Incipient Fault Detection in a Hydraulic System Using Canonical Variable Analysis Combined with Adaptive Kernel Density Estimation

Incipient fault detection in a hydraulic system is challenge the condition monitoring community. Existing research mainly monitors abnormal working conditions systems by separately detecting key parameter, which often causes high miss warning rate for incipient faults due to oversight of parameter dependence. A principal component analysis provides an effective method taking correlation multiple parameters into consideration, but this technique assumes are Gaussian-distributed, making it invalid dynamic non-Gaussian system. In paper, we combine canonical variable (CVA) and adaptive kernel density estimation (AKDE) early nonlinear systems. The collected data set was used construct typical space, state space residual divided represent characteristics different correlations between two variables, quantitatively described using Hotelling’s T2 Q. order investigate proper upper control limits, AKDE utilised estimate underlying probability functions Q nonlinearity variables consideration. advantages proposed approach illustrated via marine power plant lubrication