The fault detection of multi-sensor based on multi-scale PCA

Multi-Scale Principal Component Analysis (MSPCA) for sensor fault detection is discussed to resolve the problem that the traditional MSPCA can´t realize the comprehensive sensor fault detection. MSPCA combines the decorrelation ability of PCA for the linear variables with the ability of wavelet analysis to extract deterministic features and approximately decomposition correlation of variable. MSPCA computes wavelet coefficients of the PCA at each scale and then combines the results at relevant scales. Due to its multi-scale properties, MSPCA is appropriate for the data modeling along with the time and frequency changes. The superior performance of MSPCA for process fault monitoring is illustrated by simulation results.