Reliability-centered tool RUL prediction under time-varying processing parameters

doi:10.1016/j.ress.2025.112016

Reliability-centered tool RUL prediction under time-varying processing parameters

2025-11-22

0

PRE

AI

DOI：10.1016/j.ress.2025.112016

Original

Original request for help

Save

Abstract

En 中文

Accurate remaining useful life (RUL) prediction of a manufacturing tool under time-varying processing parameters (TVPP) is essential for sustaining the optimal performance of manufacturing systems. However, conventional time-series models suffer from lack of physical interpretation and limited training data. This study proposes a hybrid physics-based data-driven model designed to predict tool degradation while preserving physical details. The tool’s RUL is predicted by evaluating the influence of degradation on processing reliability and facilitates interfaces among heterogeneous model types. Physics-based degradation models, which are based on process and degradation mechanisms, are first deployed to generate primary data. Then, a new staged and sequential nonlinear model (S²NLM) is introduced to predict degradation under TVPP by treating processing parameters and degradation state in two sequential stages, which aligns with degradation features and is robust to limited data under time-constant processing parameter. These data-driven predictions of degradation trends under TVPP are mapped to the physical domain with explicit physical meaning. Finally, tool RUL is predicted on the basis of its impact on processing reliability, presenting accurate and rational results and prompt responsiveness to varying quality requirements. Compared with other methods, the proposed S²NLM achieves superior accuracy under limited data in a rolling processing case.

Reliability-centered tool RUL prediction under time-varying processing parameters

Abstract

Journal

Organization