Machine condition monitoring using principal component representations

The purpose of this paper is to find the low-dimensional principal component (PC) representations from the statistical features of the measured signals to characterize and hence, monitor machine conditions. The PC representations can be automatically extracted using the principal component analysis (PCA) technique from the time- and frequency-domains statistical features of the measured signals. First, a mean correlation rule is proposed to evaluate the capability of each of the PCs in characterizing machine conditions and to select the most representative PCs to classify machine fault patterns. Then a procedure that uses the low-dimensional PC representations for machine condition monitoring is proposed. The experimental results from an internal-combustion engine sound analysis and an automobile gearbox vibration analysis show that the proposed method is effective for machine condition monitoring.