关键词: 特征提取, 状态监测, 预测性维护, 变点检测

Abstract: Based on the four dimensions of thermal-pressure parameters， lubricant oil conditions， vibrations and cylinder pressures， data acquisition and feature extraction were carried out， and a method for diesel engine condition monitoring was proposed based on OCSVM anomaly detection algorithm and Fisher discriminant analysis.  The CUSUMMR was used for parameter trend detection and the D-S evidence theory and weight of evidence method were used for multi-source information fusion. the RUL(remaining useful life) of diesel engine lubricating oil was predicted by LSTM. The Paper solves the problems such as low early warning rate， poor adaptability of the model under diesel engine dynamic conditions. 

Key words: feature extraction, condition monitoring, predictive maintenance, change point detection