Wear condition prediction of hydraulic pump

Wear is a typical progressive failure of aero hydraulic pump. It is difficult to measure wear loss. To solve precision wear condition prediction problem, multi-dimensional support vector machine (SVM) prediction method was proposed, based on theoretical basis of SVM applied to time series prediction, multi-dimensional data decomposition and phase space reconstruction. The inner relationship of time series can be mined and reflected more effectively by this method. Oil-return flow was chosen to reflect the wear condition of hydraulic pump and was decomposed into trend data and random data. Multi-dimensional SVM was applied to predict oil-return flow of the aero hydraulic pump one-step ahead and multi-step ahead with grid search optimization method. The results show that multi-dimensional SVM model has higher prediction precision and is very suitable for long-term forecasting compared with the predicted results of traditional SVM.