Gearbox Fault Diagnosis Using Convolutional Neural Networks And Support Vector Machines

Fast and accurate fault diagnosis is important to ensure the reliability and the operation safety of rotating machinery, which is often based on vibration analysis. In this paper, a novel approach combining Convolutional Neural Networks (CNN) and a Support Vector Machine (SVM) classifier is proposed, in order not only to leverage upon the advantages of deep discriminative features (learnt by the CNN) but also to exploit the generalization performance of SVM classifiers. Firstly, the Continuous Wavelet Transform (CWT) is employed to obtain the pre-processed representations of raw vibration signals. Then a novel CNN with a square-pooling architecture is built to extract high-level features, without requiring extra training and fine-tuning and thus demanding reduced computation cost. Finally, a SVM is used as classifier to conduct the fault classification. Experiments are conducted on a dataset collected from a gearbox. The results demonstrate that the proposed method achieves competitive results compared to other algorithms in terms of computational cost and accuracy.