Scalable bearing fault diagnosis using metaheuristic feature selection and machine learning for diverse operating conditions

Bearings are the foremost component of machinery that fails; hence early diagnosing mechanisms aided by artificial intelligence techniques play a crucial role. The objective of this study is to develop a scalable model for real-time bearing fault diagnosis using recent evolutionary algorithms such as, Marine Predator Algorithm, Equilibrium Optimization, Henry Gas Solubility Optimization, and Poor and Rich Optimization for optimal Feature Selection. Forty time-domain features were extracted from the bearing fault recordings from Case Western Reserve University, Machinery Fault Prevention Technology, Machinery Fault Dataset, and Paderborn University, with classes ranging from 3 to 48. The performance of the algorithms is investigated in terms of feature dimension, accuracy, and processing time. The performance of the five relevant feature groups that are identified is validated using four classical machine learning models and the most promising feature set that consistently performed well across all four datasets, regardless of the classifier model is identified. The compact feature set combined with the simplicity of the ML model delivers a computationally simple, yet effective feature extraction and classification model scalable for real-time bearing fault diagnosis under diverse operating conditions. The results obtained have been compared with existing works to demonstrate the resilience of the model.