Multi-domain feature-based early detection of bearing faults using MLP classifier on NASA IMS dataset

Abstract

The degradation of bearing components in industrial machinery leads to increased maintenance costs and unexpected operational downtime. This paper presents a novel methodology that integrates multi-domain statistical feature extraction spanning both time-domain and frequency-domain characteristics to enhance the precision of bearing fault detection. A Multi-Layer Perceptron (MLP) model was trained on the NASA IMS Bearing dataset, achieving a classification accuracy of 86.5% across five degradation stages. Experimental results demonstrate that the proposed method outperforms traditional classifiers such as Support Vector Machine (SVM) and Random Forest, particularly in data-scarce environments. Furthermore, the model is well-suited for deployment on resource-constrained embedded diagnostic systems. This approach offers a practical and efficient solution for predictive maintenance, contributing to the reduction of operational costs in industrial applications.