Hybrid Machine Learning-Based Methodologies for Fault Diagnosis, Prognosis and Health Management

The in-service equipment (i.e., aircraft, vehicles, ships) currently faces new challenges of increasing operational flexibility, reducing operating costs, improving reliability and availability while mitigating the environmental impact. Traditional maintenance and health trend analysis of in-service equipment involve time-consuming, repetitive, and labor-intensive activities based on the usage time or failure. This paper aims to systematically develop and implement diagnosis, prognosis and health management (DPHM) solutions by taking advantages of the computational intelligence-based and physics model-based methodologies. The complete DPHM scheme is formalized by integrating strategies that involve nonlinear filtering algorithms, deep neural networks, supervised learning and unsupervised learning algorithms, and concepts of multiple model, among others. Various evaluations, verification and validation with respect to detectability, diagnostics, prognostics by using performance metrics are conducted. The proposed autonomous, intelligent and functional DPHM strategies represent as innovative solutions for fleet-wide, automated, proactive health monitoring systems for industrial platforms.