.

[0181] Profile deviation: Profile deviation data for each experiment.

[0182] Processing time: Processing time data for each workpiece corresponding to each experiment.

[0183] Data cleaning: Cleaning the collected data to remove outliers and erroneous data, ensuring data quality.

[0184] Next, select a model architecture, choosing a neural network model suitable for the regression task, such as a multilayer perceptron (MLP) or a convolutional neural network (CNN).

[0185] Input layer: The number of nodes in the input layer is equal to the number of process parameters.

[0186] Output layer: The output layer has 2 nodes, which correspond to the predicted values ​​of contour deviation and processing time, respectively.

[0187] Hidden layers: Design an appropriate number of hidden layers and the number of nodes in each layer based on data complexity and model performance requirements.

[0188] Data preprocessing: Normalize the process parameters and processing results data, such as min-max normalization or z-score standardization, to improve model training efficiency and stability.

[0189] Divide the initial dataset into training and testing sets, for example, 80% for training and 20% for testing. Choose a loss function suitable for the regression task, such as mean squared error (MSE) or mean absolute error (MAE). Select an appropriate optimizer, such as Adam or SGD, and set hyperparameters such as the learning rate.

[0190] The neural network model is trained using a training set and evaluated using a test set. Based on the model's performance on the test set, the model architecture, hyperparameters, etc., are adjusted to improve the model's prediction accuracy. The model performance is then evaluated using evaluation metrics, such as R-squared. 2 Coefficients, mean squared error (MSE), mean absolute error (MAE), etc.

[0191] Use independent validation sets or cross-validation methods to further evaluate the model's generalization ability.

[0192] By designing processing experiments to obtain an initial dataset, and then building and training a neural network model based on this dataset as a surrogate model, it is possible to predict the contour deviation and processing time of the workpiece after processing. This surrogate model can provide important data support for subsequent process parameter optimization, thereby improving processing efficiency and product quality.

[0193] In one possible embodiment, the process parameter data corresponding to the initial population is input into a pre-trained surrogate model to obtain the mean and standard deviation of the deviation, including:

[0194] The process parameter data corresponding to the initial population is input into the pre-trained surrogate model to obtain the predicted processing time, the mean of the deviation, and the standard deviation of the deviation.

[0195] The calculation of the fitness value based on the predicted profile deviation includes:

[0196] The fitness value is calculated based on the predicted profile deviation and the predicted processing time.

[0197] Input process parameters: Input the process parameter data of each individual in the initial population into the pre-trained surrogate model.

[0198] The surrogate model outputs the following predicted values:

[0199] Predicted processing time: Predicts the processing time corresponding to this set of process parameters.

[0200] Mean of out-of-tolerance: The predicted mean of the profile deviation corresponding to this set of process parameters.

[0201] Standard deviation of out-of-tolerance: The standard deviation of the profile deviation corresponding to the set of process parameters.

[0202] Profile deviation prediction: The profile deviation prediction can be expressed as a combination of the mean and standard deviation of the deviation, for example:

[0203] Profile deviation prediction = mean of deviation + α × standard deviation of deviation

[0204] Here, α is a weighting coefficient used to adjust the degree of influence of the standard deviation.

[0205] The mean and standard deviation of the deviation can be converted into a probability distribution, such as a normal distribution, and the expected value can be calculated as the predicted profile deviation. The fitness value is then calculated based on the predicted profile deviation and the predicted processing time, for example:

[0206] Fitness value = w1 × predicted profile deviation + w2 × predicted processing time;

[0207] Among them, w1 and w2 are weighting coefficients used to adjust the relative importance of different objectives.

[0208] The fitness value can also be defined as an objective function, such as minimizing the weighted sum of profile deviation and processing time, or maximizing processing efficiency.

[0209] Proxy models can quickly predict the results of a large number of process parameter combinations, avoiding the time and cost of actual processing. Trained using historical data, Proxy models can capture the complex relationship between process parameters and processing results, improving prediction accuracy. The fitness value calculation method and optimization algorithm can be flexibly adjusted according to different optimization objectives and constraints.

[0210] This embodiment utilizes a pre-trained surrogate model to evaluate and optimize process parameters, enabling it to efficiently and accurately find the combination of process parameters that meets processing requirements, thereby improving processing efficiency and product quality.

[0211] In this embodiment, by acquiring the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted, the starting point and range of optimization can be clearly defined, avoiding blind search and thus improving optimization efficiency. Based on the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized. Each individual in the initial population is used to indicate a set of process parameter data. The initial individuals cover the key areas of the parameter space, which helps improve the initial accuracy of the model. The process parameter data corresponding to the initial population is input into a pre-trained surrogate model to obtain the contour deviation prediction value. The surrogate model can quickly predict the impact of process parameters on processing quality, avoiding complex physical simulations or experiments. The fitness value is calculated based on the contour deviation prediction value. The fitness value is used to describe the process performance of the process parameter data indicated by the individual. The smaller the fitness value, the better the process parameter data and the better the processing effect. The initial population is iteratively optimized based on the fitness value, with the maximization of the fitness value as the iterative optimization objective, until the number of iterations reaches a preset number, and the process parameter data of the target individual is output. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual. The introduction of fitness values ​​makes the optimization results more comprehensive, taking into account both the average level and fluctuation of the deviation. The optimized process parameters can improve processing efficiency and reduce processing time, thereby improving production efficiency.

[0212] To better explain the embodiments of this application, a specific example is provided below:

[0213] Currently, we are optimizing the machining parameters for aircraft blades, including the following four parameters: spindle speed, feed rate, depth of cut, and path spacing. The goal is to use a genetic algorithm to find the optimal combination of parameters that minimizes the profile deviation of the machined blades.

[0214] First, pass in the parameter boundaries and known parameters:

[0215] The known parameters are: spindle speed (SS) = 1500 RPM.

[0216] Parameter boundaries: Spindle speed (SS): 1000 RPM to 2000 RPM;

[0217] Feed rate (FR): 50 mm / min to 150 mm / min;

[0218] Depth of cut (CD): 0.1 mm to 0.5 mm;

[0219] Path spacing (PS): 0.05mm to 0.2mm;

[0220] Then, the population for the genetic algorithm is initialized:

[0221] Assuming a population size of 10, meaning initializing 10 individuals, each representing a combination of parameters, the initial population might look like this:

[0222] Individual 1: [1500 RPM, 80 mm / min, 0.2 mm, 0.1 mm];

[0223] Individual 2: [1200 RPM, 100 mm / min, 0.3 mm, 0.15 mm];

[0224] Individual 3: [1800 RPM, 120 mm / min, 0.4 mm, 0.08 mm];

[0225] ...

[0226] Individual 10: [1600 RPM, 90 mm / min, 0.25 mm, 0.12 mm];

[0227] Next, the fitness of each individual is calculated using the predictions from the neural network surrogate model. The goal of the fitness function is to minimize the process performance metrics (the sum of the mean, standard deviation, and processing time of the out-of-tolerance).

[0228] Assume the silhouette bias predicted by the neural network model is:

[0229] Individual 1: Mean = 0.05 mm, Standard Deviation = 0.02 mm

[0230] Individual 2: Mean = 0.08 mm, Standard Deviation = 0.03 mm

[0231] Individual 3: Mean = 0.04 mm, Standard Deviation = 0.01 mm ...

[0233] Individual 10: Mean = 0.06 mm, Standard Deviation = 0.025 mm

[0234] Calculate the fitness value: Fitness value = -(mean + standard deviation)

[0235] For example: The fitness value of individual 1 = -(0.05 + 0.02) = -0.07;

[0236] The fitness value of individual 2 = -(0.08 + 0.03) = -0.11;

[0237] The fitness value of individual 3 = -(0.04 + 0.01) = -(0.05); ...

[0239] The fitness value of individual 10 is -(0.06+0.025) = -0.085.

[0240] Next, select individuals with higher fitness as parents, and select the top 5 individuals with the lowest fitness values ​​as parents. For example, select individuals 3, 1, 10, 5, and 7 as parents.

[0241] New individuals are generated through a crossover operation. Assume a single-point crossover is used, where the crossover point is the middle position of the parameter. For example, individual 3 and individual 1 are crossed to generate a new individual:

[0242] Individual 3: [1800 RPM, 120 mm / min, 0.4 mm, 0.08 mm]

[0243] Individual 1: [1500 RPM, 80 mm / min, 0.2 mm, 0.1 mm]

[0244] Crossover generates new individuals: [1800 RPM, 120 mm / min, 0.2 mm, 0.1 mm]

[0245] Randomness is introduced through mutation to ensure population diversity. Assume a mutation probability of 10%, meaning each parameter has a 10% chance of mutating.

[0246] For example, a new individual [1800 RPM, 120 mm / min, 0.2 mm, 0.1 mm] exhibits variation:

[0247] The mutated individual: [1800 RPM, 120 mm / min, 0.25 mm, 0.1 mm]

[0248] Repeat the selection, crossover, and mutation operations described above for multiple generations of iteration, gradually evolving to approach the optimal solution. Assume that after 10 generations of iteration, the individuals in the population gradually converge to the optimal solution.

[0249] Finally, 4. Output the optimal parameter combination:

[0250] The individual with the lowest fitness value is ultimately output as the optimal combination of process parameters. Assuming that after multiple iterations, individual 3 has the lowest fitness value, the final parameter combination for individual 3 is:

[0251] Optimal process parameter combination: [1800 RPM, 120 mm / min, 0.4 mm, 0.08 mm]

[0252] In this embodiment, by acquiring the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted, the starting point and range of optimization can be clearly defined, avoiding blind search and thus improving optimization efficiency. Based on the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized. Each individual in the initial population is used to indicate a set of process parameter data. The initial individuals cover the key areas of the parameter space, which helps improve the initial accuracy of the model. The process parameter data corresponding to the initial population is input into a pre-trained surrogate model to obtain the contour deviation prediction value. The surrogate model can quickly predict the impact of process parameters on processing quality, avoiding complex physical simulations or experiments. The fitness value is calculated based on the contour deviation prediction value. The fitness value is used to describe the process performance of the process parameter data indicated by the individual. The smaller the fitness value, the better the process parameter data and the better the processing effect. The initial population is iteratively optimized based on the fitness value, with the maximization of the fitness value as the iterative optimization objective, until the number of iterations reaches a preset number, and the process parameter data of the target individual is output. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual. The introduction of fitness values ​​makes the optimization results more comprehensive, taking into account both the average level and fluctuation of the deviation. The optimized process parameters can improve processing efficiency and reduce processing time, thereby improving production efficiency.

[0253] Based on the above Figure 1 The method for determining process parameter data shown in the present application also provides a device for determining process parameter data, such as... Figure 8 As shown, the device 800 for determining the process parameter data may include:

[0254] The acquisition module 810 is used to acquire the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted;

[0255] The initialization module 820 is used to initialize the population of the genetic algorithm according to the preset parameter data and parameter boundary data. Each individual in the initial population is used to indicate a set of process parameter data.

[0256] The input module 830 is used to input the process parameter data corresponding to the initial population into the pre-trained surrogate model to obtain the contour deviation prediction value; the contour deviation prediction value is used to describe the difference between the contour data of the processed workpiece obtained based on the process parameter data corresponding to the initial population and the preset contour data.

[0257] The calculation module 840 is used to calculate the fitness value based on the profile deviation prediction value; the fitness value is used to describe the process performance of the process parameter data indicated by the individual.

[0258] The iteration module 850 is used to optimize the initial population based on the fitness value, with the goal of maximizing the fitness value, until the number of iterations reaches a preset number. It outputs the process parameter data of the target individual, where the fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual.

[0259] In one possible embodiment, the input module 830 is specifically used for:

[0260] The process parameter data corresponding to the initial population is input into the pre-trained surrogate model to obtain the mean and standard deviation of the deviation; the deviation is calculated based on the contour data of the processed workpiece obtained from the process parameter data corresponding to the initial population and the preset contour data; the preset contour data is the contour data of the preset workpiece.

[0261] Calculation module 840 is specifically used for:

[0262] The fitness value is calculated based on the mean and standard deviation of the deviation.

[0263] In one possible embodiment, the device 800 may further include:

[0264] The design module is used to design M experimental schemes based on N process parameters using the orthogonal experimental design method, where N and M are both positive integers; each experimental scheme corresponds to a set of experimental process parameter data.

[0265] The experiment module is used to acquire the processing result data of the workpiece. The processing result data is obtained by conducting processing experiments based on M experimental schemes.

[0266] The normalization module is used to normalize the experimental process parameter data to obtain the normalized experimental process parameter data.

[0267] The generation module is used to generate a feature matrix based on the normalized experimental process parameter data. The feature matrix has a dimension of M*N.

[0268] The generation module is used to generate a target matrix based on the processing result data of the workpiece. The target matrix has a dimension of M*2.

[0269] The training module is used to train a neural network model based on the feature matrix and the target matrix until the neural network model meets the training stopping condition, thus obtaining a trained surrogate model.

[0270] In one possible embodiment, the generation module is specifically used for:

[0271] Based on the processing result data of the workpiece and the preset detection points on the workpiece, determine the contour deviation data of the workpiece;

[0272] For each preset detection point, the deviation is calculated based on the contour deviation data of the processed workpiece;

[0273] For each preset detection point, the deviation in the corresponding direction is calculated based on the predetermined tolerance and deviation;

[0274] For each detection point with out-of-tolerance, the mean and standard deviation of the out-of-tolerance are calculated and used as the target matrix of the neural network model. The dimension of the target matrix is ​​M*2.

[0275] In one possible embodiment, the device 800 may further include:

[0276] The building blocks are used to construct neural network models with multiple layers, including an input layer, hidden layers, fully connected layers, and an output layer. Each fully connected layer uses L2 regularization and Dropout mechanisms, and each fully connected layer uses the ReLU activation function. The output layer uses a linear activation function.

[0277] A training module for training a neural network model to determine the parameters of the trained neural network layers at each of the plurality of layers.

[0278] A determination module is used to determine the trained neural network model as a surrogate model.

[0279] In one possible embodiment, the iteration module 850 is specifically used for:

[0280] The initial population is optimized iteratively based on fitness values, with the goal of maximizing fitness values. The individual with the highest fitness value is selected as the new parent.

[0281] New individuals are generated through crossover and mutation operations;

[0282] A new population is generated based on the new parent generation and the new individuals;

[0283] The process parameter data corresponding to the new population is input into the pre-trained surrogate model to obtain the profile deviation prediction value; and the fitness value is calculated based on the profile deviation prediction value; and the new population is optimized and iterated based on the fitness value until the number of iterations reaches the preset number.

[0284] In one possible embodiment, the experimental process parameter data includes: spindle speed, feed rate, depth of cut, and path spacing.

[0285] In this embodiment, by acquiring the preset parameter data of the preset process parameters and the parameter boundary data of the process parameters to be predicted, the starting point and range of optimization can be clearly defined, avoiding blind search and thus improving optimization efficiency. Based on the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized. Each individual in the initial population is used to indicate a set of process parameter data. The initial individuals cover the key areas of the parameter space, which helps improve the initial accuracy of the model. The process parameter data corresponding to the initial population is input into a pre-trained surrogate model to obtain the contour deviation prediction value. The surrogate model can quickly predict the impact of process parameters on processing quality, avoiding complex physical simulations or experiments. The fitness value is calculated based on the contour deviation prediction value. The fitness value is used to describe the process performance of the process parameter data indicated by the individual. The smaller the fitness value, the better the process parameter data and the better the processing effect. The initial population is iteratively optimized based on the fitness value, with the maximization of the fitness value as the iterative optimization objective, until the number of iterations reaches a preset number, and the process parameter data of the target individual is output. The fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any other individual. The introduction of fitness values ​​makes the optimization results more comprehensive, taking into account both the average level and fluctuation of the deviation. The optimized process parameters can improve processing efficiency and reduce processing time, thereby improving production efficiency.

[0286] Figure 9 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application is shown.

[0287] An electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0288] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0289] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory. In a particular embodiment, memory 302 includes read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0290] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the process parameter data determination methods in the embodiment shown in the figure.

[0291] In one example, the electronic device may also include a communication interface 303 and a bus 310. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.

[0292] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0293] Bus 310 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0294] This electronic device can execute the method for determining process parameter data in the embodiments of this application, thereby achieving the combination Figure 1 The method for determining the process parameter data described.

[0295] Furthermore, in conjunction with the method for determining process parameter data in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement… Figure 1 Methods for determining process parameter data.

[0296] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0297] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0298] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0299] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for determining process parameter data, characterized in that: The method comprises: Acquiring preset parameter data of a preset process parameter and parameter boundary data of a process parameter to be predicted; According to the preset parameter data and parameter boundary data, the population of the genetic algorithm is initialized, and each individual in the obtained initial population is used to indicate a set of process parameter data; The process parameter data corresponding to the initial population is input into the pre-trained proxy model to obtain a profile deviation prediction value; the profile deviation prediction value is used to describe the difference between the profile data of the workpiece obtained by processing based on the process parameter data corresponding to the initial population and the preset profile data; The fitness value is calculated according to the predicted value of the profile deviation; the fitness value is used to describe the process performance of the process parameter data indicated by the individual; The initial population is optimized and iterated according to the fitness value, with maximization of the fitness value as the iterative optimization goal, until the number of iterations reaches a preset number, and the process parameter data of the target individual is output, and the fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any of the other individuals.

2. The method according to claim 1, characterized in that The process parameter data corresponding to the initial population is input into the pre-trained proxy model to obtain the profile deviation prediction value, including: The process parameter data corresponding to the initial population is input into the pre-trained proxy model to obtain the mean value of the deviation and the standard deviation of the deviation; the deviation is calculated based on the contour data of the processed workpiece obtained by processing the process parameter data corresponding to the initial population and the preset contour data; the preset contour data is the contour data of the preset workpiece; The step of calculating the fitness value according to the profile deviation prediction value comprises: The fitness value is calculated based on the mean of the excess and the standard deviation of the excess.

3. The method according to claim 2, characterized in that Before inputting the process parameter data corresponding to the initial population into the pre-trained proxy model to obtain the profile deviation prediction value, the method further includes: Based on the orthogonal experimental design method, M experimental schemes are designed according to N process parameters, where N and M are both positive integers; each experimental scheme corresponds to a set of experimental process parameter data of process parameters; the experimental process parameter data include: spindle speed, feed speed, cutting depth and path spacing; Acquire processing result data of the processed workpiece, where the processing result data is obtained by performing processing experiments according to M experimental schemes; Normalizing the experimental process parameter data to obtain normalized experimental process parameter data; Generate a feature matrix based on the normalized experimental process parameter data, the dimension of the feature matrix is ​​M*N; Generate a target matrix based on the processing result data of the workpiece, and the dimension of the target matrix is ​​M*2; The neural network model is trained according to the feature matrix and the target matrix until the neural network model meets the training stop condition to obtain a trained proxy model.

4. The method according to claim 3, characterized in that The step of generating a target matrix according to the processing result data of the processing workpiece includes: Determine the contour deviation data of the processed workpiece according to the processing result data of the processed workpiece and the preset detection points on the processed workpiece; For each preset detection point, the deviation is calculated based on the profile deviation data of the processed workpiece; For each preset inspection point, the deviation in the corresponding direction is calculated according to the preset tolerance and deviation; For the deviation of each detection point, the mean and standard deviation of the deviation are calculated as the target matrix of the neural network model, and the dimension of the target matrix is ​​M*2.

5. The method according to claim 3, characterized in that: The method further comprises: training the neural network model according to the feature matrix and the target matrix until the neural network model meets the training stop condition and obtains the proxy model. Construct a neural network model with multiple layers, including input layer, hidden layer, fully connected layer, and output layer; each fully connected layer uses L2 regularization mechanism and Dropout mechanism, each fully connected layer uses ReLU activation function, and the output layer uses linear activation function; training a neural network model to determine parameters of a trained plurality of neural network layers at each of the plurality of levels; The trained neural network model is determined as a proxy model.

6. The method according to claim 1, characterized in that The optimizing and iterating the initial population according to the fitness value, taking the maximization of the fitness value as the iterative optimization goal, until the number of iterations reaches a preset number, includes: The initial population is optimized and iterated according to the fitness value, with the maximization of the fitness value as the iterative optimization goal, and the individual with the highest fitness value is selected as the new parent; Generate new individuals through crossover and mutation operations; Generate a new population based on new parents and new individuals; Inputting the process parameter data corresponding to the new population into the pre-trained proxy model to obtain the contour deviation prediction value; and calculating the fitness value according to the contour deviation prediction value; and optimizing the new population according to the fitness value until the number of iterations reaches a preset number.

7. The method according to claim 3, characterized in that The process parameter data corresponding to the initial population is input into the pre-trained proxy model to obtain the mean value of the deviation and the standard deviation of the deviation, including: The process parameter data corresponding to the initial population is input into the pre-trained proxy model to obtain the processing time prediction value, the mean value of the deviation and the standard deviation of the deviation; The step of calculating the fitness value according to the profile deviation prediction value comprises: The fitness value is calculated based on the predicted value of contour deviation and the predicted value of processing time.

8. A device for determining process parameter data, characterized in that: The device for determining the process parameter data comprises: An acquisition module, used for acquiring preset parameter data of a preset process parameter and parameter boundary data of a process parameter to be predicted; An initialization module is used to initialize the population of the genetic algorithm according to preset parameter data and parameter boundary data, and each individual in the obtained initial population is used to indicate a set of process parameter data; An input module is used to input the process parameter data corresponding to the initial population into the pre-trained proxy model to obtain a profile deviation prediction value; the profile deviation prediction value is used to describe the difference between the profile data of the machined workpiece obtained based on the process parameter data corresponding to the initial population and the preset profile data; A calculation module, used for calculating a fitness value according to a profile deviation prediction value; the fitness value is used for describing a process performance of process parameter data indicated by an individual; The iterative module is used to optimize and iterate the initial population according to the fitness value, with the maximization of the fitness value as the iterative optimization goal, until the number of iterations reaches a preset number, and outputs the process parameter data of the target individual, and the fitness value corresponding to the process parameter data indicated by the target individual is greater than the fitness value corresponding to the process parameter data indicated by any of the other individuals.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for determining process parameter data according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for determining process parameter data according to any one of claims 1 to 7 is implemented.

Images