A Review of Fault Prognosis of Model-based Mechatronic Systems

Abstract

Fault prognosis represents a foundational core of modern Prognostics and Health Management (PHM) paradigms and Condition-Based Maintenance (CBM) strategies. It shifts the technological focus from localized backwards-looking fault troubleshooting to prospective, long-term remaining life forecasting. By assessing cumulative mechanical wear, thermal shifts, and electrical degradation tracking profiles, fault prognosis calculates the precise Remaining Useful Life (RUL) and Time-of-Failure (ToF) metrics of high-value industrial mechatronic assets. This comprehensive review systematically frames the structural layout of model-based fault prognosis strategies engineered for multi-domain mechatronic environments. Model-based architectures are critically examined across graphical mathematical regimes—encompassing Bayesian Networks, causal Bond Graphs, and Hybrid Bond Graphs—alongside physical thermodynamic formulations including conflict-driven algorithms and steady-state state analytical models. To present an exhaustive, state-of-the-art diagnostic landscape, these physical-principles approaches are contextualized against sequential filtering data-driven tracking algorithms (such as Particle Filtering and Hidden Markov Models) and knowledge-based expert reasoning configurations. Finally, this review isolates major deployment bottlenecks, including mathematical tracking errors under volatile ambient conditions and multi-mode operation transitions, and details unified hybrid structural frameworks engineered to elevate prognostic precision in contemporary autonomous factories.