A shot peening process parameter design method including missing data processing
By combining machine learning and artificial intelligence to process missing data in shot peening process parameter design, a multi-layer process parameter design model was established, which solved the problems of low accuracy and long processing time caused by missing data in shot peening strengthening, and achieved stable and accurate process parameter guidance.
Patent Information
- Application Number
- CN202211324639.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-27
- Publication Date
- 2025-12-19
- Estimated Expiration
- 2042-10-27
AI Technical Summary
Existing shot peening strengthening process parameter design methods suffer from low accuracy and time consumption due to missing data, especially in the determination of shot peening process parameters, where there is a lack of accurate and stable design methods.
A multi-layer process parameter design model based on machine learning and artificial intelligence is adopted. Missing data is processed by methods such as Lagrange interpolation and CatBoost regression algorithm. The correlation between shot peening process parameters and surface integrity parameters is established. The importance of process parameters is evaluated by random forest algorithm, and multi-layer process parameter design is carried out.
It enables stable and accurate guidance for determining shot peening process parameters even in the presence of missing data, improving model accuracy and design efficiency while reducing errors.
Smart Images
Figure CN115730398B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of mechanical manufacturing, and particularly relates to a shot peening process parameter design method containing missing data processing. BACKGROUND
[0002] Shot peening can improve the fatigue life of mechanical parts at low cost without being limited by the size and shape of the parts, and is widely used. In the process of shot peening, a large number of pellets continuously impact the surface of the workpiece, so that plastic deformation occurs on the surface of the part to introduce residual stress and refine the grains, thereby improving the fatigue resistance of the transmission parts. The strengthening effect is affected by various shot peening process parameters, such as pellet diameter, shot peening intensity, shot peening coverage, relative motion speed, etc. In engineering practice, the determination of the shot peening process parameters is often given by the process personnel according to the experience relationship, and there is a certain deviation between the obtained surface integrity effect and the expectation.
[0003] There are two limitations in establishing an accurate and stable correlation between the shot peening process parameters and the surface integrity after shot peening. On the one hand, due to equipment, test scheme and human factors, etc., the phenomenon of missing some key data exists widely in process tests. On the other hand, there is a nonlinear and complex correlation between the surface integrity parameters after shot peening and the process parameters, and there is a lack of accurate and stable shot peening process parameter design method.
[0004] Among them, the existing shot peening process parameter design methods can be divided into two kinds: one is a method combining test and simulation, and the other is a method based on machine learning. However, in these two methods, there are problems such as low accuracy and long time consumption caused by data missing, which makes it still a challenging work to quickly and accurately determine the shot peening process parameters. SUMMARY
[0005] The purpose of the present application is to provide a shot peening process parameter design method containing missing data processing, which comprises the following steps:
[0006] 1) Obtain a set of surface integrity parameters after shot peening containing missing data, and determine the shot peening process parameters to be designed;
[0007] According to the missing proportion of each column parameter, arrange each column in the set of surface integrity parameters in ascending order;
[0008] 2) Select the mth surface integrity parameter as the filling target column, and pre-fill the missing values of the other surface integrity parameters; the initial value of m is 1;
[0009] 3) use all the data rows of the mth column without missing values to build a machine learning-based missing data filling model, and input the other m-1 columns of the data rows of the mth column with missing values into the machine learning-based missing data filling model to obtain the estimated values of the missing data in the surface integrity parameters of the mth column, and fill the missing data estimates into the surface integrity parameters of the mth column;
[0010] 4) let m = m + 1; delete the pre-filled data of the mth surface integrity parameter, and return to step 2) until a set of surface integrity parameters without missing values is obtained;
[0011] 5) build an N-layer process parameter design model based on artificial intelligence according to the filled shot surface integrity parameters and the corresponding shot process parameters;
[0012] 6) input the target value of the shot surface integrity parameter into the jth layer process parameter design model to obtain the estimated jth shot process parameter, and write the jth process parameter into the input parameter set; j initial value is 1; j = 1, 2, …, N;
[0013] 7) let j = j + 1, and return to step 6) until N process parameters are obtained.
[0014] Further, the pre-filling method includes a statistical method.
[0015] Further, the statistical method includes Lagrange interpolation method.
[0016] The pre-filled missing value L(x) is as follows:
[0017]
[0018] In the formula, n is the number of known surface integrity parameters of the current column; x i , x j , y i are known surface integrity parameters.
[0019] Further, the artificial intelligence algorithm includes but is not limited to CatBoost and random forest.
[0020] Further, the surface integrity parameters include but are not limited to residual stress, hardness and surface roughness.
[0021] Further, the machine learning-based missing data filling model is a CatBoost regression algorithm model, and its loss function loss is as follows:
[0022]
[0023] In the formula, Y i’,expY represents the measured data of sample i i’,pre Y represents the estimated data, and n' represents the number of samples.
[0024] Further, the process parameters to be designed include, but are not limited to, shot strength, shot coverage, and projectile diameter.
[0025] Further, when building the N-layer process parameter design model based on artificial intelligence, the importance of the to-be-designed shot peening process parameters is evaluated to obtain an importance score of the to-be-designed shot peening process parameters, and the process parameter design model is built according to the importance.
[0026] Further, the method for evaluating the importance of the to-be-designed shot peening process parameters includes a random forest algorithm.
[0027] wherein the importance score of the i-th to-be-designed shot peening process parameter as follows:
[0028]
[0029] wherein Ntree is the number of decision trees in the random forest algorithm; ER it is the error rate corresponding to the t-th tree before the variable i is replaced; ER i t is the error rate corresponding to the t-th tree after the variable i is replaced.
[0030] The technical effect of the present application is self-evident. In view of the problems of low accuracy and long time consumption of the existing shot peening process parameter design method, the present application provides a shot peening process parameter design method based on missing data processing, which can realize stable and accurate realization of shot peening process parameter positive guidance while realizing missing surface integrity parameter filling.
[0031] The present application accurately and stably fills the surface integrity parameters with certain missing data, and has smaller error compared with traditional statistical methods and conventional machine learning models; a multi-layer shot peening process parameter design model is established, which can guide the selection of shot peening strengthening parameters according to the target surface integrity parameters after shot peening strengthening, and effectively improve the model precision by combining with the missing data processing method. BRIEF DESCRIPTION OF DRAWINGS
[0032] Fig. 1 is a flowchart of the method of the present application
[0033] Fig. 2 is a flowchart of the stepwise regression missing data processing method
[0034] Fig. 3 is a flowchart of the multi-layer shot peening process parameter prediction DETAILED DESCRIPTION
[0035] The application will be further described in conjunction with the following examples, but should not be understood as limiting the above-mentioned subject matter of the application to the following examples. Various substitutions and changes can be made according to ordinary technical knowledge and conventional means without departing from the technical idea of the application, and all should be included in the protection scope of the application.
[0036] Example 1
[0037] Referring to Figs. 1 to 3 A shot process parameter design method comprising missing data processing, comprising the following steps:
[0038] 1) Obtain a set of shot surface integrity parameters containing missing data, and determine the shot process parameters to be designed;
[0039] According to the missing proportion of each column parameter, arrange each column in the set of surface integrity parameters in ascending order;
[0040] 2) Select the mth surface integrity parameter as the filling target column, and pre-fill the missing values of other surface integrity parameters; the initial value of m is 1;
[0041] 3) Use all non-missing data rows in the mth column to establish a missing data filling model based on machine learning, and input the other m-1 columns of the data rows with missing values in the mth column into the missing data filling model based on machine learning, to obtain the estimated values of the missing data in the mth surface integrity parameter, and fill these missing data estimates into the mth surface integrity parameter;
[0042] 4) Let m = m + 1; delete the pre-filled data of the mth surface integrity parameter, and return to step 2) until a set of surface integrity parameters without missing data is obtained;
[0043] 5) According to the filled shot surface integrity parameters and the corresponding shot process parameters, build an N-layer process parameter design model based on artificial intelligence;
[0044] 6) Input the target value of the shot surface integrity parameter into the jth layer process parameter design model to obtain the estimated jth shot process parameter, and write the jth process parameter into the input parameter set; the initial value of j is 1; j = 1, 2, …, N;
[0045] 7) Let j = j + 1, and return to step 6) until N process parameters are obtained.
[0046] The method for pre-filling includes a statistical method.
[0047] The statistical method includes Lagrange interpolation method;
[0048] The pre-filled missing values L(x) are as follows:
[0049]
[0050] wherein n is the number of known surface integrity parameters of the current column; x i , x j , y i are known surface integrity parameters.
[0051] The artificial intelligence algorithm includes but is not limited to CatBoost, random forest.
[0052] The surface integrity parameters include but are not limited to residual stress, hardness and surface roughness.
[0053] The machine learning-based missing data filling model is a CatBoost regression algorithm model, and the loss function loss thereof is as follows:
[0054]
[0055] wherein Y i’,exp represents the measured data of sample i, Y i’,pre represents the estimated data, and n' represents the number of samples.
[0056] The process parameters to be designed include but are not limited to shot peening intensity, shot peening coverage and shot diameter.
[0057] When building an N-layer process parameter design model based on artificial intelligence, the importance of the shot peening process parameters to be designed is evaluated to obtain the importance score of the shot peening process parameters to be designed, and the process parameter design model is built from large to small according to the importance ranking. The importance determines the order of the process parameter design model, and the importance is strong and the importance is weak.
[0058] The method for evaluating the importance of the shot peening process parameters to be designed includes a random forest algorithm;
[0059] wherein the importance score of the i th shot peening process parameter to be designed is as follows:
[0060]
[0061] wherein Ntree is the number of decision trees in the random forest algorithm; ER it is the error rate corresponding to the t th tree before the variable i is replaced; ER i ′ t is the error rate corresponding to the t th tree after the variable i is replaced.
[0062] Example 2:
[0063] A shot process parameter design method comprising missing data processing, comprising the following steps:
[0064] Step 1, a step-by-step regression missing data processing method based on machine learning is constructed, the obtained data set containing missing values is taken as input, and the missing proportion of each column parameter is arranged from small to large;
[0065] Step 2, select the parameter column with the smallest missing proportion as the first round filling target column, and pre-fill each column;
[0066] Step 3, establish a machine learning algorithm model, divide the data set, input the training set into the algorithm model, and obtain the estimated value of the missing data of the target column;
[0067] Step 4, then select the parameter column with the second smallest missing proportion as the second round filling target column, and the pre-filled data in the last round is removed, that is, the temporary data set, repeat the last round of steps;
[0068] Step 5, after multiple rounds of filling, the complete surface integrity parameter is obtained. The number of filling rounds depends on the number of missing columns of the input data;
[0069] Step 6, build a multi-layer process parameter design model, input the surface integrity parameter after missing data processing as initial data into the first layer algorithm model, obtain the estimated first process parameter, then input the obtained result into the input parameter, output the second process parameter through the second layer algorithm model, then input the result into the input data. In this way, the nth process parameter is obtained through the nth layer algorithm model, and finally the design guidance of the shot process parameter is realized.
[0070] In step 1, the shot process parameter design target is determined, and the shot process parameter and the surface integrity parameter sample data set containing missing data are collected.
[0071] In step 2, the statistical method is used as pre-filling to estimate the missing data.
[0072] In step 3, a machine learning algorithm model is established, and the best adjustable parameter is determined.
[0073] In step 6, the importance of the shot process parameter is evaluated first.
[0074] According to the above importance analysis and machine learning algorithm, a multi-layer process parameter design model is built.
[0075] The artificial intelligence algorithm proposed in the application includes but is not limited to CatBoost, random forest and the like. In the process parameter design aspect, except for the shot peening strengthening, the same is applicable to the correlation between the process parameter data of finishing, grinding and the like and the workpiece surface integrity.
[0076] Embodiment 3:
[0077] A shot peening process parameter design method comprising missing data processing includes the following contents:
[0078] As Fig. 1 shown is a flowchart of the method of the application. First, the residual stress and hardness gradient under each shot peening process containing missing data are obtained through experiments. Subsequently, a stepwise regression missing data filling method based on machine learning is designed, and the missing data is estimated by using the missing feature autocorrelation and the correlation of some depth other parameters to fill the missing data completely. Finally, a shot peening process prediction model is established based on an ensemble learning algorithm, and the residual stress and hardness gradient and other parameters corresponding to different shot peening processes after filling the complete and reliable missing data are used for training of the model, and then the shot peening intensity, pellet diameter and shot peening coverage are sequentially designed in layers.
[0079] The specific steps include:
[0080] Step 1, the surface integrity parameter data of steel after being treated by different shot peening processes and the corresponding process parameters are obtained through experiments. These data include but are not limited to: shot peening intensity, shot peening coverage, pellet diameter, relative speed, residual stress, hardness and surface roughness and the like;
[0081] Step 2, the input parameters containing missing data and the shot peening process parameter design target are determined. The depth, residual stress and hardness are selected as the input parameters, and the shot peening intensity, shot peening coverage and pellet diameter are selected as the shot peening process parameter design target;
[0082] Step 3, the missing data processing method is used for the input surface integrity data containing missing data;
[0083] Step 4, a stepwise regression missing data processing method based on machine learning algorithm is established: as Fig. 2 shown is a flowchart of the method, (a) part expands the residual stress, hardness or other parameters containing missing data obtained through experiments along the depth gradient, inputs (b) part missing data processing method model, and finally (c) part obtains the complete residual stress and hardness distribution curve. In (b) part missing data processing method, firstly, the missing proportion of each column of input parameters is sorted, the parameter column with the smallest missing proportion is selected as the first round filling target column in I part, and the Lagrange interpolation method is used for pre-filling of other columns: given n depths xi Corresponding residual stress and hardness y i , i.e. points (x1, y1), (x2, y2), (x3, y3), …, (x n , y n ), a polynomial of n-1 degree can be obtained through these n points, assuming it is:
[0084] y = a0 + a1x + a2x 2 + … + a n-1 x n-1 (1)
[0085] Substitute n points into the polynomial:
[0086]
[0087] The Lagrange interpolation polynomial is obtained:
[0088]
[0089] Using known data to obtain the Lagrange interpolation polynomial, and then substituting the missing node into the polynomial L(x), the missing data (residual stress and hardness) can be estimated, and a temporary data set with only a single column missing is obtained. Part II takes each row of data in the target column with missing values in the temporary data set as the test set, and the other complete data as the training set. A CatBoost regression algorithm model is established, and the best adjustable parameters are determined, including loss function, maximum tree number, learning rate, regularization parameter, sample rate, etc. Among them, the loss function formula is as follows:
[0090]
[0091] Where Y i,exp represents the measured data of sample i, Y i,pre represents the estimated data, and n represents the number of samples. By minimizing the loss, the weight of each base learner is determined, so as to reduce the negative impact of base learners with poor prediction accuracy, and finally make the final model estimation result more accurate.
[0092] Part III inputs the training set into the CatBoost algorithm model to obtain the estimated value of the target column missing data according to the correlation between parameters. Then, return to Part I to select the parameter column with the second smallest missing proportion as the target column for the second round of filling. The data pre-filled in the last round is removed, which is the temporary data set, and repeat the last round of steps. This method estimates missing data by machine learning regression model in the order of increasing missing proportion ladder, and obtains complete surface integrity parameters after multiple rounds of filling, realizes ladder regression missing data processing, and can be used for more missing data processing of surface integrity parameters and other data sets;
[0093] Step 5, evaluate the importance of shot peening process parameter design target, calculate the importance score (Variable Importance Measures, VIM) of the shot peening process parameter design target variable according to the Mean Decrease Accuracy (MDA) using the random forest algorithm, wherein the VIM of variable i is as follows:
[0094]
[0095] where Ntree is the number of decision trees in the random forest algorithm, which is 1000 here; ER it is the error rate corresponding to the tth tree before the variable i is replaced; ER' it is the error rate corresponding to the tth tree after the variable i is replaced.
[0096] The influence degree of shot peening intensity, shot peening coverage and shot diameter on residual stress and hardness is calculated and shown in Table 1. The average VIM of shot peening intensity, shot peening coverage and shot diameter on residual stress and hardness is 0.425, 0.305 and 0.27 respectively, so the influence degree of process parameters on the model from large to small is shot peening intensity, shot peening coverage and shot diameter;
[0097] Table 1. Contribution of shot peening process parameters
[0098]
[0099] Step 6, build a multi-layer shot peening process parameter design model: according to the above contribution degree analysis result and CatBoost algorithm (the loss function and parameter determination process are consistent with step 4), build a multi-layer process parameter design model of shot peening intensity- shot peening coverage- shot diameter, as Fig. 3 is the flow chart of the model. First, the measured point depth, residual stress and hardness after shot peening are input into the first layer CatBoost model as initial data to obtain the estimated shot peening intensity, then the obtained shot peening intensity is input into the second layer CatBoost model to output the shot peening coverage, then the shot peening coverage is also input into the input data, and finally the shot peening coverage is input into the third layer CatBoost model to obtain the shot diameter, so as to realize the design guidance of the target shot peening process parameters;
[0100] Step 7, select the mean absolute error MAE as the evaluation index to reflect the final effect of the method, and the formula is as follows:
[0101]
[0102] where Y i,exp represents the true data of sample i, Y i,preThe predicted data of sample i, n is the total number of samples. MAE represents the average error distance between the predicted value and the experimental value. The smaller the value, the better the model method effect;
[0103] Step 8, the missing data processing method and the process parameter design model are combined, the missing data is estimated by using the correlation of the missing features and the correlation of the certain depth residual stress and hardness, and the missing data is filled completely. Subsequently, the residual stress and hardness corresponding to different shot peening processes after filling the complete and reliable data are used for training of the model, and finally the shot peening strength, shot peening coverage and projectile diameter are sequentially layered and estimated;
[0104] Taking the surface integrity parameter data set with a missing ratio of 20% as an example, the prediction error of the model with or without the missing data processing method is analyzed compared with the real test process parameters, as shown in Table 2. Whether the missing data processing method is used or not, the layered shot peening process parameter design model proposed in the application has good prediction accuracy. The use of the missing data processing method improves the design effect of each process parameter, and the effect of the projectile diameter is the most obvious. In engineering practice, the trained design model can be used to guide the selection of shot peening parameters according to the target surface integrity parameters after shot peening.
[0105] Table 2. Effect of the method of the application under the condition of 20% missing surface integrity parameter data
[0106]
[0107] Example 4:
[0108] A shot peening process parameter design method comprising missing data processing, comprising the following steps:
[0109] 1) Obtain a shot peening surface integrity parameter set comprising missing data, and determine the shot peening process parameters to be designed;
[0110] According to the missing ratio of each column parameter, arrange each column in the surface integrity parameter set in order from small to large;
[0111] 2) Select the mth surface integrity parameter as the filling target column, and pre-fill the missing values of the other surface integrity parameters; the initial value of m is 1;
[0112] 3) Use all non-missing data rows in the mth column to establish a missing data filling model based on machine learning, and input the other m-1 columns of the data rows with missing values in the mth column into the missing data filling model based on machine learning, to obtain the estimated values of the missing data in the mth surface integrity parameter, and fill the estimated values of the missing data into the mth surface integrity parameter;
[0113] 4) Let m = m + 1; delete the pre-filled data of the m-th surface integrity parameter, and return to step 2) until a set of surface integrity parameters without missing values is obtained;
[0114] 5) According to the filled shot surface integrity parameters and the corresponding shot process parameters, an N-layer process parameter design model based on artificial intelligence is built;
[0115] 6) The target value of the shot surface integrity parameter is input into the j-th layer process parameter design model to obtain the estimated j-th shot process parameter, and the j-th process parameter is written into the input parameter set; j is initially 1; j = 1, 2, …, N;
[0116] 7) Let j = j + 1, and return to step 6) until N process parameters are obtained.
[0117] Example 5:
[0118] A shot process parameter design method containing missing data processing, the main content of which is seen in Example 4, wherein the pre-filling method includes a statistical method.
[0119] Example 6:
[0120] A shot process parameter design method containing missing data processing, the main content of which is seen in Example 4, wherein the statistical method includes a Lagrange interpolation method.
[0121] The pre-filled missing value L(x) is as follows:
[0122]
[0123] In the formula, n is the number of known surface integrity parameters in the current column; x i , x j , y i are known surface integrity parameters; x is an unknown quantity.
[0124] Example 7:
[0125] A shot process parameter design method containing missing data processing, the main content of which is seen in Example 4, wherein the artificial intelligence algorithm includes but is not limited to CatBoost, random forest.
[0126] Example 8:
[0127] A shot process parameter design method containing missing data processing, the main content of which is seen in Example 4, wherein the surface integrity parameters include but are not limited to residual stress, hardness, and surface roughness.
[0128] Example 9:
[0129] A shot process parameter design method including missing data processing, the main content is seen in embodiment 4, wherein the missing data filling model based on machine learning is CatBoost regression algorithm model, and its loss function loss is as follows:
[0130]
[0131] In the formula, Y i’,exp Indicates the measured data of sample i, Y i’,pre Indicates the estimated data, and n' indicates the sample quantity.
[0132] Embodiment 10:
[0133] A shot process parameter design method including missing data processing, the main content is seen in embodiment 4, wherein the process parameters to be designed include but are not limited to shot intensity, shot coverage and projectile diameter.
[0134] Embodiment 11:
[0135] A shot process parameter design method including missing data processing, the main content is seen in embodiment 4, wherein when building an N-layer process parameter design model based on artificial intelligence, the importance of the shot process parameters to be designed is evaluated to obtain the importance score of the shot process parameters to be designed, and the process parameter design model is built according to the importance.
[0136] Embodiment 12:
[0137] A shot process parameter design method including missing data processing, the main content is seen in embodiment 4, wherein the method for evaluating the importance of the shot process parameters to be designed includes a random forest algorithm;
[0138] Wherein the importance score of the i-th shot process parameter to be designed As follows:
[0139]
[0140] In the formula, Ntree is the number of decision trees in the random forest algorithm; ER it Is the error rate corresponding to the t-th tree before the variable i is replaced; ER i ′ t Is the error rate corresponding to the t-th tree after the variable i is replaced.
Claims
1. A shot peening process parameter design method including missing data handling, characterized by, The method comprises the following steps: 1) obtaining a set of shot surface integrity parameters containing missing data, and determining the shot process parameters to be designed; According to the missing proportion of each column parameter, arrange each column in the set of surface integrity parameters in ascending order; 2) select the mth column of surface integrity parameters as the filling target column, and pre-fill the missing values of other columns of surface integrity parameters; the initial value of m is 1; 3) use all non-missing data rows in the mth column to establish a missing data filling model based on machine learning, and input the other m-1 columns of the data rows with missing values in the mth column into the missing data filling model based on machine learning, to obtain the estimated values of the missing data in the mth column of surface integrity parameters, and fill these estimated values of missing data into the mth column of surface integrity parameters; 4) let m = m + 1; Delete the pre-filled data of the mth column of surface integrity parameters, and return to step 2) until a set of surface integrity parameters without missing data is obtained; 5) according to the filled shot surface integrity parameters and the corresponding shot process parameters, build an N-layer process parameter design model based on artificial intelligence; 6) input the target value of the shot surface integrity parameter into the jth layer process parameter design model, obtain the estimated jth shot process parameter, and write the jth process parameter into the input parameter set; the initial value of j is 1; j = 1, 2, …, N; 7) let j = j + 1, and return to step 6) until N process parameters are obtained; When building the N-layer process parameter design model based on artificial intelligence, the importance of the shot process parameters to be designed is evaluated to obtain the importance score of the shot process parameters to be designed, and the process parameter design model is built according to the importance; The method for evaluating the importance of the shot process parameters to be designed includes random forest algorithm; wherein the importance score of the i-th to-be-designed shot process parameter as follows: where Ntree is the number of decision trees in the random forest algorithm; ER it is the error rate of the tth tree before the variable i is replaced; ER' it is the error rate of the tth tree after the variable i is replaced.
2. A shot peening process parameter design method including missing data processing according to claim 1, characterized in that, The method for pre-filling includes statistical method.
3. A shot peening process parameter design method including missing data processing according to claim 2, characterized in that, The statistical method includes Lagrange interpolation method; The pre-filled missing value L(x) is as follows: where n is the number of known surface integrity parameters for the current column; x i , x j , y i are known surface integrity parameters; x is the unknown quantity.
4. A shot peening process parameter design method including missing data processing according to claim 1, wherein, The artificial intelligence algorithm includes but is not limited to CatBoost and random forest.
5. A shot peening process parameter design method including missing data processing according to claim 1, wherein, The surface integrity parameters include but are not limited to residual stress, hardness and surface roughness.
6. A shot peening process parameter design method including missing data processing according to claim 1, wherein, The missing data filling model based on machine learning is a CatBoost regression algorithm model, and its loss function loss is as follows: where Y i’,exp represents the measured data of sample i, Y i’,pre represents the estimated data, and n’ represents the number of samples.
7. A shot peening process parameter design method including missing data processing according to claim 1, wherein, The process parameters to be designed include but are not limited to shot intensity, shot coverage and projectile diameter.