The Variational Kernel-Based 1-D Convolutional Neural Network for Machinery Fault Diagnosis

Abstract

One-dimensional convolutional neural network (1-D CNN) can be directly applied to process temporal signals in the machinery fault diagnosis. However, it requires a large amount of data to train high-quality kernels and extract meaningful features. This article develops a novel method integrating learnable variational kernels into a 1-D CNN to pay more attention to extracting important fault-related data features and offer decent performance with even limited data. In the proposed method, the variational kernel is first derived by adapting constraints and formulations of the successive variational mode decomposition (SVMD). Next, a gradient descent process based on a fault classification loss is developed to estimate the parameter of the variational kernel. Lastly, a CNN-based diagnosis model is constructed to perform machinery fault diagnoses. The proposed model is benchmarked with two standard 1-D CNNs and four wavelet kernel-based 1-D CNNs. Datasets collected from train bearing and gearbox test rigs are used for validations. Results show that the proposed model identifies important fault-related features and performs better in machinery fault diagnoses with limited training data.