A novel feature extraction algorithm for bearing fault diagnosis based on enhanced symbolic aggregate approximation

Abstract

This paper proposes a novel symbolic representation method of feature extraction for fault diagnosis and condition monitoring. The Symbolic Aggregate approXimation (SAX) technique basically transforming real-valued time series into symbol sequences, has been proven as a newly developed tool of feature extraction for fault diagnosis. However, the original SAX is based on the Piecewise Aggregate Approximation (PAA) representation, which is primarily transformed by the calculation of mean value of the equal sized data subsection for dimensionality reduction. Such mean value-based method has a high possibility of missing important information patterns in vibration signals. To overcome this limitation, an enhanced SAX (ESAX) is proposed to extract fault features well from vibration signals. The ESAX utilizes mean value together with two additional important points (the max and min points) to conduct the symbol representation in PAA process. Next, the Shannon entropy technique is conducted on the symbol sequences generated by ESAX to extract features for classification tasks. Compared with SAX, the ESAX can extract more comprehensive signal characteristics considering important information points and local fault patterns. The effectiveness and superiority of the ESAX were validated by experimental studies using the fault signals of rolling bearings.