A Symbolic Regression Based Residual Useful Life Model for Slewing Bearings

Abstract

Slewing bearings are vital functional components of large machinery. It is of far reaching significance to study their life prediction and health management. Many studies are based on data-driven approaches. However, part of them in the form of “black-box” lack actual physical meanings due to opacity model structures and have difficulty in choosing optimal parameters. Few kinds of literature focus on explicit model relationships for slewing bearings' life models. In this paper, a novel approach based on symbolic regression is proposed with the aim of exploring slewing bearings' explicit life models in depth and to predict residual useful life (RUL). The proposed method integrates the strengths of multiple signals describing a comprehensive response to slewing bearings' health and various genetic programming (GP) algorithms modeling life expressions. In addition, independent, hybrid, and piecewise strategies are introduced and explicit model relationships with respect to degradation indicators (DIs) are established via GPs. To verify the proposed method, three run-to-failure experiments under discrepant operating conditions of slewing bearings are carried out. Prediction results demonstrate that models generated by epigenetic linear genetic programming (ELGP) under hybrid and piecewise modeling strategy with similarity-based combination strategy perform best. More importantly, their life expressions are more succinct and intelligible than in other situations.