A method for quickly predicting instability of ultrasonic peening forming based on GA-BP-ANN
By constructing a nonlinear mapping between shot peening process parameters and forming instability behavior using the GA-BP-ANN model, the problem of instability prediction during shot peening is solved, achieving efficient and accurate instability prediction, improving production efficiency and reducing costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI UNIV OF SCI & TECH
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies struggle to accurately predict instability during shot peening, resulting in low production efficiency and high costs. Furthermore, they rely on experiments and experience to adjust parameters, lacking real-time predictive capabilities.
A feedforward neural network (GA-BP-ANN) model based on genetic algorithm optimization was adopted to construct a nonlinear mapping relationship between shot peening process parameters and forming instability behavior. The model was trained with experimental data to achieve rapid prediction of whether the wall panel is unstable and the unstable region during ultrasonic shot peening forming.
It improves the accuracy of shot peening forming process parameter design and engineering application efficiency, reduces the number of experiments and numerical simulation calculations, and enhances prediction accuracy and practicality.
Smart Images
Figure CN122174612A_ABST
Abstract
Description
Technical Field
[0001] This invention pertains to sheet metal forming technology, specifically relating to a rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN. Background Technology
[0002] Shot peening is an adaptive flexible surface forming technology. Its core mechanism lies in the continuous impact of a high-speed shot stream on the surface of a metal component, inducing plastic deformation of the surface material and forming residual compressive stress, thereby achieving precise surface forming and enhanced mechanical properties of the component. In recent years, integrated forming technology for aircraft panels has matured, and shot peening technology, due to its significant advantages such as process flexibility, low cost, and no need for specialized molds, has been widely used in the forming and manufacturing of large thin-walled components such as aircraft wings and fuselage integral panels. Based on traditional shot peening technology, a variety of novel shot peening technologies have emerged in recent years, such as ultrasonic shot peening, laser shot peening, and waterjet shot peening. Among them, ultrasonic shot peening, which drives the shot motion through high-frequency vibration, can achieve high-strength impact effects under relatively low macroscopic load conditions, showing unique potential in improving forming efficiency and surface properties.
[0003] However, in the actual process of ultrasonic shot peening forming of thin-walled integral panels, the accumulation and uneven distribution of residual stress may lead to severe instability phenomena such as overall instability, local plastic buckling, and even wrinkling. The occurrence of overall or local instability will significantly reduce the curvature forming accuracy of the panel and seriously affect the mechanical properties of the formed part and subsequent correction parts. However, traditional methods have limited predictive ability for shot peening forming results. In actual production, the process parameters are often repeatedly adjusted by trial and error, combined with subsequent testing to verify the forming effect. This "produce first, test later" mode is inefficient and prone to product scrapping due to improper parameter settings. This is particularly unfavorable for high-value materials such as titanium alloys and high-strength aluminum alloys commonly used in the aerospace field, and will significantly increase manufacturing costs. Although traditional numerical simulation methods can simulate the shot peening process to a certain extent, their prediction accuracy is often limited due to the large number of model simplifications and complex assumptions, and their computational efficiency is difficult to meet the needs of real-time prediction and online process control. To address potential instability issues during shot peening, existing technologies typically rely on experimental design methods or operator experience to repeatedly adjust shot peening parameters (such as shot material, shot diameter, and peening time). These methods incur significant experimental and time costs, and the parameter determination process is highly subjective, making it difficult to accurately predict instability behavior during forming and resulting in low engineering feasibility.
[0004] In recent years, data-driven methods have provided new insights for modeling and predicting complex forming processes. Artificial Neural Networks (ANNs), with their excellent nonlinear mapping capabilities, can learn the intrinsic relationship between process parameters and forming results from experimental or numerical data without requiring explicit physical governing equations. Existing research has shown that neural networks have good application potential in predicting residual stress, surface roughness, and fatigue performance during shot peening; however, research on the overall prediction of unstable morphologies during shot peening remains relatively limited.
[0005] Therefore, this invention proposes a rapid prediction method for instability in ultrasonic shot peening forming based on GA-BP-ANN. This method constructs a nonlinear mapping relationship between shot peening process parameters and forming instability behavior to predict whether instability occurs in the panel during ultrasonic shot peening forming and the distribution of the instability region. This effectively reduces the number of experimental trials and the computational load of numerical simulation, improving the accuracy of ultrasonic shot peening forming process parameter design and the efficiency of engineering applications. Summary of the Invention
[0006] Based on the above, the purpose of this invention is to provide a rapid prediction method for ultrasonic shot peening forming instability based on GA-BP-ANN, aiming to solve the technical problems of existing methods being time-consuming and labor-intensive, and the resulting shot peening parameters being inaccurate.
[0007] To achieve the above objectives, this invention provides a rapid prediction method for ultrasonic shot peening forming instability based on GA-BP-ANN, comprising the following steps:
[0008] Step S1: Conduct ultrasonic shot peening forming experiments, obtain experimental data, and evaluate the instability morphology.
[0009] Step S2: Perform numerical calculations using GA-BP-ANN.
[0010] Step S3: Prediction of instability in ultrasonic shot peening forming.
[0011] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S1 includes the following steps:
[0012] S11: Determine the experimental conditions for ultrasonic shot peening (such as shot diameter, quantity, shot peening time, etc.) and the properties of the sheet metal (such as thickness, material, etc.), which will be used as input parameters for the model later.
[0013] S12: Conduct ultrasonic shot peening forming experiments using an ultrasonic shot peening device, analyze the experimental results, and evaluate the instability morphology.
[0014] S13: Obtain surface morphology data of the specimen after the ultrasonic shot peening forming instability experiment. Before the experiment, a regular mesh was created on the specimen surface. A three-dimensional coordinate system was established based on the specimen's geometry: the lower left corner vertex of the square plate was used as the origin, the left boundary direction of the plate was defined as the y-axis, the lower boundary direction as the x-axis, and the direction perpendicular to the plate surface as the z-axis. This constructed a three-dimensional coordinate system for the specimen surface, used to characterize the arc height values of each measurement point. After the experiment, a digital dial indicator was used to measure the arc height of the 21 intersection points in the mesh, and the measurement results were used as the z-axis coordinate values of the corresponding points. Combined with the position coordinates of each measurement point in the x–y plane, a three-dimensional surface morphology database of the formed specimen was established.
[0015] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S2 includes the following steps:
[0016] S21: Preprocess the data by dividing it into training, validation, and test sets, and normalize the data.
[0017] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S21 includes the following steps:
[0018] (1) The dataset obtained in step S1 is processed into grades according to parameters such as projectile diameter, shot peening time and sampling coordinates. Within each grade, it is divided into training set, validation set and test set in a uniform random manner to ensure the representativeness and comprehensiveness of the sample distribution and improve the stability of model training and prediction accuracy.
[0019] (2) The following formula is used to normalize the data, thereby normalizing the original data to the interval [0,1].
[0020]
[0021] Where y is the normalized data; x is the original data; x max With x min These are the maximum and minimum values of the original data, respectively.
[0022] S22: Determine the structural parameters of the neural network, build the BP-ANN model, and set the training parameters of the neural network, including setting the number of neurons in the input layer, hidden layer, and output layer, etc. The number of neurons in the hidden layer is determined by the following empirical formula to determine the approximate range, and then the optimal number of nodes is determined by trial and error.
[0023]
[0024] in, Here, m represents the number of hidden layer nodes, and n represents the number of output layer nodes. a is a constant. To prevent underfitting, therefore .
[0025] S23: Encode the weights and thresholds of the BP-ANN into chromosomes for the genetic algorithm, using real-number encoding. Assume the number of nodes in the input layer is... The number of hidden layer nodes is The number of output layer nodes is The length N of the BP-ANN weight threshold can be obtained using the following formula:
[0026]
[0027] S24: Use the fitness function to calculate the fitness value of each individual in turn and record it.
[0028] S25: Perform selection, crossover, and mutation operations to generate a new generation of population.
[0029] A rapid prediction method for ultrasonic shot peening instability based on GA-BP-ANN, wherein step S25 includes the following steps:
[0030] (1) Selection operation: The selection operation is carried out in the form of roulette, with the individual fitness value as the selection probability and the population size as the cycle base to select a new individual;
[0031] (2) The crossover operation specifically includes the following steps:
[0032] a: Select an individual from the population in a random manner;
[0033] b: Determine the crossover probability: Decide whether to perform a crossover operation;
[0034] c: Compared with the simple crossover operation in traditional genetic algorithms, this patent uses the arithmetic crossover method to generate offspring individuals by linearly combining the genes of parent individuals;
[0035] d: Check whether the boundary conditions are met after the intersection;
[0036] e: Repeat steps a, b, c, and d until a new individual is selected;
[0037] (3) The mutation operation specifically includes the following steps:
[0038] a: Select an individual from the population in a random manner;
[0039] b: Determine the mutation probability: decide whether to perform the mutation operation;
[0040] c: Perform mutation operation (gene value adjustment):
[0041] Unlike traditional genetic algorithms that use a fixed mutation amount, this patent uses the following formula to determine the mutation amount:
[0042]
[0043] in, This represents the amount of mutation in the mutation operation. A random number in the range (0, 1). The iteration progress is represented by the number of iterations (n approaches N). It will approach 0, resulting in a mutation rate. It also approaches 0. This means that the algorithm has a large variation range in the early stages, while the variation range gradually decreases in the later stages. This strategy helps the algorithm to perform a global search in the early stages and a fine-grained local search in the later stages.
[0044] d: Check whether the boundary conditions are met after mutation;
[0045] e: Repeat steps a, b, c, and d until a new individual is selected;
[0046] S26: Repeat the operation in S25 until the loop condition is met. Then, assign the initial weights and thresholds of the network corresponding to the best individual obtained by the genetic algorithm to the BP-ANN to obtain the GA-BP-ANN model.
[0047] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S3 includes the following steps:
[0048] A trained GA-BP-ANN model was used to predict and analyze the arc height values at five locations on the forming surface. Based on the predicted data, response curves of the arc height at each measuring point as a function of shot peening time under different shot diameters were plotted. By identifying the fluctuation patterns of the response curves, the specific working conditions corresponding to the instability of the forming surface can be determined.
[0049] The beneficial effects of this invention are as follows: During training, BP-ANN often gets trapped in local optima due to gradient descent. The global search characteristic of genetic algorithms can optimize the initial weights and thresholds of BP-ANN, guiding it out of local optima and closer to the global optimum, thus improving the overall performance of the model. This invention significantly improves computational efficiency and successfully predicts whether instability will occur during model formation, demonstrating high practicality and convenience. Attached Figure Description
[0050] Figure 1 This is a schematic diagram of the morphological evolution process of instability in the ultrasonic shot peening forming experiment of the present invention.
[0051] Figure 2 This is a schematic diagram of the grid division of the experimental plate in this invention.
[0052] Figure 3 This is a flowchart illustrating the implementation of a rapid prediction method for ultrasonic shot peening forming instability based on GA-BP-ANN provided by the present invention.
[0053] Figure 4 This is the fitness iteration graph of the genetic optimization algorithm of this invention.
[0054] Figure 5 This is a comparison chart of the predicted and experimental values of the GA-BP-ANN test set of this invention.
[0055] Figure 6 This is the training regression graph of the prediction results of the GA-BP-ANN test set in this invention.
[0056] Figure 7 This is the response curve of the GA-BP-ANN predicting the arc height of the forming surface in this invention. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0058] To illustrate the technical solution described in this invention, specific embodiments are described below.
[0059] An embodiment of a rapid prediction method for ultrasonic shot peening forming instability based on GA-BP-ANN includes the following steps:
[0060] Step S1: Conduct ultrasonic shot peening forming experiments, obtain experimental data, and evaluate the instability morphology.
[0061] Step S2: Perform numerical calculations using GA-BP-ANN.
[0062] Step S3: Prediction of instability in ultrasonic shot peening forming.
[0063] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S1 includes the following steps:
[0064] S11: Determine the experimental conditions for ultrasonic shot peening.
[0065] In the experiment, 304 stainless steel shot with diameters of 4mm, 6mm, and 8mm was used to impact-treat 2024 aluminum alloy sheets. Ensuring that the shot of different diameters was evenly distributed across the bottom of the shot peening chamber, the corresponding quantities of the three types of shot were 251, 107, and 62, respectively. During the experiment, the sample was placed horizontally on top of the shot peening chamber and secured with a cover plate and nuts. The shot consisted of particles with a frequency of 15kHz and an amplitude of 40... Driven by ultrasonic vibration, the sample surface was subjected to single-sided impact in a sealed chamber. The ultrasonic shot peening time was determined to be 10s, 20s, 40s, 80s, 160s, 320s, 640s and 1280s through preliminary experiments.
[0066] S12: Conduct ultrasonic shot peening forming experiments using an ultrasonic shot peening device, analyze the experimental results, and refer to... Figure 1 Three forms of instability evolution were assessed.
[0067] All samples exhibited significant overall bulging deformation towards the unpeened surface, indicating overall instability based on the deformation morphology. According to the geometric characteristics after instability, the overall instability observed in the experiment manifested in two typical forms: one is "spherical instability," where the center of the sample bulges towards the unpeened surface in an approximately spherical crown shape; the other is "cylindrical instability," where the sample bulges in an approximately cylindrical shape along a single direction. When the peening time was short, the samples did not show significant deformation under the constraint of the cover plate. However, after the cover plate was removed, the samples rapidly underwent overall bulging deformation, indicating that the deformation at this stage was mainly dominated by overall instability. With further extension of the peening time, irreversible deformation gradually accumulated in the peened area based on overall instability, producing significant local undulations even under cover plate constraints. After removing the cover plate, the samples could not return to their initial flat state; their surfaces gradually evolved into wavy, locally bulging morphologies, exhibiting deformation behavior characterized by local instability.
[0068] S13: The arc height of the formed surface of the sheet metal is collected by sampling points through mesh generation as the model output data.
[0069] Reference Figure 2 A three-dimensional coordinate system was constructed on the sample surface to characterize the arc height values at each measurement point. After the experiment, a digital dial indicator was used to measure the arc height of 21 intersection points in the grid, and the measurement results were used as the z-axis coordinate values of the corresponding points. Combined with the position coordinates of each measurement point in the x–y plane, a three-dimensional surface morphology database of the sample after forming was established.
[0070] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S2 includes the following steps:
[0071] S21: Preprocess the data by dividing it into training, validation, and test sets, and normalize the data.
[0072] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, referring to Figure 3 Step S21 includes the following steps:
[0073] (1) The 504 data obtained in step S1 are classified according to the projectile diameter, peening time and sampling point coordinates. Within each classification, the data are divided into training set, validation set and test set in a uniform random manner at a ratio of 60%, 20% and 20% respectively (the final data volume is 306, 102 and 98 respectively). This ensures the representativeness and comprehensiveness of the sample distribution and improves the stability of model training and prediction accuracy.
[0074] (2) The following formula is used to normalize the data, thereby normalizing the original data to the interval [0,1].
[0075]
[0076] Where y is the normalized data; x is the original data; x max With x min These are the maximum and minimum values of the original data, respectively.
[0077] S22: Determine the structural parameters of the neural network, build the BP-ANN model, and set the training parameters of the neural network.
[0078] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein the specific implementation process of step S22 is as follows:
[0079] (1) In this embodiment, BP-ANN adopts a three-layer structure with one input layer, one hidden layer, and one input layer. The number of neurons in the input layer is 4, the number of neurons in the output layer is 1, and the number of neurons in the hidden layer is determined to be 14 by trial and error.
[0080] (2) The transfer function of the hidden layer of the BP-ANN is the logsig function, the transfer function of the output layer is the purelin linear function, the training algorithm is the trainlm algorithm, the learning rate is set to 0.01, and the network target error is 0.1×10⁻⁶. -7 ;
[0081] (3) The network performance evaluation function uses the mean square error (MSE):
[0082]
[0083] (4) A 20% validation set was specified. During training, the MSE of the training set and the MSE of the validation set are monitored simultaneously. If the validation set error does not decrease for 6 consecutive times, early stopping is triggered to prevent overfitting.
[0084] S23: Encode the weights and thresholds of the BP-ANN into chromosomes of the genetic algorithm, using real number encoding. Assuming the number of input layer nodes is Nin, the number of hidden layer nodes is Nh, and the number of output layer nodes is Nout, the length of the BP-ANN weight threshold can be calculated using the following formula:
[0085]
[0086] The length N of the BP-ANN weight threshold in this example is found to be 85.
[0087] S24: Use the fitness function to calculate the fitness value of each individual in turn and record it.
[0088] A rapid prediction method for ultrasonic shot peening forming instability based on GA-BP-ANN, wherein in step S24, the fitness value is calculated using the following fitness function:
[0089]
[0090] a: Calculate and record the fitness value for each individual in sequence, where... Let be the predicted value for the i-th sample. Let be the true value of the i-th sample.
[0091] b: Find the maximum fitness value of the current generation individuals and assign it to... Calculate the average fitness value of the current generation individuals and assign it to . Reference Figure 4 Therefore, the number of iterations for the fitness iteration graph can be determined to be 100.
[0092] S25: Perform selection, crossover, and mutation operations to generate a new generation of population.
[0093] A rapid prediction method for ultrasonic shot peening instability based on GA-BP-ANN, wherein step S25 includes the following steps:
[0094] (1) Selection operation: The selection operation is carried out in a roulette wheel manner, with the individual fitness value as the selection probability and the population size as the cycle number to select a new individual; the probability of each individual being selected is calculated as follows:
[0095]
[0096] in, Let be the probability that the i-th individual is selected, and N be the population size. This step ensures that the sum of all probabilities is 1.
[0097] (2) The crossover operation specifically includes the following steps:
[0098] a: Select an individual from the population in a random manner;
[0099] b: Determine the crossover probability: Decide whether to perform a crossover operation.
[0100] c: Compared with the simple crossover operation in traditional genetic algorithms, this patent uses the arithmetic crossover method to generate offspring individuals by linearly combining the genes of parent individuals.
[0101]
[0102] in, It is the gene of the offspring individuals that are generated. and It consists of genes from two parent chromosomes. The cross weights are randomly generated.
[0103] d: Check whether the boundary conditions are met after the intersection;
[0104] e: Repeat steps a, b, c, and d until a new individual is selected;
[0105] (3) The mutation operation specifically includes the following steps:
[0106] a: Select an individual from the population in a random manner;
[0107] b: Determine the mutation probability: decide whether to perform the mutation operation;
[0108] c: Perform mutation operation (gene value adjustment):
[0109] Unlike traditional genetic algorithms that use a fixed mutation amount, this patent uses the following formula to determine the mutation amount:
[0110]
[0111] in, This represents the amount of mutation in the mutation operation. A random number in the range (0, 1). The iteration progress is represented by the number of iterations (n approaches N). It will approach 0, resulting in a mutation rate. It also approaches 0. This means that the algorithm has a large variation range in the early stages, while the variation range gradually decreases in the later stages. This strategy helps the algorithm to perform a global search in the early stages and a fine-grained local search in the later stages.
[0112] d: Check whether the boundary conditions are met after mutation;
[0113] e: Repeat steps a, b, c, and d until a new individual is selected;
[0114] S26: Repeat step S25 until the loop condition is met. Then, assign the initial weights and thresholds of the network corresponding to the optimal individual obtained through the genetic algorithm to the BP-ANN to obtain the GA-BP-ANN model. The training results are attached. Figure 5 and attached Figure 6 As shown in the figure, the experimental and measured values of the arc height at various points in the test set are compared. This patent uses R... 2 Using RMSE as the evaluation index of the model, the obtained values are 0.9308 and 0.2242, respectively. It can be seen that GA-BP-ANN can predict the instability results of ultrasonic shot peening forming very well.
[0115] A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S3 includes the following steps:
[0116] A trained GA-BP-ANN model was used to predict and analyze the arc height values at five locations on the forming surface. Based on the predicted data, response curves of the arc height at each measuring point as a function of shot peening time under different shot diameters were plotted. By identifying the fluctuation patterns of the response curves, the specific working conditions corresponding to the instability of the forming surface can be determined.
[0117] Reference Figure 7 (a) When the shot diameter is 4 mm, during the short peening period, the arc height curves of measuring points 1 and 5 are mainly above zero and show an upward trend, indicating that a convex deformation towards the peening surface has occurred in this area. When the peening time approaches 80 s, the curve changes from an upward trend to a downward trend, corresponding to the appearance of local instability characteristics on the peened surface. As the peening time is further extended, when the peening time exceeds approximately 570 s, the arc height value of measuring point 3 changes from negative to positive and continues to increase with time, indicating that the middle of the sample gradually transforms into a convex deformation towards the peening surface, and this instability characteristic is continuously enhanced with the further increase of the peening time.
[0118] Under 6mm projectile impact conditions, the evolution trend of the material surface arc height is generally similar to that under 4mm projectile impact conditions. (Refer to...) Figure 7(b) When the shot peening time at measuring points 1 and 5 was approximately 160 s, the arc height curves changed from rising to falling, indicating that the material began to exhibit local instability characteristics. Due to the difference in shot energy, after the appearance of local buckling instability characteristics, the deformation of the plate under the 4 mm shot condition quickly shifted to being dominated by local instability; while under the action of 6 mm shot, due to the relatively small shot energy, the subsequent deformation of the plate was still affected by both the overall instability mode and the local instability mode, making the fluctuation of the arc height response curve more obvious.
[0119] Due to the relatively low impact energy of the 8mm projectile, within the experimental timeframe, the material deformation was predominantly characterized by overall instability. (Refer to...) Figure 7 (c) The arc height values of measuring points 1 and 5 increased rapidly with the shot peening time, and the rate of increase slowed down significantly after reaching about 3 mm. Subsequently, when the shot peening time exceeded about 580 s, the arc height values of measuring points 2, 3 and 4 showed a decreasing trend with time, indicating that the overall instability mode was still evolving and gradually intensifying. When the shot peening time was further extended to about 990 s, the above trend changed, and local instability characteristics began to appear on the material surface, specifically manifested as a gradual protrusion deformation in the middle of the sample towards the peened surface.
[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, characterized in that... Includes the following steps: Step S1: Conduct ultrasonic shot peening forming experiments, obtain experimental data, analyze experimental results, and evaluate instability morphology. Step S2: Perform numerical calculations using GA-BP-ANN. Step S3: Prediction of instability in ultrasonic shot peening forming.
2. A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S1 includes the following steps: S11: Determine the experimental conditions for ultrasonic shot peening forming, which will be used as input parameters for the model later. S12: Conduct ultrasonic shot peening forming experiments using an ultrasonic shot peening device, analyze the experimental results, and evaluate the instability morphology. S13: Use mesh generation to collect the arc height of the plate after it is formed as the model output data.
3. A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S2 includes the following steps: S21: Preprocess the data, dividing it into training, validation, and test sets, and then normalize the data. Details are as follows: (1) The data obtained in step S1 are classified according to the projectile diameter, peening time and sampling point coordinates. Within each classification, the data is divided into training set, validation set and test set in a uniform random manner at a ratio of 60%, 20% and 20%, respectively, so as to ensure the representativeness and comprehensiveness of the sample distribution and improve the stability of model training and prediction accuracy. (2) The following formula is used to normalize the data, thereby normalizing the original data to the interval [0,1]. S22: Determine the structural parameters of the neural network and build the BP-ANN model. S23: Encode the weights and thresholds of the BP-ANN into chromosomes for the genetic algorithm. S24: Use the fitness function to calculate the fitness value of each individual in turn and record it. S25: Perform selection, crossover, and mutation operations to generate a new generation of population. Details are as follows: (1) Selection operation: The selection operation is carried out in the form of roulette, with the individual fitness value as the selection probability and the population size as the cycle base to select a new individual; (2) The crossover operation specifically includes the following steps: a: Select an individual from the population in a random manner; b: Determine the crossover probability: Decide whether to perform a crossover operation; c: Compared with the simple crossover operation in traditional genetic algorithms, this patent uses the arithmetic crossover method to generate offspring individuals by linearly combining the genes of parent individuals. d: Check whether the boundary conditions are met after the intersection; e: Repeat steps a, b, c, and d until a new individual is selected; (3) The mutation operation specifically includes the following steps: a: Select an individual from the population in a random manner; b: Determine the mutation probability: decide whether to perform the mutation operation; c: Perform mutation operation (gene value adjustment): Unlike traditional genetic algorithms that use a fixed mutation amount, this patent uses the following formula to determine the mutation amount: in, This represents the amount of mutation in the mutation operation. A random number in the range (0, 1). The iteration progress is represented by the number of iterations (n approaches N). It will approach 0, resulting in a mutation rate. It also approaches 0. This means that the algorithm has a large variation range in the early stages, while the variation range gradually decreases in the later stages. This strategy helps the algorithm to perform a global search in the early stages and a fine-grained local search in the later stages. d: Check whether the boundary conditions are met after mutation; e: Repeat steps a, b, c, and d until a new individual is selected; S26: Repeat the operation in S25 until the loop condition is met. Then, assign the initial weights and thresholds of the network corresponding to the best individual obtained by the genetic algorithm to the BP-ANN to obtain the GA-BP-ANN model.
4. A rapid prediction method for instability in ultrasonic shot peening based on GA-BP-ANN, wherein step S3 includes the following steps: A trained GA-BP-ANN model was used to predict and analyze the arc height values at five locations on the forming surface. Based on the predicted data, response curves of the arc height at each measuring point as a function of shot peening time under different shot diameters were plotted. By identifying the fluctuation patterns of the response curves, the specific working conditions corresponding to the instability of the forming surface can be determined.