Feature extraction for engine fault diagnosis utilizing the generalized S-transform and non-negative tensor factorization

In this study, a novel feature extraction scheme was proposed for engine fault diagnosis utilizing the generalized S-transform combined with the non-negative tensor factorization (NTF). To represent the information of the non-stationary vibration signals acquired from engine, the generalized S-transform was used to get a time-frequency distribution with enhanced energy concentration. Meanwhile, a newly developed technique called NTF, which can preserve more structure information hiding in original two-dimensional matrices compared to the non-negative matrix factorization (NMF), was adopted to extract more informative features from the time-frequency matrices. Five operating states of engine were tested in an experiment for evaluating the proposed feature extraction scheme. Four different types of learning algorithms were employed to conduct the fault classification task. The NMF technique was also used for feature extraction and compared with the NTF approach. The experimental results have demonstrated that the proposed feature extraction scheme can achieve a satisfactory performance when applied to diagnose the engine faults.