Research and application of manifold learning to fault diagnosis of reciprocating compressor

Since reciprocating compressor (RC) is a key facility for many industries, the study of its fault diagnosis is thus particularly important. This paper proposes a new method for predicting the fault degree of RC by using a manifold learning method. The main idea of the proposed method can be summarized as follows: first, employ a manifold learning algorithm to directly deal with RC´s cylinder pressure signals. Based on the obtained low-dimensional representation of the pressure signals, implement the diagnosis process by weighted interpolation procedure. The experiments conducted by some simulated data demonstrate that the proposed method performs satisfactorily and it therefore provides an effective way to diagnose the fault degree of RC.