Micro-groove parameter recommendation method, device and equipment of abrasive water jet process and medium

Abstract

The application discloses a micro groove parameter recommendation method, device, equipment and medium of an abrasive water jet process, and the method comprises the following steps: S1, obtaining an experimental data set of abrasive water jet processing based on a Latin hypercube sampling method; S2, training a Gaussian process regression model using a plurality of alternative covariance kernel functions, and screening a plurality of kernel functions with high performance based on a multi-criteria scoring mechanism; S3, taking the plurality of Gaussian process regression models screened as base models, taking the prediction mean and the prediction standard deviation as meta-features, and obtaining a trained Gaussian process regression model average prediction model by adopting Lasso regression; and S4, obtaining target micro groove geometric profile parameters expected by a user, taking the trained Gaussian process regression model average prediction model as a target function, and searching to obtain an optimal process parameter combination. The application has higher prediction accuracy and stability, can effectively guide actual production, and significantly reduces trial-and-error costs of process development.

Description

Technical Field

[0001] This application relates to the fields of precision machining and intelligent manufacturing technology, and in particular to a method, apparatus, equipment and medium for recommending microgroove parameters in abrasive waterjet processes. Background Technology

[0002] Abrasive waterjet machining technology has broad application prospects in the microstructure processing of hard and brittle materials (such as quartz glass) due to its advantages such as no heat-affected zone and wide material adaptability. The geometric profile of the microgroove (including groove depth, half-width at half-maximum, groove width, and tilt angle) is a key indicator for evaluating processing quality, and it has a complex nonlinear mapping relationship with multiple process parameters such as jet angle, target distance, moving speed, abrasive concentration, and abrasive grain diameter.

[0003] Traditional methods for determining process parameters rely on extensive trial-and-error experiments, which are costly and inefficient. Existing technologies include some studies using response surface methodology (RSM) to establish polynomial models between process parameters and processing results; however, these models have limited ability to characterize complex nonlinear relationships and suffer from insufficient prediction accuracy. Other studies employ single-task Gaussian process regression (GPR) for prediction, but GPR models with a single kernel function exhibit instability under different data distributions and have limited generalization ability. Furthermore, existing methods largely follow a "predicting results from parameters" model, lacking a solution that can intelligently and efficiently recommend optimal process parameters based on the user's desired microgroove profile.

[0004] Therefore, developing an intelligent process parameter recommendation method with high prediction accuracy and good stability, capable of recommending process parameters based on the microgroove target profile, is of great significance for improving the intelligence level and processing efficiency of abrasive waterjet machining. Summary of the Invention

[0005] This application provides a method for recommending microgroove parameters in abrasive waterjet processes, which solves the technical problem that existing technologies cannot intelligently and efficiently recommend optimal process parameters based on the user's desired microgroove profile.

[0006] This application is achieved through the following solution: The recommended method for microgroove parameters in abrasive waterjet processes includes the following steps: S1. An experimental dataset for abrasive waterjet processing is obtained based on the Latin hypercube sampling method. The dataset includes process parameters and corresponding microgroove geometric profile parameters. S2. Train Gaussian process regression models using multiple alternative covariance kernel functions, and select the top-performing kernel functions based on a multi-criteria scoring mechanism. S3. Using the selected Gaussian process regression models as base models, their predicted mean and predicted standard deviation are used as meta-features. Lasso regression is used as a meta-learner for training to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the micro-groove geometric profile parameter based on the input process parameters. S4. Obtain the target microgroove geometric profile parameters expected by the user, use the average prediction model of the trained Gaussian process regression model as the objective function, and adopt a strategy that combines global exploration, clustering and local optimization to search for the optimal combination of process parameters.

[0007] Further, step S1 specifically includes the following steps: S11. Based on the Latin hypercube sampling (LHS) method, uniformly distributed experimental sample points are generated within a preset range of process parameter values. The process parameters include at least the jet angle, target distance, moving speed, abrasive concentration, and abrasive diameter. S12. Conduct abrasive waterjet processing experiments, measure the microgroove geometric profile parameters corresponding to each sample point, and construct a training dataset. The microgroove geometric profile parameters include groove depth, half-width at half-height, groove width, and tilt angle.

[0008] Furthermore, step S2 specifically includes the following steps: S21. Select multiple alternative covariance kernel functions and train the corresponding Gaussian process regression models using the training dataset; S23. Design a multi-criteria scoring mechanism to comprehensively score the performance of each trained Gaussian process regression model. The scoring mechanism comprehensively considers log marginal likelihood, cross-validation mean square error, prediction interval coverage and residual diagnostic score. S23. Based on the scoring results, select several kernel functions with the highest performance ranking as candidate basic kernel functions.

[0009] Furthermore, step S3 specifically includes the following steps: S31. Using multiple selected candidate Gaussian process regression models as base models, their predicted mean and predicted standard deviation for the same validation set are used as meta-features. The Lasso regression algorithm is used as a meta-learner to train the meta-features, learn the optimal fusion weights of each base model under different working conditions, and obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the microgroove geometry based on the input process parameters.

[0010] Furthermore, step S31 specifically includes the following steps: S311. Data preprocessing: Divide the original dataset into training and validation sets, and standardize the input features (zero mean, unit variance), while saving the standardization parameters for subsequent inverse standardization. S312. Base model training: Using the selected optimal kernel functions, construct multiple independent Gaussian process regression models as base models. For each base model, use Bayesian hyperparameter optimization (objective function is to minimize cross-validation MSE) to find the optimal hyperparameters (such as signal variance, length scale, noise variance, etc.) on the training set. After obtaining the optimal hyperparameters, train the Gaussian process regression model on the complete training set and record its prediction results on the validation set. S313. Construct the meta-feature matrix. For each sample in the validation set, each base model outputs the predicted mean and predicted standard deviation. These outputs are concatenated into a multi-dimensional meta-feature vector. The meta-feature vectors of all validation set samples constitute the meta-feature matrix. S314. Meta-model training: Lasso regression is used as the meta-learner and an optimization objective is established for training. After training, the final trained Gaussian process regression model average prediction model is obtained.

[0011] Furthermore, step S4 specifically includes the following steps: S41. Obtain the target microgroove geometry parameters expected by the user; S42. Embed the trained Gaussian process regression model average prediction model into the optimization framework as the objective function; S43. A strategy combining global exploration and local optimization is adopted. First, global candidate points are screened through Latin hypercube sampling. Then, local constraint optimization is performed on the potential regions after clustering. Within the process parameter constraint range, the optimal process parameter combination is searched to find the process parameter combination that minimizes the error between the predicted value of the microgroove geometry output by the Gaussian process regression model and the target microgroove geometry parameters.

[0012] Further, step S43 specifically includes the following steps: S431. In the global exploration phase, within the constraints of various process parameters, multiple candidate parameter points are generated using Latin hypercube sampling. The Gaussian process regression model is used to quickly predict the predicted value of the microgroove geometry profile of each candidate point. The weighted mean square error between the predicted value and the target microgroove geometry profile parameter is calculated, and the top N points with the smallest mean square error are selected as the potential point set. S432. Cluster analysis: DBSCAN clustering is performed on the potential point set to obtain several clusters. Each cluster represents a high-potential region in the parameter space, thereby identifying multiple potential optimal regions. S433. In the local optimization stage, the center point of each potential optimal region is used as the initial value, and the L-BFGS-B constraint optimization algorithm is used to perform a local fine search, and finally multiple candidate optimal solutions are obtained. S434. Output and Verification: Output multiple combinations of process parameters that minimize the error between the predicted value of the microgroove geometry and the target microgroove geometry parameters as the optimal combination of process parameters.

[0013] This application also provides a device for recommending microgroove parameters for abrasive waterjet processes, including: The data acquisition module is used to acquire experimental datasets for abrasive waterjet processing based on the Latin hypercube sampling method. The datasets include process parameters and corresponding microgroove geometric profile parameters. The model selection module is used to train Gaussian process regression models using multiple alternative covariance kernel functions, and selects several kernel functions with the best performance based on a multi-criteria scoring mechanism. The prediction model training module is used to train a selected Gaussian process regression model as a base model, using its prediction mean and prediction standard deviation as meta-features, and Lasso regression as a meta-learner to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted values ​​of the microgroove geometric profile parameters based on the input process parameters. The optimal process parameter combination output module is used to obtain the target microgroove geometric profile parameters expected by the user. The average prediction model of the trained Gaussian process regression model is used as the objective function. The optimal process parameter combination is obtained by using a strategy that combines global exploration, clustering and local optimization.

[0014] This application also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the microgroove parameter recommendation method for the abrasive waterjet process.

[0015] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the microgroove parameter recommendation method for the abrasive waterjet process.

[0016] This application also provides a computer program product, including a computer program or computer-executable instructions, which, when executed by a processor, implement the microgroove parameter recommendation method for the abrasive waterjet process.

[0017] Compared with the prior art, this application can produce the following beneficial effects: This application provides a method, apparatus, equipment, and medium for recommending microgroove parameters in abrasive waterjet processes. The method uses a multi-criteria scoring mechanism (e.g., comprehensively considering logarithmic marginal likelihood, cross-validation error, prediction interval coverage, and residual diagnosis) to screen Gaussian process regression models with various kernel functions. Furthermore, Lasso regression is used to average the selected excellent base models, constructing an average prediction model of the Gaussian process regression model. Since a single kernel function can only make prior assumptions about the function space from a fixed perspective, while Lasso regression can automatically learn the optimal fusion weights of each base model under different operating conditions and compress redundant model weights to zero, it overcomes the defect of unstable performance of a single model under different data distributions. Therefore, compared with traditional second-order response surface methodology or a single Gaussian process regression model, this application has higher prediction accuracy and... The generalization ability is significantly improved, and it can output predictive confidence intervals with interpretability. At the same time, the method embeds the Gaussian process regression model average prediction model into the optimization framework and adopts a two-stage strategy of "global sampling (Latin hypercube sampling) + clustering (DBSCAN) + local optimization (L-BFGS-B)" to search for the optimal process parameters in reverse. Because the objective function of the Gaussian process regression model average prediction model (GPR-FMA) is non-convex and multimodal, direct gradient descent is prone to getting trapped in local optima. Global sampling can pre-locate the optimal region, clustering can identify multiple candidate regions, and local optimization can quickly converge by using second-order gradient information. Therefore, the recommendation method of this application can intelligently recommend the optimal combination of process parameters according to the microgroove profile expected by the user. Experiments show that the average absolute percentage error between the measured profile and the expected profile is only 2.85%, which significantly reduces the trial and error cost and time of process development.

[0018] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. A further detailed description of this application will be provided below with reference to the figures. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart illustrating the method for recommending microgroove parameters in the abrasive waterjet process according to a preferred embodiment of this application. Figure 2This is a flowchart illustrating a method for recommending microgroove parameters in an abrasive waterjet process according to another preferred embodiment of this application. Figure 3 This is a schematic diagram of the architecture for kernel function selection and Gaussian process regression model average prediction model training in a preferred embodiment of this application; Figure 4 It is a scatter plot (groove depth) of the output results of the polynomial response surface (PRS) model; Figure 5 This is another scatter plot (full width at half maximum) of the output of the polynomial response surface (PRS) model; Figure 6 This is another scatter plot (slot width) of the output results of the polynomial response surface (PRS) model; Figure 7 This is another scatter plot (tilt angle) of the output results of the polynomial response surface (PRS) model; Figure 8 This is another scatter plot (groove depth) of the microgroove profile prediction results using the method of this application; Figure 9 This is another scatter plot (half-width at half maximum) of the microgroove profile prediction results using the method of this application; Figure 10 This is another scatter plot (groove width) of the microgroove profile prediction results using the method of this application; Figure 11 This is another scatter plot (tilt angle) of the microgroove profile prediction results using the method of this application; Figure 12 This is a schematic diagram illustrating the principle of the process parameter recommendation model of a preferred embodiment of this application; Figure 13 This is a schematic diagram of the microgroove parameter recommendation device for the abrasive waterjet process according to a preferred embodiment of this application; Figure 14 This is a schematic block diagram of an electronic device according to a preferred embodiment of this application; Figure 15 This is a schematic diagram of the internal structure of a computer device according to a preferred embodiment of this application. Detailed Implementation

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] It should be noted that the execution subject of this embodiment can be a computing service system with data processing, network communication and program running functions, such as a tablet computer, personal computer, mobile phone, etc., or a micro-groove parameter recommendation device for abrasive waterjet process that can achieve the above functions. The following uses the micro-groove parameter recommendation device for abrasive waterjet process as the execution subject to describe this embodiment and the following embodiments.

[0023] like Figure 1 As shown, to address the aforementioned technical problems, a preferred embodiment of this application provides a method for recommending microgroove parameters in abrasive waterjet processes, comprising the following steps: S1. An experimental dataset for abrasive waterjet processing is obtained based on the Latin hypercube sampling method. The dataset includes process parameters and corresponding microgroove geometric profile parameters. S2. Train Gaussian process regression models using multiple alternative covariance kernel functions, and select the top-performing kernel functions based on a multi-criteria scoring mechanism. S3. Using the selected Gaussian process regression models as base models, their predicted mean and predicted standard deviation are used as meta-features. Lasso regression is used as a meta-learner for training to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the micro-groove geometric profile parameter based on the input process parameters. S4. Obtain the target microgroove geometric profile parameters expected by the user, use the average prediction model of the trained Gaussian process regression model as the objective function, and adopt a strategy that combines global exploration, clustering and local optimization to search for the optimal combination of process parameters.

[0024] This embodiment provides a method for recommending microgroove parameters in abrasive waterjet processes. The method uses a multi-criteria scoring mechanism (e.g., comprehensively considering logarithmic marginal likelihood, cross-validation error, prediction interval coverage, and residual diagnosis) to screen Gaussian process regression models with various kernel functions. Furthermore, Lasso regression is used to average the selected excellent base models, constructing an average prediction model of the Gaussian process regression model. Since a single kernel function can only make prior assumptions about the function space from a fixed perspective, while Lasso regression can automatically learn the optimal fusion weights of each base model under different operating conditions and compress redundant model weights to zero, it overcomes the performance instability of a single model under different data distributions. Therefore, compared to traditional second-order response surface methodology or a single Gaussian process regression model, this embodiment demonstrates superior prediction accuracy and generalization ability for microgroove profiles (groove depth, half-width at half-maximum, groove width, and tilt angle). Both methods achieve significant improvements and can output interpretable prediction confidence intervals. Furthermore, this method embeds the Gaussian process regression model's average prediction model into the optimization framework, employing a two-stage strategy of "global sampling (Latin hypercube sampling) + clustering (DBSCAN) + local optimization (L-BFGS-B)" to inversely search for optimal process parameters. Because the objective function of the Gaussian process regression model's average prediction model (GPR-FMA) is non-convex and multimodal, direct gradient descent is prone to getting trapped in local optima. Global sampling can pre-locate the optimal region, clustering can identify multiple candidate regions, and local optimization utilizes second-order gradient information for rapid convergence. Therefore, the recommendation method in this embodiment can intelligently recommend the optimal combination of process parameters based on the user's desired microgroove profile. Experiments verify that the average absolute percentage error between the measured profile and the desired profile is only 2.85%, significantly reducing the trial-and-error cost and time in process development.

[0025] Preferably, step S1 specifically includes the following steps: S11. Based on the Latin hypercube sampling (LHS) method, uniformly distributed experimental sample points are generated within a preset range of process parameter values. The process parameters include at least the jet angle, target distance, moving speed, abrasive concentration, and abrasive diameter. S12. Conduct abrasive waterjet processing experiments, measure the microgroove geometric profile parameters corresponding to each sample point, and construct a training dataset. The microgroove geometric profile parameters include groove depth, half-width at half-height, groove width, and tilt angle.

[0026] Preferably, step S2 specifically includes the following steps: S21. Select multiple alternative covariance kernel functions and train the corresponding Gaussian process regression (GPR) models using the training dataset. S23. Design a multi-criteria scoring mechanism to comprehensively score the performance of each trained Gaussian process regression model. The scoring mechanism comprehensively considers log marginal likelihood, cross-validation mean square error, prediction interval coverage and residual diagnostic score. S23. Based on the scoring results, select several kernel functions with the highest performance ranking as candidate basic kernel functions.

[0027] Preferably, step S3 specifically includes the following steps: S31. Using multiple selected candidate Gaussian process regression models as base models, their predicted mean and predicted standard deviation for the same validation set are used as meta-features. The Lasso regression algorithm is used as a meta-learner to train the meta-features, learn the optimal fusion weights of each base model under different working conditions, and obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the microgroove geometry based on the input process parameters.

[0028] Preferably, step S31 specifically includes the following steps: S311. Data preprocessing: Divide the original dataset into training and validation sets, and standardize the input features (zero mean, unit variance), while saving the standardization parameters for subsequent inverse standardization. S312. Base model training: Using the selected optimal kernel functions, construct multiple independent Gaussian process regression models as base models. For each base model, use Bayesian hyperparameter optimization (objective function is to minimize cross-validation MSE) to find the optimal hyperparameters (such as signal variance, length scale, noise variance, etc.) on the training set. After obtaining the optimal hyperparameters, train the Gaussian process regression model on the complete training set and record its prediction results on the validation set. S313. Construct the meta-feature matrix. For each sample in the validation set, each base model outputs the predicted mean and predicted standard deviation. These outputs are concatenated into a multi-dimensional meta-feature vector. The meta-feature vectors of all validation set samples constitute the meta-feature matrix. S314. Meta-model training: Lasso regression is used as the meta-learner and an optimization objective is established for training. After training, the final trained Gaussian process regression model average prediction model is obtained.

[0029] Preferably, step S4 specifically includes the following steps: S41. Obtain the target microgroove geometry parameters expected by the user; S42. Embed the trained Gaussian process regression model average prediction model into the optimization framework as the objective function; S43. A strategy combining global exploration and local optimization is adopted. First, global candidate points are screened through Latin hypercube sampling. Then, local constraint optimization is performed on the potential regions after clustering. Within the process parameter constraint range, the optimal process parameter combination is searched to find the process parameter combination that minimizes the error between the predicted value of the microgroove geometry output by the Gaussian process regression model and the target microgroove geometry parameters.

[0030] Preferably, step S43 specifically includes the following steps: S431. In the global exploration phase, within the constraints of various process parameters, multiple candidate parameter points are generated using Latin hypercube sampling. The Gaussian process regression model is used to quickly predict the predicted value of the microgroove geometry profile of each candidate point. The weighted mean square error between the predicted value and the target microgroove geometry profile parameter is calculated, and the top N points with the smallest mean square error are selected as the potential point set. S432. Cluster analysis: DBSCAN clustering is performed on the potential point set to obtain several clusters. Each cluster represents a high-potential region in the parameter space, thereby identifying multiple potential optimal regions. S433. In the local optimization stage, the center point of each potential optimal region is used as the initial value, and the L-BFGS-B constraint optimization algorithm is used to perform a local fine search, and finally multiple candidate optimal solutions are obtained. S434. Output and Verification: Output multiple combinations of process parameters that minimize the error between the predicted value of the microgroove geometry and the target microgroove geometry parameters as the optimal combination of process parameters.

[0031] The following section uses the fabrication of microgrooves in quartz glass as an example to provide a more detailed explanation of this application.

[0032] Step 1: Experimental Design and Data Acquisition This embodiment uses quartz glass as the processing object and selects five key process parameters: jet angle (A), target distance (B), moving speed (C), abrasive concentration (D), and abrasive diameter (E). The value ranges of each parameter are as follows: A: 20-90°, B: 10-60mm, C: 1-3mm / min, D: 5-10g / L, E: 15, 27.5, 40μm.

[0033] To obtain high-quality training data, this embodiment employs the Latin Hypercube Sampling (LHS) method, and selects the optimal sampling scheme based on indicators such as the uniformity of sample point distribution, space filling, minimum distance, and center deviation. A total of 45 experiments were designed and conducted in this embodiment. Abrasive waterjet machining experiments were performed according to the LHS experimental scheme, and the groove depth of each machined microgroove was measured using a three-dimensional optical profilometer or microscope. d ), half height and width ( wh / 2 ), slot width ( w ) and tilt angle ( α A total of 45 sets of data were obtained, which were divided into training set according to a ratio (e.g., 80%), and the remainder were used as validation set.

[0034] Step 2: Construct a Gaussian process regression model based on kernel function performance evaluation Five candidate kernel functions were selected: RBF, Matérn (ν=2.5), Rational Quadratic, RBF_ARD, and RBF_Linear. Five GPR models were trained using the training set data.

[0035] To objectively evaluate the performance of each kernel function, a design was developed as follows: Figure 2 The multi-criteria scoring procedure shown calculates the following for each Gaussian process regression model on the validation set: Log-marginal likelihood (LML): measures how well a model fits the training data.

[0036] ; in, For the training set target value vector ( ), Input matrix for training set ( ), For kernel function hyperparameters, The covariance matrix is ​​calculated by the kernel function. For noise variance, The number of training samples, It is the identity matrix, and |•| represents the determinant of the matrix.

[0037] Log-marginal likelihood, also known as model evidence, is the probability of observed data occurring given hyperparameters. It is the most direct and comprehensive indicator for measuring the quality of GPR models, and therefore is given the highest weight.

[0038] Cross-validation mean squared error (CV-MSE): directly reflects the model's generalization ability.

[0039] ; in, For cross-validation folds, For the first i Mean squared error on the cross-validation set. Cross-validation directly evaluates the model's predictive ability on unseen data and is one of the most important evaluation criteria for generalization ability in practical applications. Therefore, given the weights, it is second only to marginal likelihood.

[0040] Predicted Interval Coverage (PICP): Evaluates the accuracy of model uncertainty quantification, targeting a 95% confidence interval.

[0041] For the validation set samples, GPR not only provides the predicted mean It also provides the prediction standard deviation. Construct a 95% prediction interval and calculate the proportion of the true values ​​on the validation set that fall within this interval: ; in, Indicates the first in the verification set i The true value of each sample m To determine the number of samples in the validation set, As an indicator function, the ideal coverage should be 95%. The smaller the deviation, the more accurate the uncertainty quantification. Since uncertainty quantification has important value in many engineering applications, its weight is second only to log-marginal likelihood and cross-validation.

[0042] Residual diagnostic score: Calculate the skewness, kurtosis, and autocorrelation of the validation set residuals.

[0043] Based on validation set residuals First, standardize: ; in, and These are the mean and standard deviation of the residuals, respectively.

[0044] Skewness score: ; Where skewness is the degree of skewness. This is the skewness score.

[0045] Kurtosis score: ; Where kurtosis is the kurtosis. This is the kurtosis score.

[0046] Autocorrelation score: ; Where DW is the autocorrelation coefficient. This is the autocorrelation score.

[0047] The final residual diagnostic score is the average of the three: ; in, This represents the residual diagnostic score.

[0048] Since this indicator is more of an auxiliary check for model applicability, it has a relatively lower weight compared to the first three indicators.

[0049] Calculation time: ; in, The total time taken from the start of training to the completion of Bayesian optimization is given the lowest weight because offline training time is not a primary consideration. The only consideration is to avoid overly complex kernel functions from unnecessarily consuming resources.

[0050] The five indicators above were scored and weighted separately (e.g., LML: 0.35, CV-MSE: 0.3, PICP: 0.2, residual: 0.15, computation time 0.05), and the weighted sum was used to obtain the comprehensive score of each kernel function. In addition, to prevent overfitting, a complexity penalty term was introduced (penalty factor = 1 - number of effective parameters / number of samples). After evaluation, the RBF, Matérn (ν=2.5), and RBF_ARD kernel functions showed the best performance and were selected as candidate base models.

[0051] Step 3: Construct the Lasso-based Gaussian process regression model average prediction model (GPR-FMA model average prediction model) This step constructs a two-layer ensemble model, such as Figure 3 As shown, it includes the following sub-steps: Step 1: Data Preprocessing The original dataset is divided into a training set (80%) and a validation set (20%). The input features are standardized (zero mean, unit variance), and the standardization parameters are saved for subsequent destandardization.

[0052] Step 2: First Layer – Base Model Training. Using the three optimal kernel functions selected in Step 2 (RBF, Matérnv=2.5, RBF_ARD), three independent Gaussian process regression models are constructed. For each model, Bayesian hyperparameter optimization (objective function: minimizing cross-validation MSE) is used to find the optimal hyperparameters (such as signal variance, length scale, noise variance, etc.) on the training set. After obtaining the optimal hyperparameters, the Gaussian process regression model is trained on the complete training set, and its prediction results on the validation set are recorded.

[0053] Step 3: Constructing the meta-feature matrix For each sample in the validation set The three base models output the predicted mean respectively. , , and the predicted standard deviation , , These outputs are concatenated into a 6-dimensional meta-feature vector: ; The meta-feature matrix is ​​formed by the meta-feature vectors of all validation set samples. The corresponding true value vector is .

[0054] Step 4: Second Layer - Meta-model Training Using Lasso regression as the meta-learner, the optimization objective is: ; in, Let be the meta-weight vector to be learned. α Regularization coefficients (automatically selected through cross-validation); L1 regularization term It can compress the weights of unimportant base models to zero, thereby achieving sparse ensemble; after training, the final Gaussian process regression model average prediction model is obtained.

[0055] Step 5: Integrated Prediction For a new test sample First, the predicted mean of each of the three base models is calculated. , , and the predicted standard deviation , , This constitutes the meta-eigenvector: ; Then, the final predicted value is calculated using the Lasso meta-model: ; Performance comparison: The average prediction model of the polynomial response surface (PRS) model and the Gaussian process regression model in this embodiment were externally validated using an independent test set.

[0056] During validation, four evaluation metrics widely used in regression prediction tasks were employed—the coefficient of determination (R²). 2 The mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) are also included.

[0057] 1. Coefficient of determination R 2 (Coefficient of Determination) The coefficient of determination reflects the extent to which a regression model explains the variability of the target variable and is a commonly used statistic for measuring the goodness of fit of the model. ; in, For the first The true observed value of each sample For the model to the first The predicted value for each sample, The total number of samples.

[0058] 2. Mean Absolute Error (MAE) Mean Absolute Error (MAE) measures the average level of the absolute deviation between model predictions and actual observations. It reflects the actual magnitude of the prediction error. The formula for calculating MAE is: ; MAE Units and Raw Data Since they share the same units, they are intuitively interpretable.

[0059] 3. Mean Absolute Percentage Error (MAPE) The mean absolute percentage error (MASE) measures the relative deviation between the predicted and actual values ​​as a percentage, and its calculation formula is as follows: ; 4. Root Mean Square Error (RMSE) Root Mean Square Error (RMSE) is another commonly used error metric in regression models. It measures the deviation between predicted and actual values ​​by calculating the square root of the mean of the squared errors. The formula for RMSE is: ; Units of RMSE and raw data same.

[0060] The output of the polynomial response surface (PRS) model is as follows: Figures 4 to 7 The output of the Gaussian process regression model average prediction model (GPR-FMA model) is: Figures 8 to 11 The R-squared value of the Gaussian process regression model as the average prediction model across the four profile parameters. 2 The values ​​are all higher than those of the PRS model, especially with significant improvements in half-width at half-maximum (+13.5%), slot width (+32.3%), and tilt angle (+30.7%). Meanwhile, its RMSE and MAE are also lower, demonstrating the superiority of this method in prediction accuracy and stability.

[0061] Step 4: Construct a process parameter recommendation model This step implements the function of solving "from desired profile to optimal parameters", and the process is as follows: Figure 12 As shown, it includes the following sub-steps: Step 1: Obtain the target profile and configure optimization parameters First, obtain the user's desired microgroove profile, such as: groove depth 200μm, half-width at half-maximum 400μm, groove width 600μm, and tilt angle 45°. Simultaneously, set the constraint ranges for each process parameter (consistent with the range of Latin hypercube sampling in step 1), as well as the number of sampling points in the global exploration phase (e.g., 2000), the number of potential points screened (e.g., 50), the cluster radius parameter, and the weights of different profile parameters. Then, use the average prediction model of the Gaussian process regression model trained in step 3 as the objective function. This function is used to calculate the mean square error between the input process parameters and the target profile. Next, the user can choose to employ a two-stage optimization strategy to find the optimal solution.

[0062] Step Two: Global Exploration Phase Within the constraints of five process parameters, 2000 candidate parameter points were generated using Latin hypercube sampling. A Gaussian process regression model was used to quickly predict the profile of each candidate point, and its weighted mean square error (MSE) relative to the target profile was calculated. The top 50 points with the smallest MSE were selected as the potential point set.

[0063] Step 3: Cluster Analysis DBSCAN clustering is performed on the potential point set. DBSCAN automatically identifies multiple density-connected potential optimal regions and excludes outliers without requiring a pre-specified number of clusters. After clustering, several clusters are obtained, each representing a high-potential region in the parameter space.

[0064] Step 4: Local Optimization Stage DBSCAN clustering was performed on the potential point set to identify multiple potential optimal regions. Using the center point of each region as the initial value, a two-stage strategy of "global sampling (Latin hypercube sampling) + clustering (DBSCAN) + local optimization" was used to perform a local fine-grained search for the optimal process parameter constraint optimization algorithm. This algorithm utilizes the second-order gradient approximation information of the objective function, enabling it to quickly converge to the minimum point within the local region. Optimization was performed independently for each cluster center, ultimately yielding multiple candidate optimal solutions.

[0065] Step 5: Result Output and Validation Finally, the algorithm outputs several differentiated process parameter recommendations that minimize the error between the predicted profile and the target profile. For example, the recommended parameters are: angle 80°, target distance 39.86 mm, speed 2.35 mm / min, concentration 6.73 g / L, and particle size 32.75 μm.

[0066] Verification experiment: After the recommended parameters were rounded for engineering purposes, the mean absolute percentage error (MAPE) between the measured microgroove profile and the target profile was only 2.85%, which fully verified the engineering practicality and reliability of the method in this application.

[0067] like Figure 13 As described in another preferred embodiment of this application, a microgroove parameter recommendation device for abrasive waterjet technology is also provided, comprising: The data acquisition module is used to acquire experimental datasets for abrasive waterjet processing based on the Latin hypercube sampling method. The datasets include process parameters and corresponding microgroove geometric profile parameters. The model selection module is used to train Gaussian process regression models using multiple alternative covariance kernel functions, and selects several kernel functions with the best performance based on a multi-criteria scoring mechanism. The prediction model training module is used to train a selected Gaussian process regression model as a base model, using its prediction mean and prediction standard deviation as meta-features, and Lasso regression as a meta-learner to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted values ​​of the microgroove geometric profile parameters based on the input process parameters. The optimal process parameter combination output module is used to obtain the target microgroove geometric profile parameters expected by the user. The average prediction model of the trained Gaussian process regression model is used as the objective function. The optimal process parameter combination is obtained by using a strategy that combines global exploration, clustering and local optimization.

[0068] The microgroove parameter recommendation device for abrasive waterjet technology provided in this embodiment adopts the microgroove parameter recommendation method for abrasive waterjet technology in the above embodiments, solving the technical problem that existing technologies cannot intelligently and efficiently recommend optimal process parameters based on the user's desired microgroove profile. Compared with the prior art, the beneficial effects of the microgroove parameter recommendation device for abrasive waterjet technology provided in this embodiment are the same as those of the microgroove parameter recommendation method for abrasive waterjet technology provided in the above embodiments, and other technical features in the microgroove parameter recommendation device for abrasive waterjet technology are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0069] like Figure 14 As shown, a preferred embodiment of this embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the microgroove parameter recommendation method for the abrasive waterjet process in the above embodiment.

[0070] This embodiment also provides an electronic device that uses the microgroove parameter recommendation method for abrasive waterjet technology in the above embodiments. This solves the technical problem that the prior art cannot intelligently and efficiently recommend the optimal process parameters according to the microgroove profile expected by the user. Compared with the prior art, the beneficial effects of the electronic device provided in this embodiment are the same as those of the microgroove parameter recommendation method for abrasive waterjet technology provided in the above embodiments. Furthermore, other technical features of the electronic device are the same as those disclosed in the method of the above embodiments, and will not be repeated here.

[0071] like Figure 15 As shown, a preferred embodiment of this application also provides a computer device, which may be a terminal or a liveness detection server, and its internal structure diagram may be as follows. Figure 15 As shown. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with other external computer devices via a network connection. When the computer program is executed by the processor, it implements the steps of the recommended method for microgroove parameters in the aforementioned abrasive waterjet process.

[0072] Those skilled in the art will understand that Figure 15 The structure shown is merely a block diagram of a portion of the structure related to the solution of this embodiment, and does not constitute a limitation on the computer device to which the solution of this embodiment is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0073] The computer device provided in this embodiment adopts the microgroove parameter recommendation method for abrasive waterjet process in the above embodiment, which solves the technical problem that the prior art cannot intelligently and efficiently recommend the optimal process parameters according to the microgroove profile expected by the user. Compared with the prior art, the beneficial effects of the computer device provided in this embodiment are the same as the beneficial effects of the microgroove parameter recommendation method for abrasive waterjet process provided in the above embodiment, and other technical features in the electronic device are the same as the features disclosed in the method of the above embodiment, which will not be repeated here.

[0074] A preferred embodiment of this application also provides a storage medium including a stored program that, when the program is executed, controls the device containing the storage medium to perform the steps of the microgroove parameter recommendation method for the abrasive waterjet process in the above embodiments.

[0075] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0076] If the functions described in this embodiment are implemented as software functional units and sold or used as independent products, they can be stored in one or more computing device-readable storage media. Based on this understanding, the parts of this embodiment that contribute to the prior art or the technical solution can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computing device (which may be a personal computer, server, mobile computing device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this embodiment. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0077] Those skilled in the art will understand that the embodiments of this example can be provided as methods, systems, or computer program products. Therefore, this example can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this example can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code. The solutions in this example can be implemented using various computer languages, such as the object-oriented programming language C++ and the embedded programming language C.

[0078] This embodiment is described with reference to flowchart illustrations and / or block diagrams of the method, apparatus (system), and computer program product according to this embodiment. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the microgroove parameter recommendation method for the abrasive waterjet process as described above.

[0082] The computer program product provided in this embodiment solves the technical problem that existing technologies cannot intelligently and efficiently recommend optimal process parameters based on the user's desired microgroove profile. Compared with the prior art, the beneficial effects of the computer program product provided in this embodiment are the same as those of the microgroove parameter recommendation method for abrasive waterjet processes provided in the above embodiments, and will not be repeated here.

[0083] Obviously, those skilled in the art can make various modifications and variations to this embodiment without departing from the spirit and scope of this embodiment. Therefore, if these modifications and variations of this embodiment fall within the scope of the claims of this embodiment and their equivalents, this embodiment is also intended to include these modifications and variations.

Claims

1. A method for recommending microgroove parameters in abrasive waterjet processing, characterized in that, Includes the following steps: S1. An experimental dataset for abrasive waterjet processing is obtained based on the Latin hypercube sampling method. The dataset includes process parameters and corresponding microgroove geometric profile parameters. S2. Train Gaussian process regression models using multiple alternative covariance kernel functions, and select the top-performing kernel functions based on a multi-criteria scoring mechanism. S3. Using the selected Gaussian process regression models as base models, their predicted mean and predicted standard deviation are used as meta-features. Lasso regression is used as a meta-learner for training to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the micro-groove geometric profile parameter based on the input process parameters. S4. Obtain the target microgroove geometric profile parameters expected by the user, use the average prediction model of the trained Gaussian process regression model as the objective function, and adopt a strategy that combines global exploration, clustering and local optimization to search for the optimal combination of process parameters.

2. The method for recommending microgroove parameters for abrasive waterjet technology according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Based on the Latin hypercube sampling method, uniformly distributed experimental sample points are generated within the preset range of process parameter values. The process parameters include at least the jet angle, target distance, moving speed, abrasive concentration, and abrasive diameter. S12. Conduct abrasive waterjet processing experiments, measure the microgroove geometric profile parameters corresponding to each sample point, and construct a training dataset. The microgroove geometric profile parameters include groove depth, half-width at half-height, groove width, and tilt angle.

3. The method for recommending microgroove parameters in the abrasive waterjet process according to claim 2, characterized in that, Step S2 specifically includes the following steps: S21. Select multiple alternative covariance kernel functions and train the corresponding Gaussian process regression models using the training dataset; S23. Design a multi-criteria scoring mechanism to comprehensively score the performance of each trained Gaussian process regression model. The scoring mechanism comprehensively considers log marginal likelihood, cross-validation mean square error, prediction interval coverage and residual diagnostic score. S23. Based on the scoring results, select several kernel functions with the highest performance ranking as candidate basic kernel functions.

4. The method for recommending microgroove parameters for abrasive waterjet technology according to claim 3, characterized in that, Step S3 specifically includes the following steps: S31. Using multiple selected candidate Gaussian process regression models as base models, their predicted mean and predicted standard deviation for the same validation set are used as meta-features. The Lasso regression algorithm is used as a meta-learner to train the meta-features, learn the optimal fusion weights of each base model under different working conditions, and obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted value of the microgroove geometry based on the input process parameters.

5. The method for recommending microgroove parameters for abrasive waterjet technology according to claim 4, characterized in that, Step S31 specifically includes the following steps: S311. Data preprocessing: Divide the original dataset into training and validation sets, standardize the input features, and save the standardization parameters for subsequent inverse standardization. S312. Base model training: Using the selected multiple optimal kernel functions, construct multiple independent Gaussian process regression models as base models. For each base model, use Bayesian hyperparameter optimization to find the optimal hyperparameters on the training set. After obtaining the optimal hyperparameters, train the Gaussian process regression model on the complete training set and record its prediction results on the validation set. S313. Construct the meta-feature matrix. For each sample in the validation set, each base model outputs the predicted mean and predicted standard deviation. These outputs are concatenated into a multi-dimensional meta-feature vector. The meta-feature vectors of all validation set samples constitute the meta-feature matrix. S314. Meta-model training: Lasso regression is used as the meta-learner and an optimization objective is established for training. After training, the final trained Gaussian process regression model average prediction model is obtained.

6. The method for recommending microgroove parameters for abrasive waterjet technology according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Obtain the target microgroove geometry parameters expected by the user; S42. Embed the trained Gaussian process regression model average prediction model into the optimization framework as the objective function; S43. A strategy combining global exploration and local optimization is adopted. First, global candidate points are screened through Latin hypercube sampling. Then, local constraint optimization is performed on the potential regions after clustering. Within the process parameter constraint range, the optimal process parameter combination is searched to find the process parameter combination that minimizes the error between the predicted value of the microgroove geometry output by the Gaussian process regression model and the target microgroove geometry parameters.

7. The method for recommending microgroove parameters for abrasive waterjet technology according to claim 6, characterized in that, Step S43 specifically includes the following steps: S431. In the global exploration phase, within the constraints of various process parameters, multiple candidate parameter points are generated using Latin hypercube sampling. The Gaussian process regression model is used to quickly predict the predicted value of the microgroove geometry profile of each candidate point. The weighted mean square error between the predicted value and the target microgroove geometry profile parameter is calculated, and the top N points with the smallest mean square error are selected as the potential point set. S432. Cluster analysis: DBSCAN clustering is performed on the potential point set to obtain several clusters. Each cluster represents a high-potential region in the parameter space, thereby identifying multiple potential optimal regions. S433. In the local optimization stage, the center point of each potential optimal region is used as the initial value, and the L-BFGS-B constraint optimization algorithm is used to perform a local fine search, and finally multiple candidate optimal solutions are obtained. S434. Output and Verification: Output multiple combinations of process parameters that minimize the error between the predicted value of the microgroove geometry and the target microgroove geometry parameters as the optimal combination of process parameters.

8. A device for recommending microgroove parameters in abrasive waterjet processing, characterized in that, include: The data acquisition module is used to acquire experimental datasets for abrasive waterjet processing based on the Latin hypercube sampling method. The datasets include process parameters and corresponding microgroove geometric profile parameters. The model selection module is used to train Gaussian process regression models using multiple alternative covariance kernel functions, and selects several kernel functions with the best performance based on a multi-criteria scoring mechanism. The prediction model training module is used to train a selected Gaussian process regression model as a base model, using its prediction mean and prediction standard deviation as meta-features, and Lasso regression as a meta-learner to obtain the trained Gaussian process regression model average prediction model. The Gaussian process regression model average prediction model is used to output the predicted values ​​of the microgroove geometric profile parameters based on the input process parameters. The optimal process parameter combination output module is used to obtain the target microgroove geometric profile parameters expected by the user. The average prediction model of the trained Gaussian process regression model is used as the objective function. The optimal process parameter combination is obtained by using a strategy that combines global exploration, clustering and local optimization.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the microgroove parameter recommendation method for the abrasive waterjet process as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the microgroove parameter recommendation method for the abrasive waterjet process as described in any one of claims 1 to 8.

Images