SVM (Support Vector Machine)-based multi-attribute aeronautical part classification and identification method, medium and equipment

Abstract

The invention discloses a multi-attribute aeronautical part classification and identification method based on SVM. The method comprises the following steps: acquiring data and preprocessing the data; analyzing and acquiring attribute information of the part based on the three-dimensional model, marking the type of the part, and then cleaning data; data coding preprocessing and feature fusion; processing the attribute information by using a large language model to obtain a multi-attribute feature vector, taking the multi-attribute feature vector as a data set, and splitting the data set into a test set and a training set; constructing an SVM model, and training the SVM model based on the training set; selecting a radial basis function as a kernel function of a support vector machine; performing cross validation and grid search; performing cross validation on the SVM model based on the test set, and then systematically traversing a parameter space to find an optimal model parameter; and using the trained SVM model to predict new data. According to the method, the semantic understanding of the LLM model and the classification capability of the SVM are combined, so that the aeronautical parts are classified, the classification accuracy and efficiency are improved, and the method has relatively good practicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of aviation part recognition, and particularly relates to a multi-attribute aviation part classification and recognition method, medium and device based on SVM. Background Art

[0002] Aviation parts and components have the characteristics of small batch, multiple types, complex structures, etc. Starting from the process design, combining different parts with the same production method according to certain rules to achieve the standardization and modularization of the process is an important way to improve the assembly efficiency of aviation parts and components. However, there are still a large number of manual operations in the current aviation product design stage, resulting in different descriptive terms, such as part names, part annotations, etc., making it difficult to group and classify parts for recognition. The traditional manual classification method is inefficient and easily affected by subjective factors.

[0003] With the development of computer technology and computer hardware, in the aircraft manufacturing industry, digital manufacturing technology based on digital quantity transfer has generally replaced two-dimensional drawings. Therefore, the present invention provides a multi-attribute aviation part classification and recognition solution based on SVM, integrating the design data of all zero-standard parts in a three-dimensional model for subsequent use, liberating process personnel from two-dimensional drawings for hundreds of years, realizing the high degree of concentration, coordination and integration of environments such as product design, process design, tooling design, part processing, component assembly, and part inspection and testing, and accumulating a large amount of empirical data. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-attribute aviation part classification and recognition method, medium and device based on SVM, aiming to realize the automatic grouping and classification recognition of aviation parts.

[0005] The present invention is mainly realized through the following technical solutions: To better implement the present invention, further, The multi-attribute aviation part classification and recognition method based on SVM includes the following steps: Step S1: Obtain data and perform preprocessing; parse the attribute information of the part based on the three-dimensional model and label its type, and then clean the data; Step S2: Data encoding preprocessing and feature fusion; use a large language model to process the attribute information to obtain a multi-attribute feature vector, and use it as a data set, and split the data set into a test set and a training set; Step S3: Construct an SVM model and train the SVM model based on the training set; select the radial basis function as the kernel function of the support vector machine, map the data to an infinite-dimensional space, and realize the linear segmentation of the data in the high-dimensional space; Step S4: Cross-validation and grid search; perform cross-validation on the SVM model based on the test set, and then systematically traverse the parameter space to find the optimal model parameters; Step S5: Use the SVM model in Step S4 to predict new data.

[0006] To better implement the present invention, further, in Step S1, for text data, in combination with an aviation dictionary, use the jieba library to clean the data, remove stop words, punctuation marks, and convert English characters to lowercase, and perform word segmentation on the remaining strings.

[0007] To better implement the present invention, further, in Step S1, the attribute information includes non-geometric attributes and geometric attributes of the part. The non-geometric attributes of the part include any one or more of drawing number, name, material name, material grade, blank size, and thickness; the geometric attributes of the part include any one or more of part size, volume, center point, surface area, and center of gravity.

[0008] To better implement the present invention, further, in Step S2, based on a large language model, encode the descriptive text in the attribute information, combine the attribute name and attribute content and convert them into high-dimensional vectors to capture the deep semantic information in the description content; finally, directly splice the feature vectors of each attribute to form a multi-attribute feature vector, and comprehensively express the part by combining text information and numerical information.

[0009] To better implement the present invention, further, in Step S2, finally, use the sum function of the numpy library to add the feature vectors of each attribute and form a multi-attribute feature vector.

[0010] To better implement the present invention, further, Step S3 includes the following steps: Step S31: Initialize the SVM model using the SVC class of the scikit-learn library; Step S32: By setting the kernel to "rbf", select the radial basis function as the kernel function; then set the penalty parameter C and gamma parameter of the SVM model; Step S33: Finally, call the fit method to complete the training of the SVM model.

[0011] To better implement the present invention, further, Step S4 includes the following steps: Step S41: First, use the GridSearchCV function, set the cross-validation parameter cv to 5, and use accuracy as the scoring function; Step S42: Use the fit function to perform a network search on the divided dataset, and use the best_params_ function to obtain the optimal parameters; Step S43: Finally, retrain the SVM model using the optimal parameters.

[0012] To better implement the present invention, further, in step S5, use the joblib library to serialize the SVM model trained in step S4, and save the joblib file locally; then, load the trained SVM model, and use the predict function to predict new data in real time and output the actual classification result.

[0013] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above method is implemented.

[0014] An electronic device, including a memory and a processor; a computer program is stored on the memory; the processor is configured to execute the computer program in the memory to implement the above method.

[0015] The beneficial effects of the present invention are as follows: Aiming at the problem of difficult group classification of aviation parts, the present invention realizes the classification of aviation parts by combining the semantic understanding of the LLM model and the classification ability of SVM. The present invention applies the support vector machine algorithm to the group classification recognition of aviation parts. For aviation parts with complex structures and diverse attributes, by combining the classification ability of SVM and multi-attribute feature data, the accuracy and efficiency of classification are improved, and it has good practicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the multi-attribute aviation part classification and recognition method based on SVM of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Example 1: A multi-attribute aviation part classification and recognition method based on SVM, as Figure 1 shown, includes the following steps: Step 1: Data preparation and cleaning.

[0018] By parsing the geometric attribute information and non-geometric attribute information of the 3D model, collect the attribute information of the parts and accurately label their categories. Then, use the industry-specific dictionary and perform operations such as stop word removal and word segmentation on the data. Specifically, the attribute information of the parts includes but is not limited to the drawing number, name, material name, material grade, blank size, part size, processing notes, etc. of the parts.

[0019] Step 2: Data encoding preprocessing and feature fusion.

[0020] In SVM classification, data preprocessing is a crucial step to ensure the performance of the model.

[0021] For text information such as names and material names, a large language model (LLM) is used to encode descriptive texts such as text attribute information. The attribute names and attribute contents are combined with each other and converted into high-dimensional vectors to capture the deep semantic information in the description content, facilitating the understanding of the SVM model. Finally, the vectors of each attribute are directly concatenated to form a multi-attribute feature vector, realizing the comprehensive feature expression of parts by combining text information and numerical information.

[0022] Step 3: SVM model construction and model training.

[0023] Step 4: Cross-validation and grid search.

[0024] The dataset formed in Step 2 is split into a test set and a training set to perform cross-validation on the data and ensure the generalization ability of the model. Then, the parameter space is systematically traversed to find the best model parameters to improve the classification accuracy.

[0025] Step 5: Predict new data. The trained model is used to predict new data.

[0026] The present invention applies the support vector machine algorithm to the group classification and recognition of aviation parts. For aviation parts with complex structures and diverse attributes, by combining the classification ability of SVM and multi-attribute feature data, the classification of aviation parts is realized, and the classification accuracy and efficiency are improved.

[0027] Example 2: A multi-attribute aviation part classification and recognition method based on SVM. In this example, the Python language is used to identify part types using multiple attributes of parts. Parts of the same type have a unified process flow and standard operations, thereby realizing the standardization of process specifications, improving production efficiency and assembly quality, and reducing production costs. The specific steps are as follows: Step 1: Data preparation and feature extraction. In this example, the three-dimensional CAD secondary development interface is used to parse the attribute information and geometric shape information of the nodes in the part model. As shown in Table 1, the following data is obtained and the types are marked: Non-geometric attributes: drawing number, name, material name, material grade, blank size, thickness, etc.; Geometric attributes: part size, volume, center point, surface area, center of gravity, etc.

[0028] Table 1 For text data, in combination with an aviation dictionary, use the jieba library to clean the data, remove stop words, punctuation marks, convert English characters to lowercase, and split the remaining strings into smaller units.

[0029] Step 2: Preprocess the above data.

[0030] Use the open-source dataset BERT-wwm dataset, combine the cleaned data with the attribute names, and input them into the pre-trained LLM model to generate vectorized representations. Specifically, for text such as the material name combined with material name 1 as: "The material name of the part is material name 1", convert the string input into a 128-dimensional vector. For numerical values such as the blank size combined with 300x200 as "The blank size is 300x200", convert the string input into a 128-dimensional vector. The dimension can be selected arbitrarily, and 128 dimensions are only used as an example for the current scenario.

[0031] Finally, use the sum function of the numpy library to add the processed feature vectors together, prepare for the training of the SVM model, and ensure the consistency of the data and the effectiveness of the model.

[0032] Step 3: Construct the SVM model. Since the relationship between part attribute values belongs to a complex non-linear relationship, the radial basis function (RBF) is selected as the kernel function of the support vector machine to map the data to an infinite-dimensional space and achieve linear separation of the data in the high-dimensional space. In addition, since there are more than 2 part categories, the one-vs-rest strategy is adopted. For k categories, train k SVM classifiers.

[0033] The specific operations are as follows: First, divide the dataset into a training set and a test set, and use the train_test_split function to divide an 80% training set and a 20% test set.

[0034] Then, use the SVC class of the scikit-learn library to initialize the SVM model. By setting the kernel to "rbf", select the radial basis function as the kernel function. Then set the model parameters: the penalty parameter C and the gamma parameter.

[0035] Finally, call the fit method to complete the model training.

[0036] Step 4: Cross-validation and grid search. First, perform cross-validation on the model. Divide the dataset in Step 3 into 5 subsets, and each subset has the opportunity to be used as a training set and a test set. Use one fold as the test set and the remaining 4 folds as the training set to train the model. Repeat this 5 times for the 5 subsets. Then, perform a grid search on the model to find the best parameter combination and find a better parameter combination.

[0037] The specific operations are as follows: First, use the GridSearchCV function to set cv to 5 (set the cross-validation parameter) and adopt accuracy as the scoring function.

[0038] Then, use the fit function to perform a network search on the divided dataset and use the best_params_ function to obtain the best parameters. Finally, use the best parameters to retrain the model, and finally complete the training of the example SVM model.

[0039] Step 5: Predict new data. Serialize the trained model in Step 4 using the joblib library, save the joblib file locally to avoid repeated training. Then load the trained model and use the predict function to predict new data in real time and output the final actual classification result.

[0040] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Any simple modifications and equivalent changes made to the above embodiments based on the technical essence of the present invention all fall within the protection scope of the present invention.

Claims

1. A multi-attribute aviation parts classification and recognition method based on SVM, characterized in that: The following steps are involved: Step S1: Acquire data and perform preprocessing; obtain attribute information of parts based on 3D model analysis and mark their types, and then clean the data; Step S2: data encoding preprocessing and feature fusion; Use a large language model to process the attribute information to obtain a multi-attribute feature vector, and use it as a data set, then split the data set into a test set and a training set; Step S3: construct an SVM model and train the SVM model based on the training set; select the radial basis function as the kernel function of the support vector machine, map the data to an infinite dimensional space, and realize linear segmentation of the data in the high dimensional space; Step S4: Cross-validation and grid search: cross-validate the SVM model based on the test set, and then systematically traverse the parameter space to find the best model parameters; Step S5: Use the SVM model in step S4 to make predictions on the new data.

2. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 is characterized in that: In step S1, the text data is cleaned using the Jieba library in combination with an aviation dictionary to remove stop words, punctuation marks, convert English characters to lowercase, and perform word segmentation on the remaining character strings.

3. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 or 2 is characterized in that: In step S1, the attribute information includes non-geometric attributes and geometric attributes of the part, and the non-geometric attributes of the part include any one or more of the drawing number, name, material name, material grade, blank size, and thickness; the geometric attributes of the part include any one or more of the part size, volume, center point, surface area, and center of gravity.

4. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 is characterized in that: In step S2, based on a large language model, the descriptive text in the attribute information is encoded, and the attribute name and attribute content are combined and converted into a high-dimensional vector to capture the deep semantic information in the description content; finally, the feature vectors of each attribute are directly spliced ​​to form a multi-attribute feature vector, and the text information and numerical information are combined to express the comprehensive features of the parts.

5. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 4 is characterized in that: In step S2, finally, the feature vectors of each attribute are added together using the sum function of the numpy library to form a multi-attribute feature vector.

6. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 is characterized in that: The step S3 comprises the following steps: Step S31: Initialize the SVM model using the SVC class of the scikit-learn library; Step S32: by setting kernel to "rbf", select the radial basis function as the kernel function; then set the penalty parameter C and gamma parameter of the SVM model; Step S33: Finally, call the fit method to complete the training of the SVM model.

7. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 is characterized in that: The step S4 comprises the following steps: Step S41: First, use the GridSearchCV function, set the cross-validation parameter cv to 5, and use accuracy as the scoring function; Step S42: Use the fit function to perform a network search on the divided data set, and use the best_params_ function to obtain the best parameters; Step S43: Finally, the SVM model is retrained using the optimal parameters.

8. The multi-attribute aviation parts classification and recognition method based on SVM according to claim 1 is characterized in that: In step S5, the SVM model trained in step S4 is serialized using the joblib library, and the joblib file is saved locally; then, the trained SVM model is loaded, and the predict function is used to predict the new data in real time, and the actual classification result is output.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method described in any one of claims 1 to 8 is implemented.

10. An electronic device, characterized in that: The invention comprises a memory and a processor; the memory stores a computer program; the processor is used to execute the computer program in the memory to implement the method according to any one of claims 1 to 8.

Images