Fault Diagnosis For Vehicle ESP System Based On Kernel Perceptron Algorithm

With the development of economy, cars have entered ordinary families. Besideseliminate price, appearance, power and other important factors, another important factorwhich cannot be ignored is safety. It’s natural for automotive safety to be paid wideimportance since it’s closely related to the drivers’ life. Nowadays automotive safetytechnology has developed very perfect, including the newly-developed stability controlsystems, among which ESP is the most powerful. ESP is not only had anti-lock braking anddriver anti-skid function, but also in times of emergency imposed different braking force oneach wheel to modified car over steer or under steer, thus ensured the safety of car driving.The system can also get the instant messages of automobiles and analyze the data, throughwhich interfere the automobiles and maintain the sustainability of driving. Different fromABS and TCS, ESP is an active safety control system with highly research value, whichmade it more complex and the risk of failure is increasing. Therefore how to maintain thesustainability of complex systems and make judges immediately are becoming newpropositions.Malfunction judging systems have a history of decades, while the traditional measuresare usually based on models. However, these measures require high precision of models.The car model is not only complex, but is non-linear dynamic model, it is great difficulty tobe modeled, thus this paper adopts another judging measure, namely the judging measurebased on data. In this paper makes an in-depth analysis about the structure and workingprinciple of ESP, and uses the AMESim software to build the vehicle system model base onthe analysis. Then we analysis the fault type of ESP, select a common ESP system wheelspeed sensor failures and actuator fault diagnosis goal. We get real-time detection aboutcar’s key operational parameters and the data by the simulation of the car model. Throughthe study and attempt, we find that the perceptron algorithm has good self-study abilities,and it’s relatively easier to practice. However, this measure is only suitable to linear data. Asfor nonlinear data, it’s less effective. To solve this problem this paper introduces kernelfunction, and gets the suitable method kernel perceptron algorithm, in the text, first byintroducing two categories perception algorithm lead multi-classification perception algorithm, then a detailed description of the kernel perceptron algorithm, At last, the papergets better judgments by adopting kernel function and using kernel perceptron algorithm.Meanwhile, the paper compares the researches using the two measures.