Boring process multi-objective collaborative optimization method and system based on RBF neural network and NSGA-II
By combining response surface model, radial basis function neural network and non-dominated sorting genetic algorithm, the problems of experience dependence and multi-objective optimization imbalance in boring process parameter optimization are solved. This achieves synergistic optimization of high precision, low roughness and long tool life, improving machining quality and production efficiency.
Patent Information
- Application Number
- CN202510979049.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-16
- Publication Date
- 2025-11-28
AI Technical Summary
Existing methods for optimizing boring process parameters rely on manual experience, resulting in low model fitting accuracy, imbalance in multi-objective optimization, and difficulty in achieving synergistic optimization of high precision, low roughness, and long tool life.
A multi-objective optimization model was constructed by combining response surface methodology (RSM), radial basis function neural network (RBF), and nondominated sorting genetic algorithm (NSGA-II). Through single-factor experiments, Box-Behnken experimental design, nonlinear regression, and multi-objective optimization, the synergistic optimization of hole cylindricity, surface roughness, and tool wear value was achieved.
It significantly improves machining accuracy and reduces machining costs. The cylindricity fluctuation of the hole is reduced from ±15% to ±5%, ensuring a balance between hole cylindricity, surface roughness and tool wear, thereby improving machining quality and production efficiency.
Smart Images
Figure CN121031276A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the field of aviation manufacturing and precision machining, and relates to a boring process multi-objective collaborative optimization method and system based on an RBF neural network and an NSGA-II. BACKGROUND
[0002] Boring is a core machining process of a helicopter precision part, and machining precision directly determines assembly fit (tolerance ≤ 5 μm), fatigue life and safety performance of the whole machine. In the field of aviation manufacturing, micron-level hole cylindricity (≤ 10 μm) and sub-micron-level surface roughness (≤ 800 nm) have become hard quality indicators, and tool wear control (≤ 250 μm) is also required. However, the existing boring process parameter optimization method has the following significant limitations: (1) Traditional experience dependence and single-target limitation: Process parameters (such as linear cutting speed, feed speed and cutting depth) have long relied on artificial experience, lack of scientific quantitative basis, and machining precision fluctuation range is as high as ± 15%. Single-factor experiment method or traditional response surface method only focuses on single target optimization (such as cylindricity), ignores the complex nonlinear coupling effect between parameters, and it is difficult to balance multi-objective conflicts (for example, improving machining precision often accompanies tool wear aggravation).
[0003] (2) Model precision and optimization algorithm defects: The existing quadratic polynomial model has insufficient fitting ability for high-dimensional nonlinear relationship, and the prediction error is large, which is difficult to accurately guide parameter optimization. In addition, local optimization algorithm is easy to fall into suboptimal solution, and does not consider the processing dynamic characteristics, which leads to tool life fluctuation of more than 20%, and batch production consistency is difficult to guarantee.
[0004] (3) Multi-objective collaborative optimization demand is urgent: Aviation part manufacturing requires high precision, low roughness and long tool life at the same time, but the existing technology lacks a systematic solution. For example, when the traditional method optimizes the hole cylindricity, the surface roughness may deteriorate to more than 1000 nm, and the tool wear value exceeds 300 μm, which significantly affects the machining economy and reliability. SUMMARY
[0005] The technical problem to be solved by the present application is to solve the problems of relying on manual experience, low model fitting accuracy and multi-objective optimization imbalance in the traditional boring process parameter optimization method, and to provide a boring process parameter optimization method based on an intelligent optimization algorithm. The method combines response surface model (RSM), radial basis function neural network (RBF) and non-dominated sorting genetic algorithm (NSGA-II), realizes multi-objective collaborative optimization of multiple processing performance indicators (such as hole cylindricity, surface roughness and tool wear value), and achieves the goal of improving quality, reducing cost and increasing efficiency.
[0006] The technical scheme of the present application is In a first aspect, a boring process multi-objective collaborative optimization method based on RBF neural network and NSGA-II is provided, comprising: Step 1: The key parameters in the boring process are studied through single-factor experiments, the value range of each parameter is changed one by one, other parameters are kept unchanged, and the influence of each parameter on the processing performance index is observed, so as to determine the feasible interval of each parameter and provide a reasonable parameter range for subsequent optimization; Step 2: Based on the determination of the parameter feasible interval, 17 groups of experimental matrices are constructed by using Box-Behnken experimental design BBD in response surface, the data of process parameters and processing performance indicators are collected, the associated data set of parameters and processing performance indicators is established, and data support is provided for subsequent multi-objective optimization; Step 3: Based on the response surface experimental data, a nonlinear regression model between the parameters and the processing performance indicators is constructed by using the response surface model RSM, and the nonlinear influence law of the parameters on the processing performance indicators is revealed; Step 4: Based on the response surface experimental data, the nonlinear relationship between the processing performance indicators and the process parameters is fitted by using support vector regression SVR and radial basis function neural network RBF; Step 5: Analyze the nonlinear influence law and the nonlinear relationship, use the trained RBF neural network model as the prediction model for optimization, and obtain a set of Pareto optimal solutions based on the non-dominated sorting genetic algorithm II; Step 6: The processing performance indicators in the Pareto optimal solution set are normalized, the normalized results are weighted and summed to construct a comprehensive evaluation formula, the minimum value of which is selected to filter out the optimal parameter combination; Step 7: The selected optimal parameter combination is used for verification experiment to evaluate the actual processing effect.
[0007] Further, the key parameters include cutting speed , feed speed , cutting depth ; the processing performance indicators include hole cylindricity , surface roughness , tool wear value ; Key parameters as response surface factors; processing performance indicators as response surface responses.
[0008] Further, the form of the response surface model is as follows:
[0009] wherein, y and x j represent the response value and the factor respectively, , , , represent the constant term coefficient, the linear term coefficient, the quadratic term coefficient and the interaction term coefficient respectively, and k is the number of response surface factors.
[0010] Further, the method further comprises: When constructing the RBF neural network, the number of neurons and its parameters are adjusted adaptively, and the spread coefficient is used to control the response range of each neuron; In the training process, the error target is set to 1e-5, and the error is required to meet the requirements until the model can accurately fit the complex nonlinear relationship between the process parameters and the processing performance.
[0011] Further, the comprehensive evaluation F opt The expression of the function is as follows: ; wherein, , , are the weights of each target, reflecting the relative importance of each index to the overall optimization target; is the normalized value of hole cylindricity, is the normalized value of surface roughness, is the normalized value of tool wear value.
[0012] In the second aspect, a boring process multi-objective collaborative optimization system based on RBF neural network and NSGA-II is provided, comprising: a data interaction module, an experimental design module, a model training module, an optimization calculation module and a result display module; The data interaction module is used to: acquire and process key input parameters of the boring process, including cutting speed , feed speed , cutting depth and other process boundary conditions, while collecting corresponding processing performance indicators, including hole cylindricity , surface roughness and tool wear value ; The experimental design module is used for: based on the Box-Behnken experimental design BBD method of the response surface, generating a reasonable experimental matrix, and performing experiments, collecting machining performance data under different process parameters; The model training module is used for: based on the experimental data and the collected machining performance indicators, training multiple data models, realizing the nonlinear fitting between the parameters and the machining performance indicators; The optimization calculation module is used for: using the non-dominated sorting genetic algorithm NSGA-II to perform multi-objective optimization on multiple objectives output by the model; through the optimization algorithm, the Pareto front solution set is generated by balancing between multiple objectives; The result display module is used for: outputting the optimal parameter combination, and drawing the Pareto optimal solution graph, the convergence speed curve and the stability analysis result, to verify the optimization result.
[0013] Further, the data model in the model training module includes: The response surface model RSM is used to describe the nonlinear relationship between the process parameters and the machining performance, and to establish a mathematical model between the parameters and the response value; The support vector regression SVR is used to fit the data through the support vector machine, especially for complex nonlinear data modeling; The radial basis function neural network RBF uses the RBF network for efficient nonlinear modeling, and improves the fitting accuracy through an adaptive learning algorithm.
[0014] The beneficial effects of the present application are: the present application realizes the multi-objective collaborative optimization of the boring process parameters by combining the response surface model (RSM), the support vector regression (SVR), the radial basis function neural network (RBF) and the non-dominated sorting genetic algorithm (NSGA-II), significantly improves the machining precision, prolongs the tool life and reduces the machining cost. Compared with the traditional method, the machining fluctuation of the hole cylindricality is reduced from ±15% to ±5%, and the machining precision is greatly improved; at the same time, through the multi-objective optimization, the performance imbalance in the single objective optimization is avoided, and the best balance of the hole cylindricality, the surface roughness and the tool wear value is ensured. The method not only effectively improves the machining quality and the production efficiency, but also significantly reduces the production cost, provides a scientific and efficient optimization scheme for the high-precision machining field such as aviation manufacturing and mold manufacturing, and has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS
[0015] Figure 1 The boring process optimization method flowchart based on the intelligent optimization algorithm provided by the embodiment of the present application; Figure 2(a) is a response surface between hole cylindricity and cutting speed, feed speed Figure 1 ; Figure 2(b) is a response surface between hole cylindricity and cutting speed, cutting depth Figure 2; Figure 2(c) is a response surface between hole cylindricity and feed speed, cutting depth Figure 3 ; Figure 2(d) is a response surface between surface roughness and cutting speed, feed speed Figure 1 ; Figure 2(e) is a response surface between surface roughness and cutting speed, cutting depth Figure 2; Figure 2(f) is a response surface between surface roughness and feed speed, cutting depth Figure 3 ; Figure 2(g) is a response surface between tool wear value and cutting speed, feed speed Figure 1 ; Figure 2(h) is a response surface between tool wear value and cutting speed, cutting depth Figure 2; Figure 2(i) is a response surface between tool wear value and feed speed, cutting depth Figure 3 ; Figure 3 A radial basis function neural network (RBF) model schematic diagram provided by an embodiment of the present application; Figure 4 A non-dominated sorting genetic algorithm (NSGA-II) optimization flowchart provided by an embodiment of the present application; Figure 5 A boring process optimization system diagram based on an intelligent optimization algorithm provided by an embodiment of the present application. DETAILED DESCRIPTION
[0016] The present application provides a process optimization method based on an intelligent optimization algorithm, comprising: Step 1: Through single-factor experiments, the key process parameters in the boring process are studied, the value range of each parameter is changed one by one, other parameters are kept unchanged, the influence of the same on the processing performance index is observed, and the feasible interval of each parameter is determined to provide a reasonable parameter range for subsequent optimization; Step 2: On the basis of determining the parameter feasible interval, 17 groups of experimental matrices are constructed by using Box-Behnken experimental design (BBD) in response surface, data of process parameters and processing performance indexes are collected, and an associated data set of parameters and processing performance indexes is established to provide data support for subsequent multi-objective optimization; Step 3: Based on the response surface experimental data, a nonlinear regression model between parameters and processing performance indexes is constructed by using a response surface model (RSM), and the nonlinear influence law of parameters on processing performance indexes is revealed. Step 4: Use machine learning algorithms such as Support Vector Regression (SVR), Radial Basis Function Neural Network (RBF), etc. to fit the non-linear relationship between the machining performance indicators and the process parameters. Step 5: Use the trained RBF neural network model as the prediction model for optimization, and use the Non-dominated Sorting Genetic Algorithm II (NSGA-II) for multi-objective optimization to obtain a set of Pareto optimal solutions.
[0017] Step 6: Normalize the machining performance indicators in the Pareto optimal solution set, and construct a comprehensive evaluation formula by weighted sum of the normalized results to minimize the combined value, and select the optimal parameter combination.
[0018] Step 7: Perform a verification experiment using the selected optimal parameter combination to evaluate the actual machining effect.
[0019] Further, the key parameters in step one include: cutting speed , feed speed , cutting depth ; Further, the machining performance indicators in step 1 include hole cylindricity , surface roughness , tool wear value , where hole cylindricity is measured using a dial gauge, surface roughness is measured using a MarSurf@PS 10 surface roughness measuring instrument, and tool wear value is measured using a body microscope.
[0020] Further, the response surface factors in step 2 are cutting speed , feed per tooth , cutting depth , and the response surface responses are hole cylindricity , surface roughness , and tool wear value , and the response surface model is established through the relationship between these variables.
[0021] Further, the form of the response surface model in step 3 is as follows:
[0022] wherein, y and x represent the response value and the factor, , , , represent the constant term coefficient, linear term coefficient, quadratic term coefficient and cross term coefficient respectively.
[0023] Further, the support vector regression (SVR) in step 4 seeks to fit the data by finding the best regression hyperplane, with the goal of minimizing the deviation of most data points from this hyperplane. The main steps are as follows: SVR maps input data to high-dimensional space through kernel functions to solve nonlinear problems; to enhance the robustness of the model, SVR introduces slack variables , allowing for errors within a certain range; by selecting support vectors (i.e., points with errors within the tolerance range), SVR defines the regression function and ensures good prediction ability for new data; the goal of SVR is to minimize the loss function, finding the optimal regression hyperplane that balances fitting errors and model complexity.
[0024] When using SVR for model training, automatic optimization of hyperparameters is set, allowing the BoxConstraint and KernelScale hyperparameters of the model to automatically adjust based on the data, optimizing the fitting accuracy of the model. In addition, using expected improvement as the optimization criterion further improves the prediction performance.
[0025] Further, the radial basis function neural network (RBF) in step 4 is a type of feedforward neural network, which uses radial basis functions (Gaussian functions) as activation functions in the hidden layer to map input data. The core steps are as follows: Input process parameters are mapped to high-dimensional space through Gaussian radial basis functions, and each input sample has a corresponding weight with each neuron in the hidden layer, which determines the distance between the input data and the neuron; the neurons in the hidden layer use radial basis functions as activation functions to calculate the distance between the input point and the neuron center to obtain the response value; based on the calculated response value, the output of each hidden layer neuron is weighted and summed to form the final prediction value. The form of the Gaussian function is as follows:
[0026] where, x is the input sample, c i is the i center, is the width parameter, which determines the range of the response; When constructing the RBF neural network, the number of neurons and their parameters are adjusted adaptively, and the spread coefficient is used to control the response range of each neuron. During training, the error target is set to 1e-5, and the error is required to meet the requirements until the model can accurately fit the complex nonlinear relationship between process parameters and machining performance.
[0027] Further, the NSGA-II algorithm in step 5 is based on the selection, crossover and mutation operations of genetic algorithm, but unlike the traditional genetic algorithm, NSGA-II adopts non-dominated sorting and crowding distance in the selection process to handle multi-objective problems; Among them, the non-dominated sorting is to sort the solution set according to the relative advantages and disadvantages of the objective function, ensure that each solution is not dominated by other solutions in the target space, and form multiple non-dominated frontiers; the crowding distance calculates the crowding degree (i.e. the distance between solutions) of each solution in the target space to maintain the diversity of the population and prevent solutions from concentrating in a certain area of the target space.
[0028] Further, the normalization processing formula in step 6 is as follows:
[0029] Among them is the normalized value, and min and max are the minimum and maximum values of the target index in the entire solution set.
[0030] Further, the expression of the comprehensive evaluation function in step 6 is as follows:
[0031] Among them, , , are the weights of each target, reflecting the relative importance of each index to the overall optimization target.
[0032] The present application provides a process optimization system based on intelligent optimization algorithm, and the system structure is as follows: Data interaction module: used for obtaining and processing the key input parameters of boring process, including cutting speed , feed speed , cutting depth and other process boundary conditions, and collecting corresponding processing performance indicators, including hole cylindricity , surface roughness and tool wear value , etc.
[0033] Experimental design module: based on response surface Box-Behnken experimental design (BBD) and other methods, generate reasonable experimental matrix and conduct experiments to collect processing performance data under different process parameters.
[0034] Model training module: based on experimental data and collected processing performance indicators, train multiple data models to realize nonlinear fitting between parameters and processing performance indicators; Optimization calculation module: use non-dominated sorting genetic algorithm (NSGA-II) to perform multi-objective optimization on multiple objectives output by the model. Through the optimization algorithm, the Pareto frontier solution set is generated by balancing between multiple objectives; Result display module: output the optimal parameter combination, and draw the Pareto optimal solution graph, convergence speed curve and stability analysis result to verify the optimization result.
[0035] Further, the data model in the model training module includes: Response surface model (RSM): used to describe the nonlinear relationship between process parameters and processing performance, and to establish a mathematical model between parameters and response values; Support vector regression (SVR): data fitting is performed by support vector machine, which is particularly suitable for complex nonlinear data modeling; Radial basis function neural network (RBF): high-efficiency nonlinear modeling is performed by using RBF network, and the fitting accuracy is improved by using adaptive learning algorithm.
[0036] The following will be described in detail in combination with the drawings the tool wear monitoring method based on multi-feature extraction and transfer learning provided by the present application.
[0037] Figure 1 The bore process optimization method provided by the embodiment of the present application is a flow chart of the bore process optimization method based on intelligent optimization algorithm, and the bore process optimization method includes: Step 1: Through single-factor experiment, the key parameters affecting the processing performance in the bore processing are studied, including cutting speed , feed speed , cutting depth . In each experiment, keep two parameters unchanged, adjust one parameter at a time, and observe the influence on hole cylindricity, surface roughness and tool wear, so as to determine the feasible range of each parameter. During the experiment, the hole cylindricity, surface roughness and tool wear value under each group of experimental settings are recorded in detail, which provides data support for the subsequent optimization model.
[0038] Step 2: On the basis of determining the feasible range of parameters, the Box-Behnken design (BBD) method is used for response surface experiment, 17 groups of experimental matrix are designed, covering different combinations of cutting speed , feed speed , cutting depth . Through the experiment, the data between the process parameters and the processing performance indicators are collected and used to establish a data correlation matrix.
[0039] Fig. 2(a) is a response surface plot of hole cylindricity and cutting speed, feed speed; Fig. 2(b) is a response surface plot of hole cylindricity and cutting speed, cutting depth; Fig. 2(c) is a response surface plot of hole cylindricity and feed speed, cutting depth; Fig. 2(d) is a response surface plot of surface roughness and cutting speed, feed speed; Fig. 2(e) is a response surface plot of surface roughness and cutting speed, cutting depth; Fig. 2(f) is a response surface plot of surface roughness and feed speed, cutting depth; Fig. 2(g) is a response surface plot of tool wear value and cutting speed, feed speed; Fig. 2(h) is a response surface plot of tool wear value and cutting speed, cutting depth; Fig. 2(i) is a response surface plot of tool wear value and feed speed, cutting depth.
[0040] Step 3: Regression analysis is performed on the experimental data by using the response surface method (RSM), and a nonlinear regression model between the process parameters and the processing performance indicators is established. These models reveal the influence law of each process parameter on the hole cylindricity, surface roughness and tool wear value.
[0041] Further, the hole cylindricity in step 1 is measured by using a dial gauge, the surface roughness is measured by using a MarSurf@PS 10 surface roughness measuring instrument, and the tool wear value is measured by using a body microscope.
[0042] Figure 3 A radial basis function neural network (RBF) model diagram provided for the embodiment of the application is shown in the figure; Step 4: Machine learning algorithms such as support vector regression (SVR) and radial basis function neural network (RBF) are used to fit the nonlinear relationship, and the prediction accuracy of the model is further improved.
[0043] Figure 4 A non-dominated sorting genetic algorithm (NSGA-II) optimization flowchart provided for the embodiment of the application is shown in the figure; Step 5: The trained RBF neural network model is used in combination with the non-dominated sorting genetic algorithm II (NSGA-II) to optimize multiple objectives, and a set of Pareto frontier optimal solutions is finally obtained.
[0044] Step 6: The processing performance indicators in the Pareto optimal solution set are normalized, and a comprehensive evaluation formula is constructed by weighted summation according to the weights of each objective, so as to screen out the optimal parameter combination.
[0045] Step 7: The selected optimal parameter combination is used for verification experiment to evaluate the effect in the actual processing process.
[0046] Figure 5The bore hole process optimization system based on intelligent optimization algorithm provided by the embodiment of the application is shown in the figure.
[0047] In another aspect, the bore hole process optimization system based on intelligent optimization algorithm comprises a plurality of modules, the main modules of the system and their functions are as follows: Data interaction module: this module is responsible for obtaining and processing key input parameters in the bore hole process, such as cutting speed , feed speed , cutting depth and other process boundary conditions. At the same time, the corresponding processing performance indicators are collected, including hole cylindricity , surface roughness and tool wear value , etc.
[0048] Experimental design module: according to the Box-Behnken experimental design (BBD) method of response surface method, a reasonable experimental matrix is designed, and experiments are performed according to the design matrix. Through these experiments, processing performance data under different process parameters are collected.
[0049] Model training module: through the analysis of experimental data and processing performance indicators, multiple data models are trained to realize the nonlinear fitting between process parameters and processing performance. The prediction models in this part include response surface model (RSM), support vector regression (SVR) and radial basis function neural network (RBF).
[0050] Optimization calculation module: non-dominated sorting genetic algorithm (NSGA-II) is used to perform multi-objective optimization on multiple objectives. This module balances between multiple objectives to generate a Pareto optimal solution set, and through normalized weighted processing, a set of optimal process parameter combinations are found.
[0051] Result display module: responsible for outputting the optimization results, including the optimal parameter combination, and displaying the Pareto optimal solution graph, convergence speed curve and stability analysis results. Through these visual results, users can intuitively evaluate the optimization effect.
[0052] Further, the data correlation matrix of process parameters and processing performance indicators in step 2 is shown in Table 1: Table 1
[0053] Further, the nonlinear regression model between hole cylindricity and cutting process parameters in step 3 is as follows:
[0054] Further, the nonlinear regression model between surface roughness and cutting process parameters in step 3 is as follows:
[0055] Further, the nonlinear regression model between tool wear value and cutting process parameters in step 3 is as follows:
[0056] Further, the SVR algorithm in step 4 models the experimental data, and the main steps are as follows: In the SVR modeling process, the input data is first mapped to a high-dimensional space through a kernel function to solve the nonlinear problem. SVR allows a certain range of errors by introducing slack variables, thereby enhancing the robustness of the model. The model selects support vectors (i.e., points with errors within the tolerance range), constructs a regression function, and ensures that the function has good prediction ability for new data.
[0057] where SVR automatically adjusts hyperparameters such as BoxConstraint and KernelScale to minimize the loss function while balancing fitting errors and model complexity. The expected improvement criterion is further improved to enhance prediction performance, ensuring the efficiency and accuracy of the SVR model.
[0058] Further, in step 4, in addition to using the SVR algorithm, RBF neural network is also used for model modeling. RBF network maps input data to high-dimensional space and uses Gaussian function as activation function for nonlinear mapping. The core idea of this network is to use the distance between each input data point and the center of the hidden layer neuron to calculate the response value and weighted sum to output the final result. The form of the Gaussian function is as follows:
[0059] where, x is the input sample, c i is the i center, is the width parameter that determines the range of response; where the training process of the RBF model includes adaptive adjustment of neuron number and parameters to ensure that the network can accurately fit the nonlinear relationship between process parameters and machining performance. During training, the error objective (such as 1e-5) is optimized to ensure that the network can achieve good fitting accuracy under complex data. In this process, the spread coefficient is used to adjust the response range of each neuron to improve the model's adaptability to different input data.
[0060] where RSM, SVR, and RBF are used to fit the relationship between machining performance indicators and machining parameters, and the performance indicators of the models are shown in Table 2: Table 2
[0061] In summary, based on the experimental data, the RBF model performs better than RSM and SVR in various performance indicators, especially in mean square error (MSE), root mean square error (RMSE), and determination coefficient (R²), showing higher prediction accuracy and stronger generalization ability, proving its significant advantage in fitting the nonlinear relationship between processing performance indicators and processing parameters.
[0062] Further, in step 5, the RBF neural network model trained in step 4 is used as a prediction model, and the NAGA-II algorithm is used for multi-objective optimization. Multiple evolved solutions are evolved through selection, crossover, and mutation. At the same time, the solutions in the current population are non-dominantly sorted and the crowding distance is calculated to generate a solution set containing multiple Pareto optimal solutions.
[0063] Further, in step 6, the processing performance indicators in the Pareto optimal solution set are normalized, and the value of each target indicator is converted to the standard range of [0, 1], so that each indicator has the same quantitative standard. The normalization formula is as follows:
[0064] wherein, is the processing performance indicator in each solution, and min and max are the minimum and maximum values of the target indicator in the entire solution set, respectively.
[0065] After normalization, the targets are weighted and summed to construct a comprehensive evaluation formula, which can obtain a comprehensive evaluation value, so as to screen out the optimal parameter combination and ensure the best balance between the targets. In this example, the expression is as follows:
[0066] wherein, in this example, the weights of hole cylindricity, surface roughness, and tool wear value are set to 0.4, 0.4, and 0.2, respectively. The optimal processing parameter combination after normalization and weighting is: cutting speed is 11.5215 m / min, feed speed is 0.0633 mm / rev, and cutting depth is 0.0235 mm.
Claims
1. A multi-objective collaborative optimization method for boring processes based on RBF neural networks and NSGA-II, characterized in that, include: Step 1: The key process parameters in the boring process are studied through single-factor experiments. The value range of each parameter is changed one by one while keeping other parameters unchanged. The impact on the machining performance index is observed to determine the feasible range of each parameter and provide a reasonable parameter range for subsequent optimization. Step 2: Based on the determination of the feasible range of parameters, 17 sets of experimental matrices are constructed using the Box-Behnken experimental design (BBD) in response surface methodology. Data on process parameters and processing performance indicators are collected, and a correlation dataset between parameters and processing performance indicators is established to provide data support for subsequent multi-objective optimization. Step 3: Based on the experimental data of the associated dataset, construct a nonlinear regression model between parameters and processing performance indicators using the response surface methodology (RSM) to reveal the nonlinear influence of parameters on processing performance indicators. Step 4: Based on the response surface experimental data, use support vector regression (SVR) and radial basis function neural network (RBF) to fit the nonlinear relationship between processing performance indicators and process parameters; Step 5: Analyze the nonlinear influence and nonlinear relationship, and use the trained RBF neural network model as the optimized prediction model; perform multi-objective optimization based on the non-dominated sorting genetic algorithm II to obtain a set of Pareto optimal solutions; Step 6: Normalize the processing performance indices in the Pareto optimal solution set, and use a weighted summation of the normalization results to construct a comprehensive evaluation formula to minimize the sum value and select the optimal parameter combination. Step 7: Conduct a verification experiment using the selected optimal parameter combination to evaluate the actual processing effect.
2. The method according to claim 1, characterized in that, Key parameters include: cutting speed Feed rate Depth of cut Machining performance indicators include hole cylindricity. Surface roughness Tool wear value ; Key parameters are used as response surface factors; processing performance indicators are used as response surface responses.
3. The method according to claim 2, characterized in that, The response surface model takes the following form: in, y and x j These represent the response value and the factor, respectively. , , , These represent the constant term coefficient, linear term coefficient, quadratic term coefficient, and interaction term coefficient, respectively, and k is the number of response surface factors.
4. The method according to claim 3, characterized in that, The method further includes: When constructing an RBF neural network, the number of neurons and their parameters are adaptively adjusted, and the spread coefficient is used to control the response range of each neuron. During training, the error target was set to 1e-5 until the error met the requirements, ensuring that the model could accurately fit the complex nonlinear relationship between process parameters and processing performance.
5. The method according to claim 4, characterized in that, Overall evaluation F opt The function expression is as follows: ； in, , , These are the weights of each objective, reflecting the relative importance of each indicator to the overall optimization objective; It is the normalized value of the cylindricity of the hole. The normalized value of surface roughness This is the normalized value of the tool wear.
6. A multi-objective collaborative optimization system for boring processes based on RBF neural networks and NSGA-II, characterized in that, include: The module includes a data interaction module, an experimental design module, a model training module, an optimization calculation module, and a results display module. The data interaction module is used to acquire and process key input parameters of the boring process, including cutting speed. Feed rate Depth of cut Equal process boundary conditions, and simultaneously collect corresponding machining performance indicators, including hole cylindricity. Surface roughness and tool wear value ; The experimental design module is used to: use the Box-Behnken experimental design method based on response surface methodology (BBD) to generate a reasonable experimental matrix, conduct experiments, and collect processing performance data under different process parameters; The model training module is used to train multiple data models based on experimental data and collected processing performance indicators, and to achieve nonlinear fitting between parameters and processing performance indicators. The optimization computation module is used to: perform multi-objective optimization on multiple objectives of the model output using the non-dominated sorting genetic algorithm NSGA-II; and generate a Pareto front solution set by weighing the trade-offs among multiple objectives through the optimization algorithm. The results display module is used to output the optimal parameter combination, and simultaneously plot the Pareto optimal solution graph, convergence rate curve, and stability analysis results to verify the optimization results.
7. The system according to claim 6, characterized in that, The data models in the model training module include: Response surface methodology (RSM) is used to describe the nonlinear relationship between process parameters and processing performance, and to establish a mathematical model between parameters and response values. Support Vector Regression (SVR) fits data using support vector machines, and is especially useful for modeling complex nonlinear data. Radial Basis Function Neural Network (RBF) utilizes RBF networks for efficient nonlinear modeling and improves fitting accuracy through adaptive learning algorithms.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the method described in any one of claims 1-5.