A Lightweighting Methodology for Diagnostic Models Based on Depthwise Separable Convolution

Current deep learning-based fault diagnosis methods are often hindered by substantial computational demands, including large parameter counts, significant memory footprint, and prolonged inference times. These limitations challenge the deployment capabilities of standard computers and prevent real-time monitoring on edge devices. To address this, we propose a model lightweighting method employing depthwise separable convolution. This study systematically explores both global and local architectural adaptations of a fault diagnosis model. Through comparative experiments, we identify the better model that maintains diagnostic accuracy while achieving a significant reduction in computational complexity and storage requirements.