Fault Diagnosis Of Three Way Catalytic Converters In Vehicles Under High Dimensional Heterogeneous Data

Frequent three-way catalytic converter(TWCC)faults seriously affect the exhaust emission level of fuel vehicles and aggravate the environmental pollution caused by vehicles.The TWCC fault diagnosis model based on physicochemical principle is difficult to describe its fault conditions in complex and uncertain traffic environment because of the simplification of the model by using constraints,which leads to the insufficient generalization ability and low diagnostic accuracy of the fault diagnosis model.With the development of modern sensing and communication technology,the monitoring of automobile exhaust is more convenient and feasible.A large amount of exhaust emission data is generated,which provides a feasible way to study fault diagnosis of TWCC from the perspective of data-driven.Based on the vehicle exhaust data with fault information of TWCC this research aims at fault diagnosis theory and application for complex TWCC system by means of machine learning methods such as support vector machine(SVM)and neural networks.Specifically as follows:(1)Aiming at the feature extraction of high-dimensional and multi-scale faults,the sparse representation learning method is used to diagnose the faults of TWCC.A complete feature dictionary for sparse coding is constructed by using robust principal component analysis(RPCA).Sparse feature vectors of exhaust data are extracted by orthogonal matching pursuit(OMP)algorithm.Least squares support vector machine(LS-SVM)classification model driven by sparse fault feature vectors is constructed to realize multiclassification of TWCC faults.(2)Aiming at the problem of loss of fault information in the process of sparse fault features,a deep learning method is used to diagnose the fault of TWCC.A deep recursive neural network is constructed based on Huffman coding theory,and the exhaust gas data with different dimensions are used as the input of the recurrent neural network.Through the layer-by-layer statistical learning of fault state frequencies,a unified and refined description of fault characteristics of TWCC is realized in high-dimensional space.A deep recurrent neural network is constructed as a fault classifier.The fine characterization of the fault features is presented to the recurrent neural network.Driven by the cyclic neural network,the accurate classification and location of multi-class faults in TWCC are realized by supervised residual learning.The experimental results show that,driven by sparse features,the average accuracy of fault diagnosis method based on LS-SVM for five types of state fault diagnosis is 91.96%,while the accuracy of overheating aging fault diagnosis is lower,only 84.16%.Driven by refined features,the generalization performance of fault diagnosis method based on deep neural network is better than that of the former.The average accuracy of fault diagnosis for five kinds of states is 96.40%,and that for overheating aging is 98.00%.