Remaining useful life prediction in embedded systems using an online auto-updated machine learning based modeling

Systems on Chips are increasingly involved in critical equipment the fields of aeronautics, transportations, and energy. Therefore, monitoring their life cycle is a crucial issue for safety hazard-prevention. This paper deals with data-driven method online prediction Remaining Useful Life (RUL) safety-critical System-on-Chips (SoC). based detection drifts operating temperatures. The work starts description formal relationships between temperature degradation process SoCs to justify choice as an indicator level system. Then, temperature-based physical health indicators constructed using analytical redundancy. Since varies not just according state system, but also its various normal points, redundancy makes it possible obtain that has well-defined meaning, which only sensitive SoC process. To predict remaining useful chip, trend drift modeled auto-regressive neural (NAR) network. latter updated evolution Finally, forecasts obtained combination temporal projection threshold data. Simulations experimental results highlight effectiveness accuracy proposed approach.