Fault Diagnosis Using Wavelet Based Denoising Integrated with Neural Networks: A Signal Processing Informed Neural Network Framework

Utilising a signal processing-informed neural network (SPINN) built around wavelet network-driven denoising, this research introduces an innovative paradigm for fault identification. Traditional defect detection approaches often struggle with noisy signals, resulting in reduced diagnostic accuracy. We address this problem using the learning capability of neural networks and the denoising ability of wavelet networks. The proposed framework uses a multi-phase process: in the first step, a wavelet-based denoising module is used to process the raw signal to clean it of noise while preserving relevant signal characteristics related to faults; the second phase involves taking the denoised data and inputting it into a neural network that would be able to detect and locate faults with higher accuracy. By embedding domainspecific signal processing knowledge directly into the framework of the neural network, the system can adapt better in noisy conditions. Experiments on standard datasets show notable gains in the generalisation, resilience, and precision of defect identification. For real-time, noise-resistant problem diagnostics in engineering and industrial applications, this method presents a viable avenue. The findings of the study shows that 94.7 % accuracy 92.7 % precision 93.1 % recall and F1 score is 92.9 during validation phase.

Перевод пока недоступен