Anomaly detection and reasoning with embedded physical model

Abstract

The anomaly detection/reasoning system has been viewed by many experts, in industry as the cornerstone towards maturing onboard health management systems for complex flight articles such as commercial aircraft, space vehicles, and military aircraft. The technical ingredients that are necessary for an effective anomaly system include: 1) models, particularly that which are based on physical principles; 2) detection algorithms; and 3) reasoning rules. The Boeing Company has been working with Scientific Monitoring, Inc. (SMI) to develop a physics-based modeling framework to improve the accuracy of data-centric anomaly detection algorithms. We developed strategies to derive such data-driven physical models, which are designated as the low cost physical models (LCPM) or the heuristic physical models (HPM). To prove that this modeling framework is applicable to all the physical systems or subsystems of an aerospace vehicle, we selected a space flight rocket engine to prove the concept. The engine represents a complex aero-thermal-mechanical system, and it offers a high degree of modeling difficulty. The engine also offers a rich data environment for validating the concept. This paper describes briefly the modeling approach, model fidelity, and the feasibility of this model-based anomaly detection approach.