A Computationally Efficient and Adaptive Approach for Online Embedded Machinery Diagnosis in Harsh Environments

Condition-based monitoring (CBM) has advanced to the stage where industry is now demanding machinery that possesses self-diagnosis ability. This need has spurred the CBM research to be applicable in more expanded areas over the past decades. There are two critical issues in implementing CBM in harsh environments using embedded systems: computational efficiency and adaptability. In this paper, a computationally efficient and adaptive approach including simple principal component analysis (SPCA) for feature dimensionality reduction and K-means clustering for classification is proposed for online embedded machinery diagnosis. Compared with the standard principal component analysis (PCA) and kernel principal component analysis (KPCA), SPCA is adaptive in nature and has lower algorithm complexity when dealing with a large amount of data. The effectiveness of the proposed approach is firstly validated using a standard rolling element bearing test dataset on a personal computer. It is then deployed on an embedded real-time controller and used to monitor a rotating shaft. It was found that the proposed approach scaled well, whereas the standard PCA-based approach broke down when data quantity increased to a certain level. Furthermore, the proposed approach achieved 90% accuracy when diagnosing an induced fault compared to 59% accuracy obtained using the standard PCA-based approach