Real-time adaptable fault analysis of rotating machines based on Marine predator algorithm optimised LightGBM approach

Abstract

ABSTRACT Intelligent fault diagnosis (IFD) techniques are important for fault analysis. It is challenging to constitute a real-time adaptable technique that can handle different types of fault conditions and fault severity values under variable machine settings. Based on this challenge, this study proposes an effective approach called Marine Predator Algorithm optimised Light Gradient Boosting Machine (MPA-LightGBM) for fault diagnosis and severity analysis of rotating machines regarding both single and compound faults. For the fault diagnosis, the MPA-LightGBM classifier is employed. The suitability of the proposed approach for real-world applications is validated by performing a real-time analysis using unseen data collected from a rotating machine operated at different rotational speeds. The computational load is alleviated by eliminating the redundant features up to 85.00% via MPA. According to experimental findings, the suggested method was able to accurately identify the individual cases for the testing set by 99.59% and 94.88% for compound faults. Regarding real-time analysis, the identification performance is 99.48% for individual faults while it is 92.33% for compound faults. The fault severity prediction error of the proposed approach considering the testing set is 0.55% and 1.15% for the individual and compound faults, while it is 0.50% and 1.88% regarding the real-time analysis.