Engine wear fault diagnosis based on improved semi-supervised fuzzy c-means clustering

A improved semi-supervised fuzzy c-means clustering algorithm (ISS-FCM) is proposed to diagnose engine wear faults with small oil samples. An optimized objective function, which is defined through introducing average distance measure between unlabeled samples and training samples with weighting values, is used to conduct the clustering process. To avoid local extrema originating from initialing partition matrix randomly, the training samples are utilized in partition matrix initialing work. By reason that engine wear condition can not be effectively characterized by original oil data with unobvious cluster trendency, Autoregression (AR) model is used to abstract the residual variance features from oil data. The atomic emission spectrometric oil data of engine bench test are analyzed with the proposed method. The cylinder scoring and bushing ablating faults are diagnosed successfully. Experimental results demonstrate the validity of the presented method in the field of engine wear fault diagnosis. © 2011 Journal of Mechanical Engineering.