Failure Prognosis for electro-mechanical actuators based on improved SMO-SVR method

This paper deals with the problem of fault prediction and failure prognosis for electro-mechanical actuators. Firstly, traditional C-C method to reconstruct phase space of time series is developed and searching radius is expanded to determine the optimal delay time and embedded dimension. Then support vector regression (SVR) is utilized on fault prediction and failure prognosis. With the large size of sample data taken into account, the improved sequential minimal optimization (ISMO) algorithm is employed to solve the SVR model. And to raise the training speed, the duality gap ratio is introduced to improve the stopping criteria for iterations. The highlight is that real experimental data from NASA Prognostics Center are used to train and test the prediction model. Finally comparing with the BP network method, the simulation result demonstrates that the ISMO-SVR method has characteristics of high prediction accuracy and time efficiency, all of which will help take preventive measures before failure occurs.