Comparisons of different deep convolutional neural network and machine learning based methods on gearbox fault diagnosis using small dataset

Abstract

Modern industry prioritizes condition monitoring and problem diagnostics due to safety and quality standards. Modern gearboxes, one of the most common components, break under intense operating conditions and require problem detection. Vector assessment and vibration signal analysis have successfully used deep learning to extract representative information and sensitive features from raw data to diagnose gearbox faults. Deep learning for mechanical diagnostics is relatively restricted, and few research have compared feature learning with varied data sources. This study uses vibration signal temporal data to train a convolutional neural network (CNN) using multiple architectures. UoC gearbox data verifies the technique against seven typical intelligent ways. Adaptive learning from temporal data enhances diagnostic accuracy.