Bearing fault diagnosis

What is it about?

The provided research presents an innovative methodology for bearing fault diagnosis in mechanical systems by combining Continuous Wavelet Transform (CWT) with Derivatives of Gaussian (DoG) wavelets, Dynamic Time Warping (DTW), and a Multi-Layer Perceptron (MLP) classifier. The framework initially extracts and decomposes raw vibration signals from machinery to capture non-stationary characteristics, followed by an energy-based selection strategy to isolate dominant scales. Subsequently, DTW is applied to these filtered signals to quantify temporal similarities and generate a robust feature set capable of handling operational variability. This refined data is then normalized and fed into an MLP neural network designed with multiple hidden layers and dropout regularization to prevent overfitting. Experimental evaluations conducted on the standard Case Western Reserve University (CWRU) dataset demonstrate that this integrated CWT–DTW–MLP framework achieves an exceptional classification accuracy of 99.83% in distinguishing between healthy and various faulty bearing conditions. Ultimately, the system provides a reliable, accurate, and computationally sound solution for real-time industrial condition monitoring and predictive maintenance.

Featured Image

Photo by Glen Carrie on Unsplash

Photo by Glen Carrie on Unsplash