Unified IGBT prognostic using natural computation

Within the field of Integrated System Health Management, ISHM, there is still a lack of adequate prognostic models for critical applications while systems are missioned to work in harsh environments. One of the main challenges is that prognostic models should be adaptive to changes in working conditions so that prediction of the remaining useful life of systems/components can be well adjusted with the dynamic of variation of both system and working conditions. Among various modeling and prediction techniques, natural computation and soft computing techniques, such as neural networks, offer interesting solutions to adjust prediction of the remaining useful life of systems/components while the complexity of modeling and real-time calculation is also reduced. This paper presents a radically-novel approach for building per-unit prognostic models applied on one of the most critical components of power modules, IGBT, that presents high failure rates in power electronic systems. An advantage is that the prognostic model can be generalized in a per-unit form; and then, its features are adjusted depending on the application, working condition, and dynamic of changes.