Fault Detection And Diagnosis Method Of Commercial Vehicle Operation Process Based On Multivariate Time Series Data

As an important part of the traditional manufacturing industry,commercial vehicle manufacturing integrates with the Internet technology and gives birth to the Internet of Vehicles.With the help of the Internet of Vehicles,commercial vehicle owners can be provided with more intelligent services.Once the commercial vehicle has a running fault,it will lead to running stagnation.If the fault of commercial vehicles can be detected and diagnosed in time,the occurrence of more heavier loss can be avoided.Massive multi-time-series data contained in the Internet of Vehicles can provide supports for fault detection and diagnosis in the operation of commercial vehicles.The traditional fault detection and diagnosis methods of commercial vehicles include artificial way and setting threshold to start the fault alarm device.There are some problems in these two ways,such as low efficiency,weak accuracy and difficult to locate the cause of fault.Therefore,it is necessary to make use of the relevant algorithm to detect and diagnose faults in the running process of commercial vehicles.Existing fault detection algorithms for time series data can be roughly divided into two categories:one is fault detection based on traditional machine learning algorithm,and the other is fault detection based on deep learning algorithm.Fault detection based on traditional machine learning algorithms include fault detection algorithms based on probability statistics,distance and isolation,etc.This kind of method requires the data to meet certain characteristic rules.In addition,commonly used fault detection algorithms based on classification,such as Naive Bayes,SVM and so on,are difficult to obtain high detection accuracy under the condition of unbalanced positive and negative samples.The fault detection method based on deep learning mainly determines the fault by calculating the prediction error.The algorithms include LSTM,GRU,etc.Most of them only consider the time dependence characteristics of time series data and ignore the correlation characteristics of time series data.In addition,existing algorithms are more about making judgment on the fault of time series data than diagnosing the cause of the fault.The time series data generated by commercial vehicles during operation are stochastic,and it is difficult to obtain characteristic rules.There are complex correlations between the data and have the problem of imbalance of positive and negative samples.In view of these problems,the existing methods are not suitable for the fault detection and diagnosis of commercial vehicles.Based on that,this paper studies the fault detection and diagnosis methods of commercial vehicle running process based on multivariate time series data.The problems to be solved in this research are as follows:first,how to extract the characteristic information of multiple time series from the operation data of commercial vehicles to effectively detect faults;second,how to ensure the accuracy of fault detection under the condition of unbalanced positive and negative samples of Internet of Vehicles data;third,how to identify the important factors affecting the occurrence of the fault to accurately locate the cause of the fault.In view of the above problems,this paper has done in-depth research and discussion.A fault detection and diagnosis model FDD_MTS（Fault Detection and Diagnosis Model Based on Multivariate Time Series Data）based on multivariate time series data which can solve these problems is proposed.FDD_MTS mainly includes the following three modules:The first module is the extraction of correlation features among multivariate time series data.By using CNN’s convolutional layer,the correlation characteristics of multivariate time series data are obtained.The second module is the time dependent feature extraction of multivariate time series data.The time dependent characteristics are extracted by using Transformer Encoder.The third module is fault detection and diagnosis of multivariate time series data.The GAN network based on generation model is used to detect and diagnose the fault of data.The main works and contributions of this paper are summarized as following:1.This paper proposes a feature extraction scheme for multivariate time series data of commercial vehicles.In this scheme,CNN is used to extract the correlation features of multivariate time series data,and Transformer Encoder is used to extract the time dependent features of time series data.There are complex correlations between the components of commercial vehicles in operation,and the data change constantly over time.By fully extracting the feature information of multivariate time series data,including time dependent features and correlation features,the model can improve the accuracy of fault detection.2.In this paper,the GAN network based on the generation model is adopted to solve the problem of the imbalance of positive and negative samples of Internet of Vehicles data.This paper uses only normal data to train the model,and the difference between the discriminant score of the data and the reconstruction error of the data of the generator is used as the basis for the faulty judgment.At the same time,the difference of each dimension between the input data and the reconstructed data is compared,and the dimension with a larger difference value is taken as the factor contributing more to the fault,so as to diagnose the fault.Compared with the fault detection algorithm based on classification,this algorithm can ensure the accuracy of fault detection under the condition of unbalanced positive and negative samples of Internet of Vehicles data.3.In this paper,the real data set of a large commercial vehicle manufacturing industry is used for experiments,to verify the validity of the FDD_MTS model.Precision,Recall,and F1 were used as evaluation metrics to evaluate the experimental results.The FDD_MTS model is compared with four baseline models for fault detection of time series data.Experiments have shown that FDD_MTS is more suitable for the fault detection of commercial vehicles.In addition,the validity of the model is further verified by modifying the model parameters and replacing the model components.Finally,the results of fault diagnosis by the model are presented.