Method for predicting repair depth of paint surface of vehicle body

Abstract

The invention relates to a method for predicting the repair depth of a vehicle body paint surface, and belongs to the technical field of surface processing, and the method comprises the steps: obtaining the polishing process parameters of the vehicle body paint surface; determining a first polishing depth according to the polishing process parameters and the target physical model; the target physical model is a physical model for describing the relationship between the polishing process parameters and the polishing depth; determining a second polishing depth according to the polishing process parameters and the target neural network model; the target neural network model is a polishing depth prediction neural network model; determining a target polishing depth according to the first polishing depth and the second polishing depth; and the vehicle body paint surface is polished according to the target polishing depth. The target polishing depth is obtained by integrating the polishing depth prediction result of the neural network and the polishing depth prediction result of the physical model, the accuracy of the target polishing depth can be improved, the paint surface defect can be accurately removed by polishing the vehicle body paint surface according to the target polishing depth, and a bottom layer is not worn through.

Description

Technical Field

[0001] The invention relates to the technical field of surface processing, and in particular to a method for predicting the depth of vehicle body paint repair. Background Art

[0002] During the manufacturing or maintenance of high-speed rail, if the defects of the paint film need to be repaired by grinding, the grinding depth has a great impact on the repair process and the quality of the repair. For example, since the thickness of the paint film is generally thin, once the underlying putty layer is worn through during the grinding process, the repair process will become more complicated. Therefore, higher requirements are placed on the prediction of the grinding depth and the grinding process parameters.

[0003] In the related art, the grinding depth can be predicted by a neural network model. However, the neural network model requires a large number of experimental samples for training, and there is currently almost no experimental data on paint film grinding processing available, which results in the neural network model's prediction of the grinding depth not being accurate enough. Summary of the invention

[0004] In view of this, it is necessary to provide a method for predicting the depth of vehicle body paint repair to solve the problem that the neural network model is not accurate enough in predicting the polishing depth.

[0005] In order to solve the above problems, the present invention provides a method for predicting the depth of vehicle body paint repair, comprising: Obtain the polishing process parameters of the car body paint surface; Determining a first grinding depth according to the grinding process parameters and a target physical model; the target physical model is a physical model describing the relationship between the grinding process parameters and the grinding depth; Determining a second polishing depth according to the polishing process parameters and a target neural network model; the target neural network model is a polishing depth prediction neural network model; Determining a target grinding depth according to the first grinding depth and the second grinding depth; The vehicle body paint surface is polished according to the target polishing depth.

[0006] Optionally, determining a target grinding depth according to the first grinding depth and the second grinding depth includes: Acquire a first weight parameter of the target physical model and a second weight parameter of the target neural network model; The sum of the product of the first polishing depth and the first weight parameter and the product of the second polishing depth and the second weight parameter is taken as the target polishing depth.

[0007] Optionally, the target physical model is expressed by the following formula: ; in, Indicates the grinding center point The grinding depth at is the model coefficient, To describe the grinding center point Grinding process parameters and The function of the relationship, The calculation result is the first grinding depth; Then, taking the product of the first grinding depth and the first weight parameter and the product of the second grinding depth and the second weight parameter as the target grinding depth includes: The target grinding depth is determined by a weighted network model: The weighted network model is represented by the following formula: ; in, Grind depth for your target, is the first weight parameter, is the second polishing depth output by the target neural network model, is the second weight parameter.

[0008] Optionally, the ,in, is the grinding tool speed, is the grinding normal force, is the grinding tool feed speed.

[0009] Optionally, the method further includes: The weighted network model is trained by the following method: Acquire multiple groups of sample data; each group of sample data includes: sample process parameters and actual grinding depth corresponding to the sample process parameters; Determining the target grinding depth corresponding to each group of sample process parameters according to the weighted network model; Determining the loss value of the weighted network model according to the difference between the actual grinding depth and the target grinding depth corresponding to each group of the sample process parameters; The first weight parameter and the second weight parameter are adjusted according to the loss value until the loss value meets the preset requirement.

[0010] Optionally, the method further includes: A metal plate sprayed with a preset thickness of paint is used as a grinding material; Determining sample process parameters, and performing a grinding experiment on the grinding material according to the sample process parameters; Determine the height difference of the grinding center of the grinding material before and after each grinding experiment, and use the height difference as the actual grinding depth; The sample process parameters and the actual grinding depth corresponding to each of the grinding experiments are taken as a set of sample data.

[0011] Optionally, determining the sample process parameters includes: The sample process parameters are determined through a preliminary grinding experiment.

[0012] Optionally, the method further includes: Determine the mean square error, mean absolute percentage error and determination coefficient of the weighted network model according to the actual grinding depth and the target grinding depth corresponding to each group of the sample process parameters; The prediction effect of the weighted network model is determined according to the mean square error, mean absolute percentage error and determination coefficient.

[0013] Optionally, the method further includes: The target neural network model is trained by the following method: The input of the target neural network model is set as the polishing process parameters, and the output is ; The target neural network model is trained using the multiple groups of sample data.

[0014] Optionally, the target neural network model includes 6 hidden layers, the hidden layers are linear layers, each of the linear layers is connected to a LeakyReLU activation function, and a Dropout layer is connected to the first and fifth linear layers respectively.

[0015] The beneficial effects of the present invention are: After obtaining the polishing process parameters of the vehicle body paint surface, the first polishing depth is calculated according to the polishing process parameters and the target physical model, and the second polishing depth is predicted according to the polishing process parameters and the target neural network model. The target polishing depth is calculated by combining the first polishing depth and the second polishing depth, and finally the vehicle body paint surface is polished and repaired according to the target polishing depth. When determining the target polishing depth, the present invention uses the calculation results of the target physical model as a constraint condition, which can reduce the demand for samples of the target neural network model during the training process. By determining the final target polishing depth by combining the calculation results of the comprehensive physical model and the prediction results of the neural network model, the accuracy of the target polishing depth can be ensured when there are not many training samples of the target neural network model. Finally, the vehicle body paint surface is polished according to the target polishing depth, which can achieve accurate removal of vehicle body paint defects and avoid the situation of grinding through the putty layer under the paint surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of an embodiment of a grinding depth prediction method provided by the present invention; Figure 2 A schematic diagram of contact between a robot grinding tool and a workpiece provided by the present invention; Figure 3 A schematic diagram of a weighted network model provided by the present invention; Figure 4 A target grinding depth prediction flow chart provided by the present invention; Figure 5 A schematic diagram showing the comparison between the predicted value and the true value of the sample provided by the present invention on the target physical model; Figure 6 A schematic diagram showing the comparison between the predicted value and the true value of the sample provided by the present invention on the target neural network model; Figure 7 A schematic diagram showing the comparison between the predicted value and the true value of the sample provided by the present invention on the weighted network model. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0018] In the description of the embodiments of the present invention, unless otherwise specified, "multiple" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone.

[0019] The descriptions of "first", "second", etc. involved in the embodiments of the present invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the technical features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0020] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] Reference Figure 1 , shows a flow chart of an embodiment of a grinding depth prediction method provided by the present invention, the method comprising: Step S101, obtaining polishing process parameters of the vehicle body paint surface.

[0022] Step S102, determining a first grinding depth according to the grinding process parameters and a target physical model; the target physical model is a physical model that describes the relationship between the grinding process parameters and the grinding depth.

[0023] Step S103, determining the second grinding depth according to the grinding process parameters and the target neural network model; the target neural network model is a grinding depth prediction neural network model.

[0024] Step S104, determining a target grinding depth according to the first grinding depth and the second grinding depth.

[0025] Step S105, polishing the paint surface of the vehicle body according to the target polishing depth.

[0026] The polishing process parameters of the car body paint surface may include: polishing pressure, polishing tool rotation speed, polishing tool feed speed, etc. Among them, the polishing pressure may be the polishing normal force, and the polishing normal force refers to the vertical force applied to the workpiece during the polishing process.

[0027] The target physical model may be a physical model describing the relationship between the grinding process parameters and the grinding depth, and the relationship between the grinding process parameters and the grinding depth may be expressed by a mathematical formula. The first grinding depth may be calculated by combining the target physical model and the grinding process parameters.

[0028] The target neural network model may be a neural network model for predicting the grinding depth. The target neural network model may be a feedforward neural network model, a convolutional neural network model, a recurrent neural network model, and the like. This embodiment does not specifically limit the structure of the target neural network model. After obtaining the grinding process parameters, the grinding process parameters may be input into the target neural network model, and the target neural network model outputs the second grinding depth.

[0029] Furthermore, the weight information of the two models can be obtained, and the target polishing depth can be calculated according to the weight information of the two models and the first polishing depth and the second polishing depth output by the two models respectively. Finally, the car body is polished according to the target polishing depth to remove the paint defects of the car body.

[0030] If only the neural network model is used to predict the grinding depth, there will be insufficient training samples, resulting in poor model training effect and inaccurate model prediction results. If only the physical model is used to predict the grinding depth, due to the multiple restrictions in the construction of the physical model, the prediction effect of the constructed physical model is not ideal in actual prediction.

[0031] After obtaining the polishing process parameters of the vehicle body paint surface, this embodiment calculates the first polishing depth according to the polishing process parameters and the target physical model, and predicts the second polishing depth according to the polishing process parameters and the target neural network model, and calculates the target polishing depth by combining the first polishing depth and the second polishing depth; the vehicle body paint surface is polished according to the target polishing depth. By determining the final target polishing depth by combining the calculation results of the comprehensive physical model and the prediction results of the neural network model, the accuracy of the target polishing depth can be improved. This makes up for the defect that the neural network model makes inaccurate predictions due to insufficient sample size when the neural network model is used alone for prediction. And by polishing the vehicle body paint surface according to the target polishing depth, the defects of the vehicle body paint surface can be accurately removed, avoiding the situation of grinding through the putty layer under the paint surface.

[0032] In one embodiment, step S103 may include: obtaining a first weight parameter of the target physical model and a second weight parameter of the target neural network model; taking the sum of the product of the first polishing depth and the first weight parameter and the product of the second polishing depth and the second weight parameter as the target polishing depth.

[0033] The weight information of the two models may include a first weight parameter and a second weight parameter. The product of the first polishing depth and the first weight parameter and the product of the second polishing depth and the second weight parameter may be calculated respectively, and the sum of the two products is used as the target polishing depth.

[0034] In one embodiment, the target physical model is represented by the following formula: ; in, Indicates the grinding center point The grinding depth at is the model coefficient, is a constant, which can be obtained by least squares fitting. To describe the grinding center point Grinding process parameters and The function of the relationship, or yes The expression after removing the model coefficients from the function, The calculated result is the first grinding depth.

[0035] After obtaining the grinding process parameters, the grinding process parameters can be input into the target neural network model to obtain the second grinding depth output by the target neural network model. , and calculate the first grinding depth according to the grinding process parameters and the above target physical model , and then combine the weight information of the two models to calculate the target grinding depth. However, this method has a large number of parameters (parameters are model coefficients and the first weight parameter) is rather cumbersome to determine.

[0036] Therefore, in one embodiment, the step of taking the sum of the product of the first grinding depth and the first weight parameter and the product of the second grinding depth and the second weight parameter as the target grinding depth may specifically include: determining the target grinding depth through a weighted network model: The weighted network model is represented by the following formula: ; in, Grind depth for your target, is the first weight parameter, is the second polishing depth output by the target neural network model, is the second weight parameter. The first weight parameter and the second weight parameter are obtained by training the weighted network model.

[0037] In the previous example, the formula for the weighted network model can be understood as: , then in the process of calculating the target grinding depth, it is necessary to first calculate the model coefficient in the target physical model , and then calculate the first grinding depth , and finally the first grinding depth and the second grinding depth The target grinding depth is calculated by inputting it into the weighted network model. In this embodiment, ,and It is obtained by training the weighted network model. Therefore, in this embodiment, it is not necessary to first calculate the model coefficients The exact value of , saving one step of calculation operation.

[0038] In one embodiment, the grinding pressure includes a grinding normal force, , , ,in, is the grinding tool speed, is the grinding normal force, is the grinding tool feed speed. This embodiment solves the problem of insufficient grinding sample data and difficulty in determining the grinding coefficient, improves the accuracy of the grinding depth prediction model, and thus more effectively guides grinding process decisions.

[0039] Reference Figure 2 , showing a schematic diagram of contact between a robot grinding tool provided by the present invention and a workpiece. Figure 2 , a portion of a robot grinding tool is shown, which includes a rigid connector, a grinding disc rigid fixing plate, and a grinding disc buffer layer. The grinding disc buffer layer contacts the workpiece to be ground. It is the feeding direction of the grinding disc, which is the same as the positive half-axis direction of the y-axis. It is the grinding normal force, perpendicular to the workpiece plane. It is the grinding disc speed.

[0040] The derivation process is as follows: The formula for grinding depth during static end surface machining is as follows: , in, is the grinding depth per unit time, is the grinding coefficient, It is the grinding pressure. is the linear speed of the grinding disc.

[0041] Taking the center of the grinding disc as the origin, the feed direction as the positive half axis of the y-axis, and the vertical feed direction as the x-axis, the grinding depth formula during the continuous feed end face processing is as follows: , , , in, is the grinding removal depth per unit displacement, represents the pressure distribution in the contact area, is the speed of the grinding disc, is the feed rate of the robot.

[0042] in, Pressure at the location The mapping model between the normal force and the load is based on the finite element simulation software Abaqus, and the formula is as follows:

[0043] in is the fitting parameter value, which can be calculated from the simulation results. is the contact normal force.

[0044] The formula of the grinding depth physical model is as follows:

[0045] Further simplification to obtain the center point of the grinding disc Grinding depth formula: .

[0046] In one embodiment, the weighted network model can be trained by the following method: obtaining multiple groups of sample data; each group of sample data includes: sample process parameters and actual polishing depths corresponding to the sample process parameters; determining the target polishing depths corresponding to each group of sample process parameters according to the weighted network model; determining the loss value of the weighted network model according to the difference between the actual polishing depth corresponding to each group of sample process parameters and the target polishing depth; adjusting the first weight parameter and the second weight parameter according to the loss value until the loss value meets the preset requirements.

[0047] In this embodiment, the sample data can be obtained in the following manner: a metal plate sprayed with a preset thickness of paint is used as a polishing material; sample process parameters are determined, and a polishing experiment is performed on the polishing material according to the sample process parameters; the height difference between the polishing centers of the polishing material before and after each polishing experiment is determined, and the height difference is used as the actual polishing depth; the sample process parameters and the actual polishing depth corresponding to each polishing experiment are used as a set of sample data.

[0048] In this embodiment, the sample process parameters can be determined through a preliminary grinding experiment. Through the preliminary grinding experiment, a suitable sample process parameter range can be obtained, so that when the grinding robot performs a grinding experiment within the sample process parameter range, the actual grinding depth is less than or close to the paint film thickness of the vehicle body. Then the sample process parameters are determined according to the sample process parameter range.

[0049] In this embodiment, the trained weighted network model can also be evaluated in the following way: determine the mean square error, mean absolute percentage error and determination coefficient of the weighted network model according to the actual grinding depth and target grinding depth corresponding to each group of sample process parameters; determine the prediction effect of the weighted network model according to the mean square error, mean absolute percentage error and determination coefficient.

[0050] In this embodiment, the target neural network model can be trained by the following method: the input of the target neural network model is set as the polishing process parameters, and the output is ; Train the target neural network model through multiple groups of sample data.

[0051] Sample data can be obtained through experimental testing. After obtaining the sample data, the sample data can be divided into a training set and a test set. The target neural network model and the weighted network model are trained through the training set. The target neural network model and the weighted network model are tested through the test set.

[0052] In the process of training the model, the loss value of the model can be calculated by the loss function, which can be: in, is the sample label value, is the model prediction value.

[0053] When the loss value cannot be reduced further, the training phase is terminated.

[0054] In one embodiment, the target neural network model may include 6 hidden layers, wherein the hidden layers are linear layers, each of the linear layers is connected to a LeakyReLU activation function, and a Dropout layer is connected to the first and fifth linear layers respectively to reduce the risk of overfitting.

[0055] Reference Figure 3 , showing a schematic diagram of a weighted network model provided by the present invention. The grinding process parameters are respectively input into the target physical model and the target neural network model to obtain the physical model calculation result (first grinding depth) and the neural network prediction result (second grinding depth). The physical model calculation result and the neural network prediction result are input into the weighted network to obtain the target grinding depth.

[0056] Reference Figure 4 , showing a target grinding depth prediction flow chart provided by the present invention.

[0057] S1, the physical model of material removal during continuous end face grinding is constructed using the Preston equation and Abaqus simulation.

[0058] Determine the workpiece to be polished and the polishing disc tool. Obtain the material properties of the polishing disc, and perform simulation using Abaqus simulation software based on the material properties of the polishing disc. Simulate and fit to obtain the relationship function between the contact pressure distribution and the normal force during polishing. Based on the relationship function obtained by simulation fitting, combine the Preston equation to obtain the target physical model.

[0059] S2, the grinding process parameters are used as variables and the grinding depth is used as the quantity to be measured to design an experiment and obtain a small sample data set.

[0060] Preliminary testing is performed to obtain the range of variation of the grinding process parameters. According to the range of variation of the grinding process parameters, several sets of process parameter combinations are generated, and grinding experiments are performed to measure the actual grinding depth during the experiment.

[0061] S3. Construct a weighted network model that integrates the physical model and the neural network model.

[0062] Define a fully connected neural network model and a weighted physical model, and train the weighted network model.

[0063] S4. Train and test the model and calculate the evaluation function.

[0064] After training is completed, the weighted network model is tested and evaluated.

[0065] In one embodiment, in order to better demonstrate the beneficial effects of the present invention, the present invention tests the target physical model, the target neural network model and the weighted physical model.

[0066] The testing process is as follows: Prepare an aluminum alloy plate with a paint surface of about 50 microns sprayed on it as a workpiece to be polished; conduct a preliminary polishing test to obtain a suitable range of process parameters; generate several sample process parameter combinations, conduct polishing experiments, and obtain the actual polishing depth under different sample process parameters. In this embodiment, the polishing operation is controlled by a robot. The hardware used for the polishing operation mainly includes: a six-degree-of-freedom processing robot (IRB 6700-200 / 2.60), an F / T force control sensor (ATI Omega 160), and a central control console (CPU / Core i9-10900K @3.70GHz; GPU / Nvidia GeForce RTX 3060 (12GB)). The polishing tool is a homemade flexible polishing disc (rated speed 8000r / min), and a 5-inch sandpaper (P240) is used. The normal force is used for polishing. , Tool speed , robot feed speed Height difference between the center point before and after grinding As characteristic variables, a total of 117 samples were collected.

[0067] After training the target neural network model and weighted network model using the training samples from the above 117 samples, the target physical model, target neural network model and weighted network model are tested using the test samples from the 117 samples.

[0068] The effect in the target physical model: the mean square error MSE is 9.015×10 -0.5 The mean absolute percentage error MAPE is 0.256 and the coefficient of determination R2 is 0.69.

[0069] The prediction effect on the target neural network model: the mean square error MSE is 7.130×10 -0.5 , the mean absolute percentage error MAPE is 0.213, and the coefficient of determination R2 is 0.731.

[0070] The prediction effect on the weighted network model obtained by fusing the target physical model and the target neural network model: the mean square error MSE is 5.193×10 -0.6The mean absolute percentage error MAPE is 0.063 and the coefficient of determination R2 is 0.986.

[0071] And, refer to Figure 5 , which shows a schematic diagram of the comparison between the predicted value and the true value of the sample provided by the present invention on the target physical model, Figure 5 It can be seen that the predicted value and the true value of the sample on the target physical model are quite different. Figure 6 , which shows a schematic diagram of the comparison between the predicted value and the true value of the sample provided by the present invention on the target neural network model, Figure 6 It can be seen that the predicted value and the true value of the sample on the target neural network model are also quite different. Figure 7 , which shows a schematic diagram of the comparison between the predicted value and the true value of the sample provided by the present invention on the weighted network model, Figure 7 It can be seen that the difference between the predicted value and the true value of the sample on the weighted network model is small, and the prediction results are more accurate than those of the target physical model and the target neural network model.

[0072] Figure 5-Figure 7 In the figure, the vertical axis is the grinding depth, and the unit is , the horizontal axis is the sample number, and one sample number corresponds to a set of sample process parameters. This embodiment forms 3 test sets by random sampling. Figure 5-Figure 7 Schematic diagram for comparing the predicted values ​​and true values ​​obtained from testing the three test sets.

[0073] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0074] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the depth of vehicle body paint repair, characterized in that: include: Obtain the polishing process parameters of the car body paint surface; Determining a first grinding depth according to the grinding process parameters and the target physical model; The target physical model is a physical model that describes the relationship between the grinding process parameters and the grinding depth; Determining a second polishing depth according to the polishing process parameters and a target neural network model; the target neural network model is a polishing depth prediction neural network model; Determining a target grinding depth according to the first grinding depth and the second grinding depth; The vehicle body paint surface is polished according to the target polishing depth.

2. The method for predicting the depth of vehicle body paint repair according to claim 1, characterized in that: The step of determining a target grinding depth according to the first grinding depth and the second grinding depth includes: Acquire a first weight parameter of the target physical model and a second weight parameter of the target neural network model; The sum of the product of the first polishing depth and the first weight parameter and the product of the second polishing depth and the second weight parameter is taken as the target polishing depth.

3. The method for predicting the depth of vehicle body paint repair according to claim 2, characterized in that: The target physical model is expressed by the following formula: ； in, Indicates the grinding center point The grinding depth at is the model coefficient, To describe the grinding center point Grinding process parameters and The function of the relationship, The calculation result is the first grinding depth; Then, taking the product of the first grinding depth and the first weight parameter and the product of the second grinding depth and the second weight parameter as the target grinding depth includes: The target grinding depth is determined by a weighted network model: The weighted network model is represented by the following formula: ； in, Grind depth for your target, is the first weight parameter, is the second polishing depth output by the target neural network model, is the second weight parameter.

4. The method for predicting the depth of vehicle body paint repair according to claim 3, characterized in that: Said ,in, is the grinding tool speed, is the grinding normal force, is the grinding tool feed speed.

5. The method for predicting the depth of vehicle body paint repair according to claim 3, characterized in that: The method further comprises: The weighted network model is trained by the following method: Acquire multiple groups of sample data; each group of sample data includes: sample process parameters and actual grinding depth corresponding to the sample process parameters; Determining the target grinding depth corresponding to each group of sample process parameters according to the weighted network model; Determining the loss value of the weighted network model according to the difference between the actual grinding depth and the target grinding depth corresponding to each group of the sample process parameters; The first weight parameter and the second weight parameter are adjusted according to the loss value until the loss value meets the preset requirement.

6. The method for predicting the depth of vehicle body paint repair according to claim 5, characterized in that: The method further comprises: A metal plate sprayed with a preset thickness of paint is used as a grinding material; Determining sample process parameters, and performing a grinding experiment on the grinding material according to the sample process parameters; Determine the height difference of the grinding center of the grinding material before and after each grinding experiment, and use the height difference as the actual grinding depth; The sample process parameters and the actual grinding depth corresponding to each of the grinding experiments are taken as a set of sample data.

7. The method for predicting the depth of vehicle body paint repair according to claim 5, characterized in that: The step of determining the sample process parameters comprises: The sample process parameters are determined through a preliminary grinding experiment.

8. The method for predicting the depth of vehicle body paint repair according to claim 5, characterized in that: The method further comprises: Determine the mean square error, mean absolute percentage error and determination coefficient of the weighted network model according to the actual grinding depth and the target grinding depth corresponding to each group of the sample process parameters; The prediction effect of the weighted network model is determined according to the mean square error, mean absolute percentage error and determination coefficient.

9. The method for predicting the depth of vehicle body paint repair according to claim 5, characterized in that: The method further comprises: The target neural network model is trained by the following method: The input of the target neural network model is set as the polishing process parameters, and the output is ; The target neural network model is trained using the multiple groups of sample data.

10. The method for predicting the depth of vehicle body paint repair according to claim 1, characterized in that: The target neural network model includes 6 hidden layers, each of which is a linear layer, a LeakyReLU activation function is connected to each linear layer, and a Dropout layer is connected to the first and fifth linear layers respectively.

Images