Fault Prediction via Symptom Pattern Extraction Using the Discretized State Vectors of Multi-Sensor Signals

Abstract

Fault prediction and early degradation detection have received considerable attention in many engineering disciplines. Fault symptoms can be identified by abnormal values or unusual trends in the monitored sensor signals over a certain period prior to fault occurrence. However, how to extract abnormal pattern, particularly those with conditional relations among multiple sensor signals, remains unclear. Pattern extraction is further difficult particularly when there is no gradient relationship between measurements and operational states due to highly scattered data and unclear boundaries for distinguishing operational states. Additionally, defining the time period for symptom periods is challenging. To resolve these issues, we define the terms symptom pattern and symptom period, and then present a symptom pattern extraction method that collects all evidences of potential fault occurrence from multiple sensor signals. We postulate that, given time markers of fault occurrences, a symptom period precedes the occurrence of a fault. Symptom patterns are defined as either only found in the symptom periods or not found in the given time series, but similar to fault patterns. We further discuss an iterative search procedure for determining the length of symptom periods and propose a severity assessment method for symptom patterns. Finally, we apply the symptom pattern extraction and severity assessment methods to an online fault prediction procedure. By assessing the total severity of patterns in the monitoring window, early warning decision can be made. The procedure is tested in the early detection of abnormal cylinder temperature in a marine diesel engine and automotive gasoline engine knocking.