Data-Driven Fault Detection Based on Process Monitoring using Dimension Reduction Techniques

One goal of integrated vehicle health management (IVHM) for commercial airplane customers is to monitor sensor data to anticipate problems before flight deck effects (FDEs) ground the airplane for unplanned maintenance. Airplane subsystems - such as flight and environmental control systems, electrical and hydraulic power - can have a high number of associated parameters. Monitoring sensor data streams individually can be inefficient, and fail to detect problems. A research effort at The Boeing Company is investigating anomaly detection algorithms for multivariate time series of parametric data. The multivariate process monitoring techniques account for correlation between parameters, and therefore alert when relationships between parameters change, as well as when mean levels of individual parameters change. Since many traditional multivariate process monitoring techniques are not suited for the high number of parameters in airplane subsystems, this paper discusses using dimension reduction techniques. One example is principal component analysis (PCA). If the assumptions behind PCA are not met, then monitoring charts based on conventional PCA alone can show false alarms and bad detectability. Independent component analysis (ICA) is a recently developed method in which the goal is to decompose observed data into linear combinations of statistically independent components. ICA can be considered an extension of PCA since it uses PCA as an initial pre-whitening stage. Like projection pursuit density estimation, ICA searches for projections of the data that are most non-Gaussian.