Data-driven Based Torque-lost Fault Diagnosis For In-wheel Motor Vehicle

With the gradual rise of electric vehicles,the development of wheel-drive vehicles is becoming faster and will surely become the future of electric vehicles.Four-wheel motor-drive vehicles have revolutionized the existing vehicle architecture and brought new chassis implementations.There is no more need to place a large motor in the space,nor the mechanical structure of the transmission.The development and application of new technology has opportunities and challenges.The introduction of the hub motor not only puts high demands on the four-wheel coordinated control,the characteristics of the four-wheel independent drive also make the fault diagnosis technology of the wheel drive vehicles become an important research field.This paper mainly studies the motor fault diagnosis method for the vehicle controller.The traditional fault diagnosis is mostly based on the model.The theory of this part is mature and the technology is relatively perfect.However,some methods are under the premise of having an accurate model,but the vehicle is a nonlinear model.so it is very difficult to model the vehicle with high precision.Based on the above problems,this paper proposes a method for fault diagnosis by constructing a data model.Firstly,according to the analysis of local model modeling method,ARX model is selected as the modeling method of this paper,and JITL(just in time learning)is combined to complete the relevant theoretical derivation,so the local data model is obtained.The WLTC cycle condition was selected as the initial database source,and the residual fault-based fault diagnosis was studied and then realize the design of the support vector machine diagnostic tool.Aiming at the problem of low efficiency of database traversal query,this paper proposes a deduplication algorithm that reduces the size of the database by using fuzzy C-means clustering.According to the octree-theory,the hierarchical search algorithm of data is designed.The simulation verified the efficiency optimization algorithm and make the algorithm reach the real-time operation.Finally,the accuracy of the prediction algorithm and the fault diagnosis algorithm are simulated in different working conditions.The real-time performance of the algorithm is verified based on the dSPACE real-time simulation platform.The matrix storage method of the tree structure is proposed,and the database is re-sampled.It can be applied to hardware with different sampling periods while ensuring accuracy.This paper establishes an offline database and builds a data model based on the database.Then uses the residual vector-based support vector machine fault diagnosis method to complete the fault diagnosis of the hub motor without the vehicle model,and the hardware in the real-time control system On the ring platform,the method has certain engineering practical significance.