A vehicle fault diagnosis model training method and device adapted to cloud server

Abstract

The present invention discloses a vehicle fault diagnosis model training method and device adapted to a cloud server, wherein the method comprises the following steps: collecting data with vehicle fault category labels; uploading the data with vehicle fault category labels to a cloud server; preprocessing the data with vehicle fault category labels in the cloud server to obtain preprocessed fault data; extracting features from the preprocessed fault data using a variational autoencoder to obtain feature data; using the feature data as input of an improved support vector machine SVM to construct a fault diagnosis model combined with a variational autoencoder; wherein the improved support vector machine SVM is a one-to-many-to-one classifier. In the technical solution provided by the present invention, the use of this classifier has the advantages of being simpler, more widely applicable, more flexible and independent than the traditional SVM.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle fault diagnosis, and in particular to a vehicle fault diagnosis model training method and device adapted to a cloud server. Background Art

[0002] With the rapid development of social economy and automobile industry, traditional vehicle fault diagnosis methods are increasingly unable to adapt to the increasingly complex fault environment. The traditional way of vehicle fault diagnosis is for maintenance personnel to manually inspect and repair the vehicle, that is, to understand the fault symptoms and problems of the vehicle, and to conduct external visible inspections of the vehicle, including observing whether there is obvious damage, oil leakage, wear or corrosion on the vehicle's appearance, reading and analyzing recorded data to understand the state of the vehicle when the fault occurs, and further determine the possible cause of the fault, and conduct a more detailed inspection of the components that may be involved in the fault. However, traditional vehicle fault diagnosis methods usually rely on experienced maintenance personnel to perform manual inspections and judgments, which are not effective and accurate enough for complex faults and the interaction of multiple systems.

[0003] In recent years, many intelligent vehicle fault diagnosis methods have emerged. Using vehicle fault diagnosis models to diagnose vehicle faults can be more effective and accurate. With the further development and progress of technology, the industry continues to improve vehicle fault diagnosis models to improve the performance of vehicle fault diagnosis and meet user needs. Summary of the invention

[0004] The purpose of the present invention is to provide a vehicle fault diagnosis model training method and device adapted to a cloud server to improve the performance of vehicle fault diagnosis, thereby meeting the user's usage needs.

[0005] To achieve the purpose of the present invention, the technical solution provided by the present invention is as follows:

[0006] First aspect

[0007] The present application provides a vehicle fault diagnosis model training method adapted to a cloud server, comprising the following steps:

[0008] Step S1: Collect data with vehicle fault category labels;

[0009] Step S2: uploading the data with the vehicle fault category label to the cloud server;

[0010] Step S3: preprocessing the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data;

[0011] Step S4: extracting features from the preprocessed fault data using a variational autoencoder to obtain feature data;

[0012] Step S5: using the feature data as input of an improved support vector machine (SVM) to construct a fault diagnosis model combined with a variational autoencoder;

[0013] The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows:

[0014] In the first stage, K one-vs-all SVM classifiers are trained as follows:

[0015] Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class;

[0016] In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows:

[0017] Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

[0018] Second aspect

[0019] The present application provides a vehicle fault diagnosis model training device adapted to a cloud server, comprising the following units: a data acquisition unit, a data upload unit, a data preprocessing unit, a feature data extraction unit, and a fault diagnosis model construction unit;

[0020] The data collection unit is used to collect data with vehicle fault category labels;

[0021] The data uploading unit is used to upload the data with the vehicle fault category label to the cloud server;

[0022] The data preprocessing unit is used to preprocess the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data;

[0023] The feature data extraction unit is used to extract features from the preprocessed fault data using a variational autoencoder to obtain feature data;

[0024] The fault diagnosis model building unit is used to use the feature data as input of an improved support vector machine (SVM) to build a fault diagnosis model combined with a variational autoencoder;

[0025] The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows:

[0026] In the first stage, K one-vs-all SVM classifiers are trained as follows:

[0027] Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class;

[0028] In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows:

[0029] Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

[0030] Compared with the prior art, the technical advantages of the present invention are as follows:

[0031] In the technical solution provided by the present invention, the obtained fault diagnosis model combined with the variational autoencoder includes an improved support vector machine SVM, and the improved support vector machine SVM is a new type of one-to-many-to-one classifier. Compared with the traditional SVM, the use of this classifier has the advantages of being simpler, more widely applicable, more flexible and independent, etc.

[0032] In addition, in the technical solution provided by the present invention, the data with vehicle fault category labels are preprocessed, including data cleaning, processing missing values, processing outliers, and segmenting vehicle fault categories and removing isolated noise points. The fault data after preprocessing has higher accuracy and completeness, so that the prediction effect of the subsequent constructed fault diagnosis model is better;

[0033] In addition, in the technical solution provided by the present invention, the powerful computing and storage capabilities of the cloud server are utilized to process data, which can save local computing resources and reduce maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a method flow chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] like Figure 1 As shown, this embodiment provides a vehicle fault diagnosis model training method adapted to a cloud server, comprising the following steps:

[0037] Step S1: Collect data with vehicle fault category labels;

[0038] Step S2: uploading the data with the vehicle fault category label to the cloud server;

[0039] It should be noted that the cloud server may be an Alibaba Cloud cloud server.

[0040] Step S3: preprocessing the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data;

[0041] In step S3, the data with the vehicle fault category label is preprocessed in the cloud server, specifically including the following:

[0042] Step S3.1: performing data cleaning, processing missing values, processing abnormal values ​​on the data with vehicle fault category labels, and obtaining fault data after preliminary preprocessing;

[0043] It should be noted that the data cleaning is to remove duplicate data, check whether there are duplicate data records in the data set, and delete them; the processing of missing values ​​is to analyze the missing values ​​in the data. According to the missing value situation, you can choose to fill the missing values ​​(such as mean filling, median filling or interpolation method), or delete the data records containing missing values ​​for processing; the processing of outliers is to detect and process outliers, usually using statistical methods such as the 3σ principle.

[0044] Step S3.2: segment the fault data after preliminary preprocessing into vehicle fault categories and remove isolated noise points to obtain preprocessed fault data.

[0045] In step S3.2, Euclidean clustering is used to segment and denoise the vehicle fault categories, which specifically includes the following:

[0046] Step S3.2.1: Set the distance threshold of Euclidean clustering to t, and set the minimum number of vehicle failures in a single cluster to N;

[0047] Step S3.2.2: Clustering is performed according to the distance threshold t, and all clusters with the number of vehicle faults less than N are removed.

[0048] Step S4: extracting features from the preprocessed fault data using a variational autoencoder to obtain feature data;

[0049] It should be noted that the variational autoencoder includes two parts: an encoder network and a decoder network. The encoder network converts the original input samples of the network into hidden layer vector representations, and the decoder network reconstructs the original input samples based on the hidden layer vector representations.

[0050] In the actual training process, the present invention uses the mean square error to measure the similarity between the original input sample and the reconstructed original sample. The objective function of the variational autoencoder is:

[0051] ;

[0052] in, is the mean square error, is the sample size, is the sample serial number, For the i samples, For the i The output corresponding to each sample.

[0053] Step S5: using the feature data as input of an improved support vector machine (SVM) to construct a fault diagnosis model combined with a variational autoencoder;

[0054] The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows:

[0055] In the first stage, K one-vs-all SVM classifiers are trained as follows:

[0056] Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class;

[0057] In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows:

[0058] Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

[0059] Usage: When using a one-to-many-to-one classifier, for the feature data to be classified, first use the one-to-many classifier to obtain the confidence of different categories. For the top T categories ranked from high to low according to confidence, select the corresponding T(T-1) / 2 one-to-one classifiers from the one-to-one classifier, and determine the final classification (i.e. the result of fault diagnosis) according to the voting mechanism based on the results of the one-to-one classifier.

[0060] Accordingly, the present embodiment provides a vehicle fault diagnosis model training device adapted to a cloud server, characterized in that it comprises the following units: a data acquisition unit, a data uploading unit and a data preprocessing unit, a feature data extraction unit, and a fault diagnosis model building unit;

[0061] The data collection unit is used to collect data with vehicle fault category labels;

[0062] The data uploading unit is used to upload the data with the vehicle fault category label to the cloud server;

[0063] The data preprocessing unit is used to preprocess the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data;

[0064] The feature data extraction unit is used to extract features from the preprocessed fault data using a variational autoencoder to obtain feature data;

[0065] The fault diagnosis model building unit is used to use the feature data as input of an improved support vector machine (SVM) to build a fault diagnosis model combined with a variational autoencoder;

[0066] The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows:

[0067] In the first stage, K one-vs-all SVM classifiers are trained as follows:

[0068] Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class;

[0069] In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows:

[0070] Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

[0071] The preprocessing of the data with the vehicle fault category label in the cloud server specifically includes the following steps:

[0072] Step S3.1: performing data cleaning, processing missing values, processing abnormal values ​​on the data with vehicle fault category labels, and obtaining fault data after preliminary preprocessing;

[0073] Step S3.2: segment the fault data after preliminary preprocessing into vehicle fault categories and remove isolated noise points to obtain preprocessed fault data.

[0074] Among them, Euclidean clustering is used to segment and denoise the vehicle fault categories, including the following:

[0075] Step S3.2.1: Set the distance threshold of Euclidean clustering to t, and set the minimum number of vehicle failures in a single cluster to N;

[0076] Step S3.2.2: Clustering is performed according to the distance threshold t, and all clusters with the number of vehicle faults less than N are removed.

[0077] Among them, the objective function of the variational autoencoder is as follows:

[0078] ;

[0079] in, is the mean square error, is the sample size, is the sample serial number, For the i samples, For the i The output corresponding to each sample.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate and explain the present invention, and are not intended to limit the present invention to the scope of the described embodiments. In addition, those skilled in the art can understand that the present invention is not limited to the above embodiments, and more variations and modifications can be made according to the teachings of the present invention, and these variations and modifications all fall within the scope of the protection claimed by the present invention.

Claims

1. A vehicle fault diagnosis model training method adapted to a cloud server, characterized in that: The steps include: Step S1: Collect data with vehicle fault category labels; Step S2: uploading the data with the vehicle fault category label to the cloud server; Step S3: preprocessing the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data; Step S4: extracting features from the preprocessed fault data using a variational autoencoder to obtain feature data; Step S5: using the feature data as input of an improved support vector machine (SVM) to construct a fault diagnosis model combined with a variational autoencoder; The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows: In the first stage, K one-vs-all SVM classifiers are trained as follows: Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class; In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows: Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

2. A vehicle fault diagnosis model training method adapted to a cloud server according to claim 1, characterized in that: In step S3, the data with the vehicle fault category label is preprocessed in the cloud server, specifically including the following: Step S3.1: performing data cleaning, processing missing values, processing abnormal values ​​on the data with vehicle fault category labels, and obtaining fault data after preliminary preprocessing; Step S3.2: segment the fault data after preliminary preprocessing into vehicle fault categories and remove isolated noise points to obtain preprocessed fault data.

3. A vehicle fault diagnosis model training method adapted to a cloud server according to claim 2, characterized in that: In step S3.2, Euclidean clustering is used to segment and denoise the vehicle fault categories, which specifically includes the following: Step S3.2.1: Set the distance threshold of Euclidean clustering to t, and set the minimum number of vehicle failures in a single cluster to N; Step S3.2.2: Clustering is performed according to the distance threshold t, and all clusters with the number of vehicle faults less than N are removed.

4. A vehicle fault diagnosis model training method adapted to a cloud server according to claim 3, characterized in that: In step S4, the objective function of the variational autoencoder is as follows: ； in, is the mean square error, is the sample size, is the sample serial number, For the i samples, For the i The output corresponding to each sample.

5. A vehicle fault diagnosis model training device adapted to a cloud server, characterized in that: It includes the following units: data acquisition unit, data upload unit, data preprocessing unit, feature data extraction unit, and fault diagnosis model building unit; The data collection unit is used to collect data with vehicle fault category labels; The data uploading unit is used to upload the data with the vehicle fault category label to the cloud server; The data preprocessing unit is used to preprocess the data with the vehicle fault category label in the cloud server to obtain preprocessed fault data; The feature data extraction unit is used to extract features from the preprocessed fault data using a variational autoencoder to obtain feature data; The fault diagnosis model building unit is used to use the feature data as input of an improved support vector machine (SVM) to build a fault diagnosis model combined with a variational autoencoder; The improved support vector machine SVM is a one-to-many-to-one classifier, and its construction method is as follows: In the first stage, K one-vs-all SVM classifiers are trained as follows: Among the K vehicle fault categories, the following processing is performed in sequence: for each vehicle fault category, the vehicle fault category is set as the positive class, and the other K-1 vehicle fault categories are set as a negative class to form a binary classification problem; for each binary classification problem, a one-to-many SVM classifier is trained, and the output of the classifier is the confidence that the test sample is a positive class; In the second stage, K(K-1) / 2 one-to-one classifiers are trained as follows: Among the K vehicle fault categories, for each pair of vehicle fault categories, a one-to-one SVM classifier is constructed to distinguish the pair of vehicle fault categories.

6. A vehicle fault diagnosis model training device adapted to a cloud server according to claim 5, characterized in that: The preprocessing of the data with the vehicle fault category label in the cloud server specifically includes the following: Step S3.1: performing data cleaning, processing missing values, processing abnormal values ​​on the data with vehicle fault category labels, and obtaining fault data after preliminary preprocessing; Step S3.2: segment the fault data after preliminary preprocessing into vehicle fault categories and remove isolated noise points to obtain preprocessed fault data.

7. A vehicle fault diagnosis model training device adapted to a cloud server according to claim 6, characterized in that: Euclidean clustering is used to segment and denoise vehicle fault categories, including the following: Step S3.2.1: Set the distance threshold of Euclidean clustering to t, and set the minimum number of vehicle failures in a single cluster to N; Step S3.2.2: Clustering is performed according to the distance threshold t, and all clusters with the number of vehicle faults less than N are removed.

8. A vehicle fault diagnosis model training device adapted to a cloud server according to claim 7, characterized in that: The objective function of the variational autoencoder is as follows: ； in, is the mean square error, is the sample size, is the sample serial number, For the i samples, For the i The output corresponding to each sample.

Images