Gear Fault Detection with InceptionResNetV2 Transfer Learning

This study addresses the critical need for reliable gearbox fault detection in industrial machinery, aiming to enhance operational efficiency and reduce downtime. The research explores the application of deep learning techniques, specifically focusing on the InceptionResNetV2 transfer learning model, for analysing vibration signals using Constant Q Transform (CQT) spectrograms. CQT’s adaptive resolution and noise robustness make it superior for capturing time-varying frequency content in non-stationary signals, such as those from gearboxes. The custom dataset, derived from SpectraQuest’s Gearbox Fault Diagnostics Simulator, includes labeled CQT spectrograms of healthy and faulty gearboxes, preprocessed and augmented for model training. The InceptionResNetV2 architecture, combining Inception modules and residual connections, effectively captures multiscale features, achieving 98.04% accuracy in fault classification. Evaluation metrics, including precision, recall, and F1 score, confirm the model’s robustness. Comparative analysis with other methods highlights its superiority in handling limited data and noisy conditions. This work demonstrates the viability of transfer learning-based fault detection systems for industrial applications, offering a template for predictive maintenance solutions. Future improvements will focus on multi-modal fusion and explainable AI techniques for enhanced interpretability.