Defect state and severity analysis using discretized state vectors

Abstract

The time series of sensor data for condition monitoring of a system is often characterized as very-short, intermittent, transient, highly nonlinear and non-stationary random signals, which hinder the straightforward pattern analysis. In order to identify meaningful features in measured sensor data, we transform the continuous time series into a set of contiguous discretized state vectors using a multivariate discretization approach. We then search for important patterns that are only found in defective systems. We discuss how to measure the severity degree of each defect pattern and assess the criticality of a defective state. We consider a defective state to be more severe if various defect patterns are observed in the state. Similarly, if a particular defect pattern describes multiple defect states, the pattern is treated as significant. The proposed procedure is utilized to detect defective car door trims that generate small but irritating noises. We analyzed the datasets obtained from a typical acoustic sensor array and acoustic emission sensors. The defective door trims were efficiently identified including the severity degrees of the identified patterns.