Abstract
Prognostics and Health Management (PHM) plays a crucial role in Industry 4.0 revolution by providing smart predictive maintenance solutions. Early failure detection and prediction of remaining useful life (RUL) of critical industrial machines/components are the main challenges addressed by PHM methodologies. Today, most devices and systems have embedded electronic components for monitoring, control, and enhanced functionality. Hence, it is vital to monitor these electronic devices/systems for failure to avoid catastrophic events and manage operational cost. The prognostic methods commonly used rely on an accurate mathematical model (Model-based methods) based on the underlying physics-of-failure mechanism or run-to-failure historical data (Data-driven methods). In many cases, the lack of accurate physics-based models emphasizes the need to resort to machine learning-based prognostic algorithms. However, data-driven methods require extensive machine failure data incorporating all possible operating conditions and all possible failure modes pertaining to that particular machine/component, which are seldom available in their entirety. With emerging technologies and advancements in manufacturing processes, new devices/components are being developed to cater to these emerging needs. The significant challenges inhibiting reliability studies on such newly developed devices/systems are the lack of sufficient failure data and the lack of full-fledged physics-of-failure models. To address the generalization problem, hybrid approaches combining model-based and data-driven methods have been widely used in the recent past. Hybrid prognostic approaches have an upper edge on such new devices as they neither require an accurate degradation model nor a large amount of training data for the purpose of remaining useful life (RUL) estimation. In other words, they make the best use of partial knowledge and sparse data available for the new device/component under prognostic investigation. This work proposes an intelligent and adaptive hybrid approach wherein we combine a Bayesian inference-based state estimation method to warm-start a neural network model. The informed stochastic parameter initialization helps overcome the generalization issues faced by other hybrid approaches in the literature. The developed hybrid prognostic approach is tested out on commonly used electronic devices such as Lithium-ion batteries, Power MOSFETs and LEDs. Our overall results are accurate and robust, and, therefore, they can form the basis for condition-based maintenance and performance-based evaluation of complex systems.