Special Session on Computational Intelligence for Anomaly Detection, Diagnosis, and Prognosis, IEEE World Congress on Computational Intelligence (WCCI 2008),2008

A Hybrid Prognostics Methodology for Electronic Products

Sachin Kumar
Myra Torres
Y. C. Chan
Michael Pecht

Prognostics Health Management Group
Center for Advanced Life Cycle Engineering (CALCE)
Department of Mechanical Engineering
University of Maryland
College Park, MD 20742

Abstract:

Prognostics and health management enables in-situ assessment of a product’s performance degradation and deviation from an expected normal operating condition. A unique hybrid prognostics and health management methodology combining both data-driven and physics-of-failure models is proposed for fault diagnosis and life prediction. The shortcomings of using data-driven and physics-of-failure methodologies independently are discussed. These approaches estimate future system health, based on a systems current health status, historical performance, and operating environmental conditions. Although these methodologies are applicable to legacy, current, and future electronics, and ranging from components to circuit assemblies and electronic products, the hybrid approach is preferred due to its capability to include potential failure precursor parameters with failure mechanism, thus improving accuracy in prognostic estimates. Various works on data-driven and physics-of-failure approaches to prognostics for electronics are summarized and a hybrid methodology case study is presented.

Full Text