A multivariate statistical analysis technique for on-line fault prediction

This paper describes a generalized multivariate statistical analysis technique for prediction of impending failures in electronic and electromechanical equipment. This data-driven prognostic technique is useful in health-monitoring situations where equipment physical models are unavailable or of limited fidelity. Statistical analysis algorithms, integrated into a predictive fault detection (PFD) statistical analysis engine, operate on heterogeneous streams of data from sensors that monitor selected equipment structural and functional parameters. The statistical analysis engine processes input data in two stages - similarity testing followed by trend determination. The input stage algorithm extracts multidimensional feature data samples from the arriving sensor data streams. It then performs statistical comparisons of the feature data samples to corresponding feature patterns from equipment considered to be in nominal operating condition. The nominal-condition data may be obtained from equipment specifications, or it may be derived in situ from the sensor data streams. The algorithm maintains a set of scalar similarity metrics for the equipment being monitored, which it periodically compares with pre-computed thresholds. The thresholds may be adaptive, with values typically functions of the amount and variability of the feature data. The trend determination stage is triggered when a threshold value is exceeded. The trend determination algorithm projects trends of feature data from the similarity analysis onto the future. The automatic data trending computation for a given feature is performed by statistically analyzing a window of the feature data. This analysis addresses a collection of different characteristics of a well-fitting trend, which are then fused into a single trend result per data window. The statistical analysis engine applies the trending results to determine the most probable trend, which in the PFD context is related to the requirements for schedulin- - g of equipment maintenance actions.