A METHOD FOR MACHINE FAILURE DETECTION AND ISOLATION USING PERSONALIZED DIAGNOSTIC MODEL

摘要

Diagnostic models are often designed based on physics, known as a first-principles model, or using expert knowledge derived from a set of similar machinery, termed a "fleet-based" model. Diagnostic models measure and compare sensor change patterns with known machinery "failure signatures." By analyzing the similarities among various failure signatures and the actual data trends of a specific machine, an application can detect, characterize and diagnose the root causes of anomalous behavior. These so-called, fleet-based diagnostic models perform adequately when the analyzed machine performs close to the overall set's common behavior. However, when individual machines vary substantially from the average, a fleet-based diagnostic model behaves increasingly poorly, providing inaccurate results, and driving up both false positives and false negatives. This paper presents an approach to dynamically personalize a signature-based diagnostic model based on individual machine data characteristics, and then demonstrates the effectiveness of this technique for improving signature sensitivity and false alarm reduction in an aircraft engine diagnostic application.