Prognostic Health Management for Mission and Safety Critical Electronics

Abstract

The increasing reliance on electrical systems to fulfill mission and safety critical applications has motivated the need for in situ monitoring of functioning electrical systems and a priori warnings of failure. In this research the drawbacks of traditional reliability methods when applied to mission and safety critical applications are discussed and a new paradigm for the reliability of electronics, prognostic health management, is demonstrated. In the prognostic health management reliability model wear and damage to individual electrical systems are monitored and advanced warnings of failure are issued to allow adequate time to plan mitigating action and avoid unplanned failures. The occurrence of an unplanned failure in a mission and safety critical electrical component is considered by definition to have an associated cost that is unbearably high. Leading indicators of failure have been developed that allow in-situ monitoring of the structural health of electronics, a method coined resistance spectroscopy. The presented techniques are non-destructive in nature and were purposefully designed to cheaply embed into new electrical systems. Methods for processing the stream of information from real time observations are presented in a manner that facilitates statistically defendable decision making and the optimization of safety, availability, and operating costs. A variety of recursive filters –least squares, Kalman, extended Kalman, and particle – are combined with prognostic methods (forecasting) to create seamless real time monitoring and prediction algorithms. Solder joint configurations studied include SnPb eutectic, high lead, lead-free (SnAgCu), copper columns, and micro coil springs. Architectures studied include ball grid array, land grid array, a novel spring interconnect and pin/spring electrical connectors. Test environments include drop/shock, vibration, and simultaneous temperature and vibration. Investigations relating to the implementation of the presented techniques demonstrate the practical nature of the work. A cost justification method that is accessible to engineers, technical managers, and executives is developed to quantify the business case for implementing prognostic health management for electronics. Particle swarm optimization methods have been used to demonstrate the expected future performance of implemented prognostic algorithms given a set of test data. The described methods have been shown to be particularly sensitive to damage in the novel spring interconnect designed for long duration space travel. Methods for verifying the correct operation of prognostic algorithms and validating that algorithms meet specified requirements (both online and offline) are discussed. A prognostic health management toolbox, coded in MatlabTM, has been created. The toolbox provides a foundation of verified and validated algorithms for implementation of the presented methods, and is highly extendable for the development of new prognostic algorithms. The code is generic enough to apply to electronics as well as other application domains where suitable leading indicators of failure exist. All of the work presented in this document was created in part with the prognostic health management toolbox.