Stamping part machining procedure determination method and device, equipment and storage medium

Abstract

The invention relates to the technical field of stamping processes, in particular to a stamping part machining procedure determining method and device, equipment and a storage medium, and the method comprises the steps that three-dimensional point cloud data of a to-be-machined stamping part is acquired; the three-dimensional point cloud data of the to-be-machined stamping part is input into the stamping part recognition model, and the stamping part recognition model outputs the part type of the to-be-machined stamping part; a stamping part feature recognition model is determined based on the part type of the to-be-machined stamping part, the three-dimensional point cloud data of the to-be-machined stamping part is input into the stamping part feature recognition model, and the stamping part feature recognition model outputs part features of the to-be-machined stamping part; and determining a machining procedure of the to-be-machined stamping part based on the part characteristics. Therefore, the problems of low efficiency, poor accuracy and the like due to the fact that the machining procedure of the stamping part is manually determined in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of stamping process, in particular to a stamping part processing procedure determination method, device, equipment and storage medium. BACKGROUND

[0002] The stamping process is an indispensable key technology in manufacturing industry, widely used in the fields of automobile, aerospace, etc., and its core is to apply pressure to the plate through the mold to make it produce plastic deformation or separation, so as to obtain the target shape of the part. In the traditional design process of automobile stamping part processing procedure, artificial methods are adopted to manually identify the feature area of the part processing procedure and measure the data of the related feature points, and finally determine the processing procedure of the part according to the experience after calculation. The experience gap of different engineers, the accuracy and efficiency of manual identification all have differences, and each part procedure design needs to be carried out for multiple rounds, and a large amount of time is needed for each round, which is time-consuming and laborious. SUMMARY

[0003] The present application provides a stamping part processing procedure determination method, device, equipment and storage medium to solve the problems of low efficiency and poor accuracy in determining the processing procedure of stamping parts by manual method in related technologies.

[0004] The first aspect embodiment of the present application provides a stamping part processing procedure determination method, comprising the following steps: obtaining three-dimensional point cloud data of a stamping part to be processed; inputting the three-dimensional point cloud data of the stamping part to be processed into a stamping part recognition model, and the stamping part recognition model outputs the part type of the stamping part to be processed; determining a stamping part feature recognition model based on the part type of the stamping part to be processed, and inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, and the stamping part feature recognition model outputs the part feature of the stamping part to be processed; determining the processing procedure of the stamping part to be processed based on the part feature.

[0005] Optionally, before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part recognition model, the method comprises: obtaining training data of the stamping part, wherein the samples in the training data carry type identification of the stamping part, and the type identification is used to identify the type of the stamping part; constructing the stamping part recognition model according to the training data, dividing the training data into a training set and a test set, and iteratively training the stamping part recognition model by using the training set; testing the part recognition accuracy of the trained stamping part recognition model by using the test set until the stamping part recognition model converges, and completing the iterative training of the stamping part recognition model.

[0006] Optionally, in the iterative training process of the stamping part recognition model, if the part recognition accuracy of the stamping part recognition model is lower than a preset value, the training data is expanded.

[0007] Optionally, constructing the stamping part recognition model according to the training data comprises: extracting point cloud data in the training data; iteratively sampling the point cloud data, in the first sampling, randomly selecting a reference point from the point cloud data, recording the Euclidean distance between each point and the reference point, in each iteration, selecting a point with the maximum Euclidean distance between all points in the selected sampling point set as a new sampling point from the remaining points, updating the Euclidean distance between all points and the new sampling point until the number of sampling points in the sampling point set reaches the target number; grouping and segmenting the sampling points in the sampling point set into multiple regions according to spatial proximity, dimensionally increasing the feature channels of the grouped sampling points to obtain feature matrices of different radii; concatenating the feature matrices of different radii to obtain a complete feature matrix, using maximum pooling to aggregate local features in the complete feature matrix into a feature vector, calculating the exponential value of all elements in the feature vector, and normalizing the exponential value; determining the class corresponding to the maximum probability according to the normalized result, taking the class corresponding to the maximum probability as the final classification result, and generating the stamping part recognition model according to the final classification result.

[0008] Optionally, before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, the method further comprises: obtaining training data of the stamping part, wherein the training data carries a feature identifier of the stamping part, and the feature identifier is used to identify the feature region of the stamping part; constructing the stamping part feature recognition model according to the training data, dividing the training data into a training set and a test set, and iteratively training the stamping part feature recognition model using the training set; testing the feature recognition accuracy of the trained stamping part feature recognition model using the test set until the stamping part feature recognition model converges, and completing the iterative training of the stamping part feature recognition model.

[0009] Optionally, in the iterative training process of the stamping part feature recognition model, if the feature recognition accuracy of the stamping part feature recognition model is lower than a preset value, the training data is expanded.

[0010] Optionally, determining the processing procedure of the stamping part to be processed based on the part feature comprises: obtaining a corresponding relationship between the part feature and the processing procedure; and determining the processing procedure of the stamping part to be processed by searching the corresponding relationship.

[0011] The second aspect embodiment of the application provides a stamping part processing procedure determination device, comprising: an acquisition module configured to acquire three-dimensional point cloud data of a stamping part to be processed; a first input module configured to input the three-dimensional point cloud data of the stamping part to be processed into a stamping part recognition model, and the stamping part recognition model outputs a part type of the stamping part to be processed; a second input module configured to determine a stamping part feature recognition model based on the part type of the stamping part to be processed, and input the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, and the stamping part feature recognition model outputs a part feature of the stamping part to be processed; and a determination module configured to determine a processing procedure of the stamping part to be processed based on the part feature.

[0012] Optionally, the device further comprises a first training module configured to, before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part recognition model, acquire training data of the stamping part, wherein a sample in the training data carries a type identification of the stamping part, and the type identification is used to identify the type of the stamping part; construct the stamping part recognition model according to the training data, divide the training data into a training set and a test set, and iteratively train the stamping part recognition model by using the training set; and test the part recognition accuracy of the trained stamping part recognition model by using the test set, until the stamping part recognition model converges, and the iterative training of the stamping part recognition model is completed.

[0013] Optionally, during the iterative training of the stamping part recognition model, if the part recognition accuracy of the stamping part recognition model is lower than a preset value, the training data is expanded.

[0014] Optionally, the first training module is further configured to: extract point cloud data in the training data; iteratively sample the point cloud data, in the first sampling, randomly select a reference point from the point cloud data, and record the Euclidean distance between each point and the reference point, in each iteration, select a point with the maximum Euclidean distance between all points in the selected sample point set as a new sample point from the remaining points, update the Euclidean distance between all points and the new sample point, until the number of sample points in the sample point set reaches a target number; group and segment the sample points in the sample point set into multiple regions according to spatial proximity, obtain feature matrices of different radii by dimensionally increasing the feature channels of the grouped sample points; splice the feature matrices of different radii together to obtain a complete feature matrix, aggregate local features in the complete feature matrix into a feature vector using maximum pooling, calculate the exponential value of all elements in the feature vector, and normalize the exponential value; determine a category corresponding to the maximum probability according to the normalized result, take the category corresponding to the maximum probability as a final classification result, and generate the stamping part recognition model according to the final classification result.

[0015] Optionally, the method further comprises: a second training module configured to obtain training data of the stamping part before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, wherein the training data carries a feature label of the stamping part, and the feature label is used to identify a feature region of the stamping part; constructing the stamping part feature recognition model according to the training data, dividing the training data into a training set and a test set, and iteratively training the stamping part feature recognition model by using the training set; testing the feature recognition accuracy of the trained stamping part feature recognition model by using the test set until the stamping part feature recognition model converges, and completing the iterative training of the stamping part feature recognition model.

[0016] Optionally, during the iterative training of the stamping part feature recognition model, if the feature recognition accuracy of the stamping part feature recognition model is lower than a preset value, the training data is expanded.

[0017] Optionally, the determining module is further configured to: obtain a corresponding relationship between the part feature and the processing procedure; and find the corresponding relationship to determine the processing procedure of the stamping part to be processed.

[0018] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the stamping part processing procedure determination method of the above-mentioned embodiments.

[0019] The fourth aspect of the present application provides a computer readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions are executed by a processor to perform the stamping part processing procedure determination method of the above-mentioned embodiments.

[0020] The fifth aspect of the present application provides a computer program product comprising a computer program or instructions, wherein the computer program or instructions are executed to implement the stamping part processing procedure determination method of the above-mentioned embodiments.

[0021] Therefore, the present application has at least the following beneficial effects: The embodiment of the present application can realize determination of the machining process of the stamping part based on the pre-generated and trained stamping part recognition model and stamping part feature recognition model. Specifically, the stamping part recognition model is used to determine the type of the stamping part, the corresponding stamping part feature recognition model is determined based on the type of the stamping part, the features of the stamping part are output by using the stamping part feature recognition model, and then the machining process of the stamping part is determined according to the features, so as to realize automatic determination of the machining process from the three-dimensional point cloud data of the to-be-machined part, without manual determination, and improve the efficiency and accuracy of the machining process planning. Thus, the technical problems of low efficiency and poor accuracy of manually determining the machining process of the stamping part in the related art are solved.

[0022] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein: Figure 1 A flowchart of a stamping part machining process determination method according to an embodiment of the present application is provided. Figure 2 A flowchart of automatic determination of a stamping part machining process according to an embodiment of the present application is provided. Figure 3 A flowchart of determination of a side wall stamping part machining process according to an embodiment of the present application is provided. Figure 4 An example diagram of a stamping part machining process determination device according to an embodiment of the present application is provided. Figure 5 A structural schematic diagram of an electronic device according to an embodiment of the present application is provided. DETAILED DESCRIPTION

[0024] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0025] A stamping part machining process determination method, device, equipment and storage medium are described below with reference to the accompanying drawings. In view of the low accuracy and efficiency of manual identification of stamping parts to determine the machining process of the parts in the background art, the present application provides a stamping part machining process determination method. In this method, the machining process of the stamping part can be determined based on a pre-generated and trained stamping part recognition model and a stamping part feature recognition model. Specifically, the stamping part recognition model is used to determine the type of the stamping part, the corresponding stamping part feature recognition model is determined based on the type of the stamping part, the features of the stamping part are output using the stamping part feature recognition model, and then the machining process of the stamping part is determined according to the features, thereby realizing automatic determination of the machining process from the three-dimensional point cloud data of the parts to be machined, without the need for manual determination, improving the efficiency and accuracy of machining process planning. Thus, the problems of low efficiency and poor accuracy of manual determination of the machining process of stamping parts in related art are solved.

[0026] Specifically, Figure 1 A flowchart of a stamping part machining process determination method provided by an embodiment of the present application is shown.

[0027] As Figure 1 shown, the stamping part machining process determination method includes the following steps: In step S1, three-dimensional point cloud data of a stamping part to be machined is obtained.

[0028] The stamping part includes a side wall, a wing panel and the like of a vehicle body; the three-dimensional point cloud data is a data set constructed by the coordinates of a large number of points in a three-dimensional space, and is used to represent the geometric shape of the stamping part.

[0029] In step S2, the three-dimensional point cloud data of the stamping part to be machined is input into a stamping part recognition model, and the stamping part recognition model outputs the part type of the stamping part to be machined.

[0030] The part type is the side wall, the wing panel and the like mentioned above.

[0031] It can be understood that the three-dimensional point cloud data of the stamping part to be machined can be input into the stamping part recognition model, and the stamping part recognition model outputs the part type of the stamping part to be machined.

[0032] In the embodiment of the present application, before the three-dimensional point cloud data of the stamping part to be processed is input into the stamping part recognition model, the following steps are included: obtaining training data of the stamping part, wherein the sample in the training data carries a type identification of the stamping part, and the type identification is used to identify the type of the stamping part; constructing a stamping part recognition model according to the training data, dividing the training data into a training set and a test set, and iteratively training the stamping part recognition model using the training set; testing the part recognition accuracy of the trained stamping part recognition model using the test set until the stamping part recognition model converges, and completing the iterative training of the stamping part recognition model.

[0033] It can be understood that the embodiment of the present application can obtain training data of the stamping part, construct a stamping part recognition model using the training data, iteratively train the stamping part recognition model using the training set, test the part recognition accuracy of the trained stamping part recognition model using the test set, until the stamping part recognition model converges, and complete the iterative training of the stamping part recognition model.

[0034] In the embodiment of the present application, constructing a stamping part recognition model according to the training data includes: extracting point cloud data in the training data; iteratively sampling the point cloud data, in the first sampling, randomly selecting a reference point from the point cloud data, recording the Euclidean distance between each point and the reference point, in each iteration, selecting a point with the largest Euclidean distance between all points in the selected sample point set as a new sample point, updating the Euclidean distance between all points and the new sample point, until the number of sample points in the sample point set reaches the target number; grouping and dividing the sample points in the sample point set into multiple regions according to spatial proximity, and increasing the dimension of the feature channel to obtain a feature matrix of different radii; concatenating the feature matrices of different radii to obtain a complete feature matrix, using maximum pooling to aggregate local features in the complete feature matrix into a feature vector, calculating the exponential value of all elements in the feature vector, and normalizing the exponential value; determining the class corresponding to the maximum probability according to the normalized result, taking the class corresponding to the maximum probability as the final classification result, and generating a stamping part recognition model according to the final classification result.

[0035] Specifically, the construction and training process of the stamping part recognition model in the embodiment of the present application is as follows: After obtaining the design data of different stamping parts of the automobile, the design data is converted into point cloud data format. Further, in order to cope with the different scales of the point cloud data of different stamping parts, the input point cloud data is adjusted to a uniform scale through the mean and variance method, so as to ensure the robustness of the subsequent model to different scale data. Further, in order to enhance the generalization ability of the subsequent model, the existing original stamping part point cloud data is stretched, scaled and flipped to generate new stamping part point cloud data. A manual labeling method is used, and all stamping part and feature point cloud data are divided into a model training set and a model test set at a certain ratio, thereby forming a stamping part point cloud database.

[0036] In the stamping part model preparation process, the stamping part point cloud data is sampled, grouped, aggregated and normalized. Further, a point is randomly selected from the point cloud data as the first sampling point, the Euclidean distance between each point and the nearest selected sampling point is recorded, and the iteration operation is performed. In each iteration, the point with the maximum distance from all points in the selected sampling point set is selected from the remaining points as the new sampling point. The distance between all points and the newly selected sampling point is updated. The iteration is performed until a specified number of sampling points are selected. Further, the point cloud data is grouped according to spatial proximity and divided into multiple local regions. Each region represents the local environment of a point to capture local geometric features. Further, the grouped point cloud information is used to realize feature channel dimensionality increase using a series of multi-layer perceptrons. The feature matrices of different radii are spliced together to retain local details and capture global structures, forming a complete feature matrix suitable for both macro and micro scales. A maximum pooling is used to aggregate the local features in the neighborhood into a single feature vector to represent the sampling point feature. Further, the exponential values of all elements of the vector are calculated, the above exponential values are normalized to ensure that the sum of all outputs is 1, and the class corresponding to the maximum probability is selected as the final classification result, thereby obtaining the initial model of the stamping part data; In the stamping part model generation process, the training set samples are sent into the initial model of the stamping part data in batches, and forward propagation is performed. The model predicts and classifies the sample data, calculates the cross-entropy loss between the classification result and the true label, calculates the gradient through back propagation, updates the model parameters using the optimizer function, evaluates the validation set data after each training round, saves the best model in the current round, and repeats the above process until the model converges, thereby forming a stamping part recognition model.

[0037] In the embodiment of the present application, during the iterative training of the stamping part recognition model, if the part recognition accuracy of the stamping part recognition model is lower than a preset value, the training data is expanded.

[0038] The preset value can be set according to specific conditions, such as 90% or 95%.

[0039] It can be understood that in the iterative training process of the stamping part recognition model, when the part recognition accuracy of the stamping part recognition model is lower than the preset value, the training data is expanded, the sample diversity is increased, and then the robustness and accuracy of the stamping part recognition model are improved.

[0040] In step S3, the stamping part feature recognition model is determined based on the type of the stamping part to be processed, and the three-dimensional point cloud data of the stamping part to be processed is input into the stamping part feature recognition model. The stamping part feature recognition model outputs the part features of the stamping part to be processed.

[0041] The feature recognition model of different types of stamping parts is different, and each type of stamping part corresponds to a stamping part feature recognition model. The feature type can include surface features, hole features, etc.

[0042] It can be understood that the stamping part feature recognition model corresponding to the type of the part to be processed can be determined according to the type of the part to be processed, and the three-dimensional point cloud data is input into the stamping part feature recognition model. The stamping part feature recognition model outputs the part feature type of the stamping part to be processed.

[0043] In the embodiment of the application, before the three-dimensional point cloud data of the stamping part to be processed is input into the stamping part feature recognition model, it further includes: obtaining training data of the stamping part, wherein the training data carries a feature identifier of the stamping part, and the feature identifier is used to identify the feature area of the stamping part; constructing a stamping part feature recognition model according to the training data, dividing the training data into a training set and a test set, and using the training set to iteratively train the stamping part feature recognition model; using the test set to test the feature recognition accuracy of the trained stamping part feature recognition model until the stamping part feature recognition model converges, and completing the iterative training of the stamping part feature recognition model.

[0044] Similarly, the stamping part feature recognition model also needs to be trained in the specific application of the stamping part feature recognition model, and the training process is similar to that of the stamping part recognition model.

[0045] In the embodiment of the application, in the iterative training process of the stamping part feature recognition model, if the feature recognition accuracy of the stamping part feature recognition model is lower than the preset value, the training data is expanded.

[0046] The preset value can be set according to specific conditions, such as 90% or 95%.

[0047] It can be understood that in the iterative training process of the stamping part feature recognition model, when the feature recognition accuracy of the stamping part feature recognition model is lower than the preset value, the training data can be expanded, the sample diversity can be increased, and then the robustness and accuracy of the stamping part feature recognition model can be improved.

[0048] Specifically, the construction and training process of the stamping part feature recognition model according to the embodiments of the present application is as follows: The single stamping part feature recognition model preparation process is to perform data sampling, grouping, aggregation and normalization processing on the point cloud data of the feature region of the stamping part selected in advance, and the initial model of the current stamping part feature region data is obtained according to the repeated stamping part model preparation process. The single stamping part feature recognition model generation process is to put the current stamping part feature training set sample into the initial model of the current stamping part feature region data, and the current stamping part feature recognition model is formed according to the repeated stamping part model generation process. The single stamping part feature recognition model generation process is to put the current stamping part feature training set sample into the initial model of the current stamping part feature region data, and the current stamping part feature recognition model is formed according to the repeated stamping part model generation process.

[0049] In step S4, the machining process of the stamping part to be machined is determined based on the part feature.

[0050] It can be understood that the embodiments of the present application can determine the machining process of the stamping part to be machined based on the part feature, so as to realize automatic determination from the three-dimensional point cloud data of the part to be machined to the machining process, without human determination, and improve the efficiency and accuracy of the machining process planning.

[0051] In the embodiments of the present application, the machining process of the stamping part to be machined is determined based on the part feature, including: obtaining the correspondence between the part feature and the machining process; and finding the correspondence to determine the machining process of the stamping part to be machined.

[0052] It can be understood that the embodiments of the present application can determine the machining process of the stamping part to be machined based on the part feature and the correspondence.

[0053] In summary, the automatic determination method of the stamping part machining process according to the embodiments of the present application includes: 1. Stamping part sample processing.

[0054] After obtaining the design data of different stamping parts of the automobile, they are converted into point cloud data format. Further, to cope with the different scales of the point cloud data of different stamping parts, the input point cloud data is adjusted to a uniform scale through the mean and variance method to ensure the robustness of the subsequent model in dealing with different scale data. Further, to enhance the generalization ability of the subsequent model, the existing original stamping part point cloud data is stretched, scaled, and flipped to generate new stamping part point cloud data. A human-labeled method is used, and all stamping part and feature point cloud data are divided into a model training set and a model test set at a certain ratio. Thus, a point cloud database of stamping parts is formed.

[0055] 2. Stamping part recognition model preparation process.

[0056] The stamping part point cloud data is sampled, grouped, aggregated, and normalized. Further, a point is randomly selected from the point cloud data as the first sampling point, the Euclidean distance between each point and the nearest selected sampling point is recorded, and the iteration operation is performed. In each iteration, the point with the maximum distance from all points in the selected sampling point set is selected as the new sampling point. The distance between all points and the newly selected sampling point is updated. The iteration is continued until a specified number of sampling points are selected. Further, the point cloud data is grouped according to spatial proximity and divided into multiple local regions. Each region represents the local environment of a point to capture local geometric features. Further, the grouped point cloud information is used to increase the dimension of the feature channel using a series of multilayer perceptrons. The feature matrices of different radii are spliced together to preserve local details and capture global structures, forming a complete feature matrix suitable for both macro and micro scales. The local features in the neighborhood are aggregated into a single feature vector using max pooling to represent the sampling point features. Further, the exponential values of all elements of the vector are calculated, the above exponential values are normalized to ensure that the sum of all outputs is 1, and the class corresponding to the maximum probability is selected as the final classification result. Thus, the initial stamping part recognition model is obtained.

[0057] 3. Stamping part recognition model generation process.

[0058] The training set samples are sent into the initial stamping part data model in batches for forward propagation. The model predicts and classifies the sample data, calculates the cross-entropy loss between the classification result and the true label, calculates the gradient through backpropagation, updates the model parameters using the optimizer function, evaluates the validation set data after each training round, saves the best model in the current round, and repeats the above process until the model converges, thereby forming the stamping part recognition model.

[0059] 4. Single stamping part feature recognition model preparation process.

[0060] Data sampling, grouping, aggregation and normalization processing are performed on the point cloud data of a certain feature area of a certain stamping part, and the initial model of the data of the certain feature area of the current stamping part is obtained by repeating the stamping part model preparation process; 5. A certain feature recognition model generation process of a single stamping part.

[0061] The certain feature training set sample of the current stamping part is put into the initial model of the data of the certain feature area of the current stamping part, and the certain feature recognition model of the current stamping part is formed by repeating the stamping part model generation process.

[0062] 6. A certain feature recognition model generation process of a single stamping part.

[0063] The certain feature recognition model generation process of a single stamping part is repeated to complete model recognition of all features of the stamping part, and the certain feature recognition model of the current stamping part is formed. 7. A stamping part feature recognition process.

[0064] Any stamping part is put into the stamping part recognition model, and the stamping part is identified as what kind of stamping part, and the stamping part feature recognition model is used to obtain the overall feature of the stamping part.

[0065] 8. A stamping part process generation process.

[0066] According to the correspondence between the known stamping part process and the features thereof, the overall feature of a certain stamping part that has been identified is used to correspond to the required process, so as to realize intelligent design of the stamping part machining process.

[0067] Based on the above automatic determination method of the stamping part machining process, the application further provides an automatic design system, comprising: A stamping part sample processing module is used for stamping part design data conversion, point cloud data unification, and training set and test set construction. A stamping part recognition model generation module is used for data sampling, grouping, aggregation and normalization processing of stamping part point cloud data, and the stamping part recognition model is constructed through multiple rounds of training until the model converges. A stamping part feature recognition model generation module is used for data sampling, grouping, aggregation and normalization processing of the overall feature point cloud data of a single stamping part, and the stamping part feature recognition model is constructed through multiple rounds of training until the model converges. A stamping part feature recognition module is used for feature recognition of any stamping part. The stamping part processing procedure automatic generation module automatically generates a procedure according to a known correspondence between a stamping part processing procedure and a feature thereof.

[0068] According to the stamping part processing procedure determination method provided in the embodiments of the present application, the processing procedure of a stamping part can be determined based on a pre-generated and trained stamping part recognition model and a stamping part feature recognition model. Specifically, the stamping part recognition model is used to determine the type of the stamping part, the corresponding stamping part feature recognition model is determined based on the type of the stamping part, the features of the stamping part are output by using the stamping part feature recognition model, and then the processing procedure of the stamping part is determined according to the features, so as to realize the automatic determination of the processing procedure from the three-dimensional point cloud data of the part to be processed, without the need for manual determination, and improve the efficiency and accuracy of the processing procedure planning.

[0069] In general, the stamping part processing procedure determination method provided in the embodiments of the present application converts different stamping part data into a point cloud format, adjusts a unified scale and shape transformation, constructs a model training set and a test set, performs data sampling, grouping, aggregation and normalization processing on the stamping part point cloud data, forms an initial model of the stamping part data, trains the model by using the stamping part training set, verifies the model by using the stamping part test set, forms a stamping part recognition model, repeats the labeling and training by using the method steps of constructing the stamping part model for a single stamping part feature recognition model, forms a stamping part feature recognition model, puts a new stamping part into the constructed stamping part feature recognition model, obtains the features of the part, and obtains the processing procedure of the part according to the correspondence between the features and the procedures. Compared with the existing method, the automation rate is 100%, the flexibility degree is improved by 80%, and the time consumption is 1% of the original time.

[0070] As shown in Figure 2 the stamping part processing procedure automatic determination process includes: Step 101: Convert different stamping part data into a point cloud format, adjust a unified scale, and perform shape transformation to generate new stamping part point cloud data, then divide the model training set and the model test set by using a certain proportion, and form a point cloud database of the stamping part; Step 102: Perform data sampling, grouping, aggregation and normalization processing on the stamping part point cloud data to form an initial model of the stamping part data; Step 103: Send the stamping part training set samples into the initial model of the stamping part data in batches for multi-round training, evaluate the data of the validation set after each round of training, save the model with the best performance in the current round, and repeat the above process until the model converges, so as to form a stamping part recognition model; Step 104: If the part recognition rate is less than 90% in the verification process, it means that the data sample needs to be expanded for further training, or a new algorithm needs to be reconstructed to complete the model training.

[0071] Step 105: The point cloud data of a pre-selected feature area of a certain stamping part is sampled, grouped, aggregated and normalized to obtain an initial model of the feature area data of the current stamping part; Step 106: The training set sample of a pre-selected feature area of a certain stamping part is sent into the initial model of the feature area data of the current stamping part in batches for multiple rounds of training. After each round of training, the validation set data is used for evaluation, the best model in the current round is saved, and the above process is repeated until the model converges, thereby forming a complete feature recognition model of the current stamping part. The above operation is performed on all features of the stamping part to obtain a complete feature recognition model of the known stamping part. Step 107: Verify the part feature recognition model. If the feature recognition rate is less than 90%, optimize the algorithm for local features or increase the data volume to expand the training sample.

[0072] Step 108: Put any stamping part into the stamping part recognition model to identify what kind of stamping part it is, and then obtain the complete feature information of the stamping part by the complete feature recognition model of the stamping part. According to the correspondence between the known stamping part process and its features, the required process is obtained from the complete feature information of the identified stamping part, thereby realizing intelligent design of stamping part processing.

[0073] The following describes the processing procedure determination method of the application by taking the left and right side wall stamping parts of a vehicle as an example. The specific process is shown in Figure 3 , which includes: Step 211: After obtaining the side wall stamping part design data of 46 vehicle models, the data is first converted into standard three-dimensional STL data format, and then into point cloud data format; Step 212: The obtained side wall point cloud data of different sizes is adjusted to a uniform scale by mean and variance method, and the point cloud coordinates are standardized to a Gaussian distribution with a mean of 0 and a variance of 1, to ensure the robustness of the subsequent side wall part model to different scale data; Step 213: Stretch, scale and flip operations are performed on the existing original side wall stamping part point cloud data. The three axes of the side wall three-dimensional model are introduced into the scaling and stretching transformation with a scale of 0.7 to 1.3, and the original 46 side wall samples are expanded to 388 samples; Step 214: The 14 features of the side wall stamping part are marked by manual annotation; Step 215: The point cloud data of all side body stamping parts and features is divided into 80% training set and 20% test set according to the 8:2 principle to form a point cloud database taking the side body stamping parts as a whole; Step 221: Sampling of side body stamping part point cloud data is performed, mainly using the farthest point sampling method, a point is randomly selected from the point cloud data as the first sampling point, the Euclidean distance between each point and the nearest selected sampling point is recorded, and in each iteration, the point with the maximum distance from all points in the selected sampling point set is selected from the remaining points as the new sampling point. Update the distance between all points and the newly selected sampling point. Iterate until 10,000 sampling points are selected; Step 222: Grouping of side body part sampling points is performed, the point cloud data is grouped according to spatial proximity and divided into multiple local regions. If the number of points in the neighborhood exceeds n, n points are randomly selected, and if not, the points are resampled. Each region represents the local environment of a point to capture local geometric features; Step 223: Aggregation of side body part sampling points is performed, the grouped point cloud information is used to realize feature dimensionality increase of 1024 channels using a series of multi-layer perceptrons. Feature matrices of different radii are spliced together to preserve local details and capture global structure, forming a complete feature matrix suitable for both macro and micro scales. Maximum pooling is used to aggregate local features in the neighborhood into a single feature vector representing the sampling point feature; Step 224: Normalization of side body part sampling points is performed, the exponential value of all elements of the vector is calculated, the above exponential value is normalized to ensure that the sum of all outputs is 1, and the class corresponding to the maximum probability is selected as the final classification result. Thus, the initial model of the side body part data is obtained; Step 231: The training set samples are sent into the initial model of the side body part data in batches for forward propagation, the model predicts and classifies the sample data, calculates the cross-entropy loss between the classification result and the true label, calculates the gradient through back propagation, updates the model parameters using the optimizer function, evaluates the validation set data after each training round, saves the best model in the current round, and repeats the above process 200 times until the model converges, thereby forming a side body part recognition model; Step 232: The side body part recognition model formed by training is verified, and the 20% data used for verification can be identified, otherwise the algorithm is optimized or the data training amount is increased; Step 240: Side wall part internal feature recognition model preparation, since the side wall part has 14 feature points to be recognized, the feature types include holes, surfaces and distances, but the 14 feature points are not uniformly distributed, therefore, according to the distribution, the side wall part point cloud data is manually segmented into 4 parts by manual method, data sampling, grouping, aggregation and normalization processing are performed, and the specific implementation repeats steps 221 to 224, thereby obtaining the initial model of the current stamping part feature area data.

[0074] Step 250: After the initial construction of the side wall part internal feature recognition model is completed, the steps of steps 231 and 232 are repeated to form the side wall part internal feature recognition model training Step 261: Then, upsampling is also needed, each point in the point cloud file with a relatively coarse resolution is taken as a center point, and an interpolation method is used to find the k nearest points to the center point in the original point cloud, thereby forming a more dense point cloud, and each point is classified to obtain the class label of each point, so as to determine which part of the original point cloud the point is in, that is, which area of the side wall part the point belongs to.

[0075] Step 262: After the class label of each point is determined, one of the 14 features in the side wall part to which the point belongs can be located, and the position information of the feature is returned, and the processing method of the feature can be obtained, taking the distance feature as an example, the application first obtains the candidate area through model forward reasoning, then obtains the final endpoint of the point in the candidate area through average pooling, and finally calculates the specific value of the distance feature, solidifies the feature result, and the implementation manners of other features are the same as above. Repeat the above process to obtain the feature results corresponding to all points to form the side wall part internal feature recognition model.

[0076] Step 270: The side wall part of the new vehicle model is put into the side wall part recognition model, and the side wall part is accurately recognized, and the complete feature recognition model of the side wall part is obtained. According to the correspondence between the known side wall part process and the feature, the complete feature of the recognized side wall part is obtained, and the required process is obtained, thereby realizing intelligent design of the side wall part processing process, and the other similar parts, such as the fender, the front hood, etc. Stamping parts are also applicable.

[0077] In addition, it should be noted that the embodiments of the application improve the PointNet++ algorithm as follows: 1. Dynamic downsampling multiple is adopted.

[0078] Different features have different scales. If the same down-sampling factor is taken, the recognition accuracy of small features will be poor and the recognition speed of large features will be slow. According to the scale of the detection target, the application takes a dynamic down-sampling factor. For small holes and other features, a down-sampling factor of 1 / 10 is taken. The accuracy of recognition can be ensured. For surface features, a down-sampling factor of 1 / 50 is taken. The inference speed is ensured.

[0079] 2. A multi-scale data enhancement strategy is taken.

[0080] In the case of less data, the application takes a multi-scale data enhancement strategy. Stretch or scale in three axes with a factor of 0.7-1.3. Randomly crop small areas on the point cloud, which is equivalent to regularizing the data.

[0081] 3. Different loss functions are designed for different features.

[0082] Since the feature often only occupies a small part of the entire part, the size of the foreground and background is very uneven, so a dynamically adjusted loss function is necessary. On the basis of cross-entropy loss, a dynamic weight function is added to calculate the proportion of foreground and background, and the inverse of the weight is taken.

[0083] 4. Post-processing.

[0084] For different features, the application designs different post-processing methods. For the surface, since the escape of abnormal points may occur, the highest confidence area is first obtained by density clustering, and then the least square fitting surface is used to remove points with a distance greater than the threshold from the fitting surface. For the distance, the candidate area is first obtained by forward inference of the model, and the final end point is obtained by average pooling of the points in the candidate area.

[0085] Secondly, the stamping part machining process determination device according to the embodiment of the application is described with reference to the accompanying drawings.

[0086] Figure 4 is a block schematic diagram of the stamping part machining process determination device of the embodiment of the application.

[0087] As shown in Figure 4 , the stamping part machining process determination device 10 comprises an acquisition module 100, a first input module 200, a second input module 300 and a determination module 400.

[0088] The acquisition module 100 is configured to acquire three-dimensional point cloud data of a stamping part to be processed; the first input module 200 is configured to input the three-dimensional point cloud data of the stamping part to be processed into a stamping part recognition model, and the stamping part recognition model outputs a part type of the stamping part to be processed; the second input module 300 is configured to determine a stamping part feature recognition model based on the part type of the stamping part to be processed, and input the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, and the stamping part feature recognition model outputs a part feature of the stamping part to be processed; and the determination module 400 is configured to determine a processing procedure of the stamping part to be processed based on the part feature.

[0089] In the embodiment of the present application, the device 10 further comprises a first training module.

[0090] The first training module is configured to acquire training data of the stamping part before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part recognition model, wherein the sample in the training data carries a type identification of the stamping part, and the type identification is used to identify the type of the stamping part; the stamping part recognition model is constructed according to the training data, the training data is divided into a training set and a test set, and the stamping part recognition model is iteratively trained by using the training set; the part recognition accuracy of the trained stamping part recognition model is tested by using the test set, until the stamping part recognition model converges, and the iterative training of the stamping part recognition model is completed.

[0091] In the embodiment of the present application, if the part recognition accuracy of the stamping part recognition model is lower than a preset value during the iterative training of the stamping part recognition model, the training data is expanded.

[0092] In the embodiment of the present application, the first training module is further configured to: extract point cloud data in the training data; iteratively sample the point cloud data, in the first sampling, randomly select a reference point from the point cloud data, and record the Euclidean distance between each point and the reference point, in each iteration, select a point with the maximum Euclidean distance between all points in the selected sample point set as a new sample point from the remaining points, update the Euclidean distance between all points and the new sample point, until the number of sample points in the sample point set reaches a target number; group and segment the sample points in the sample point set into multiple regions according to spatial proximity, obtain feature matrices of different radii by dimension increasing of the grouped sample points; splice the feature matrices of different radii together to obtain a complete feature matrix, aggregate local features in the complete feature matrix into a feature vector using maximum pooling, calculate the exponential value of all elements in the feature vector, and normalize the exponential value; determine the category corresponding to the maximum probability according to the normalized result, take the category corresponding to the maximum probability as the final classification result, and generate the stamping part recognition model according to the final classification result.

[0093] In the embodiment of the present application, the device 10 of the embodiment of the present application further comprises a second training module.

[0094] The second training module is configured to: obtain training data of the stamping part before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, wherein the sample in the training data carries a feature identifier of the stamping part, and the feature identifier is used to identify the feature region of the stamping part; construct the stamping part feature recognition model according to the training data, divide the training data into a training set and a test set, and iteratively train the stamping part feature recognition model by using the training set; test the feature recognition accuracy of the trained stamping part feature recognition model by using the test set until the stamping part feature recognition model converges, and complete the iterative training of the stamping part feature recognition model.

[0095] In the embodiment of the present application, during the iterative training of the stamping part feature recognition model, if the feature recognition accuracy of the stamping part feature recognition model is lower than a preset value, the training data is expanded.

[0096] In the embodiment of the present application, the determination module 400 is further configured to: obtain a correspondence between the feature of the part and the processing procedure; and find the processing procedure of the stamping part to be processed according to the correspondence.

[0097] It should be noted that the above explanation of the embodiment of the method for determining the processing procedure of the stamping part is also applicable to the device for determining the processing procedure of the stamping part, which will not be described here.

[0098] According to the device for determining the processing procedure of the stamping part provided in the embodiment of the present application, the processing procedure of the stamping part can be determined based on the stamping part recognition model and the stamping part feature recognition model which are generated and trained in advance. Specifically, the stamping part recognition model is used to determine the type of the stamping part, the corresponding stamping part feature recognition model is determined based on the type of the stamping part, the features of the stamping part are output by using the stamping part feature recognition model, and then the processing procedure of the stamping part is determined according to the features, so as to realize the automatic determination of the processing procedure from the three-dimensional point cloud data of the part to be processed, without the need for manual determination, thereby improving the efficiency and accuracy of the processing procedure planning.

[0099] Figure 5 The structure schematic diagram of the electronic device provided in the embodiment of the present application is shown. The electronic device can include: The memory 501, the processor 502, and the computer program stored in the memory 501 and executable on the processor 502.

[0100] The processor 502 implements the stamping part machining process determination method provided in the above embodiments when executing a program.

[0101] Further, the electronic device further comprises: The communication interface 503 is configured to communicate between the memory 501 and the processor 502.

[0102] The memory 501 is configured to store a computer program executable on the processor 502.

[0103] The memory 501 can include a high-speed RAM memory, and can further include a non-volatile memory, for example, at least one disk memory.

[0104] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 5 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0105] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can complete communication between each other through an internal interface.

[0106] The processor 502 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the present application.

[0107] The embodiments of the present application further provide a computer readable storage medium, which stores a computer program or instructions, and the computer program or instructions are executed by a processor to implement the stamping part machining process determination method as above.

[0108] The embodiment of the present application further provides a computer program product comprising a computer program or instructions, which, when executed, implement the stamping part processing procedure determination method as above.

[0109] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or N embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples, without contradiction.

[0110] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0111] Any process or method descriptions in flow charts or described elsewhere herein can be understood as representing code modules, segments, or portions of code which include one or more executable instructions for implementing the specified logical functions or steps, and the preferred embodiments of the present application also include the possibility that the functions described can be implemented using hardware, software, firmware, or any combination thereof. The preferred embodiments of the present application should be understood to include the possibility that the functions described can be implemented using hardware, software, firmware, or any combination thereof.

[0112] It should be understood that parts of the present application can be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. As in another embodiment, if implemented in hardware, any one or more of the following technologies known in the art can be used: discrete logic circuit with logic gates for implementing logical functions on data signals, application specific integrated circuit with suitable combination logic gates, programmable gate array, field programmable gate array, etc.

[0113] Those skilled in the art can understand that all or part of the steps of the foregoing method embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium. When the programs are executed, the steps of the method embodiments or a combination thereof are included.

Claims

1. A method of determining a press part processing procedure, characterized by, The method comprises the following steps: acquiring three-dimensional point cloud data of a stamping part to be processed; inputting the three-dimensional point cloud data of the stamping part to be processed into a stamping part recognition model, the stamping part recognition model outputting a part type of the stamping part to be processed; determining a stamping part feature recognition model based on the part type of the stamping part to be processed, and inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, the stamping part feature recognition model outputting a part feature of the stamping part to be processed; determining a processing procedure of the stamping part to be processed based on the part feature.

2. The press part machining process determination method according to Claim 1, characterized by, Before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part recognition model, the method comprises the following steps: acquiring training data of a stamping part, wherein a type identifier of the stamping part is carried in a sample in the training data, and the type identifier is used to identify a type of the stamping part; constructing a stamping part recognition model according to the training data, dividing the training data into a training set and a test set, and iteratively training the stamping part recognition model by using the training set; testing part recognition accuracy of the trained stamping part recognition model by using the test set, until the stamping part recognition model converges, and the iterative training of the stamping part recognition model is completed.

3. The press part processing procedure determination method according to Claim 2, characterized by, During the iterative training of the stamping part recognition model, if the part recognition accuracy of the stamping part recognition model is lower than a preset value, the training data is expanded.

4. The press part processing procedure determination method according to Claim 2, characterized by The method of constructing a stamping part recognition model according to the training data comprises the following steps: extracting point cloud data in the training data; iteratively sampling the point cloud data, in the first sampling, randomly selecting a reference point from the point cloud data, recording the Euclidean distance between each point and the reference point, in each iteration, selecting a point with the maximum Euclidean distance between all points in the selected sampling point set from the remaining points as a new sampling point, updating the Euclidean distance between all points and the new sampling point, until the number of sampling points in the sampling point set reaches a target number; grouping and dividing the sampling points in the sampling point set into multiple regions according to spatial proximity, and increasing the dimension of the feature channel of the grouped sampling points to obtain feature matrices of different radii; splicing the feature matrices of different radii together to obtain a complete feature matrix, using maximum pooling to aggregate local features in the complete feature matrix into a feature vector, calculating the exponential value of all elements in the feature vector, and normalizing the exponential value; determining a category corresponding to a maximum probability according to the normalized result, taking the category corresponding to the maximum probability as a final classification result, and generating a stamping part recognition model according to the final classification result.

5. The press part machining process determination method according to Claim 1, characterized by, Before inputting the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, the method further comprises the following steps: acquiring training data of a stamping part, wherein a feature identifier of the stamping part is carried in a sample in the training data, and the feature identifier is used to identify a feature region of the stamping part; According to the training data, the stamping part feature recognition model is constructed, the training data is divided into a training set and a test set, and the stamping part feature recognition model is iteratively trained using the training set; The feature recognition accuracy of the trained stamping part feature recognition model is tested using the test set until the stamping part feature recognition model converges, and the iterative training of the stamping part feature recognition model is completed.

6. The press part processing procedure determination method according to Claim 5, characterized by, During the iterative training of the stamping part feature recognition model, if the feature recognition accuracy of the stamping part feature recognition model is less than a preset value, the training data is expanded.

7. The press part machining process determination method according to Claim 1, characterized by, The processing procedure of the stamping part is determined based on the part feature, including: Obtaining the correspondence between the part feature and the processing procedure; Finding the processing procedure of the stamping part to be processed by searching the correspondence.

8. A device for determining the processing steps of stamped parts, characterized in that, It includes: An acquisition module is configured to acquire three-dimensional point cloud data of a stamping part to be processed. A first input module is configured to input the three-dimensional point cloud data of the stamping part to be processed into a stamping part recognition model, and the stamping part recognition model outputs a part type of the stamping part to be processed. A second input module is configured to determine a stamping part feature recognition model based on the part type of the stamping part to be processed, and input the three-dimensional point cloud data of the stamping part to be processed into the stamping part feature recognition model, and the stamping part feature recognition model outputs a part feature of the stamping part to be processed. A determination module is configured to determine a processing procedure of the stamping part to be processed based on the part feature.

9. An electronic device, comprising: It includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the stamping part processing procedure determination method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the stamping part processing procedure determination method according to any one of claims 1-7.

Images