Sinc-based Multiplication-Convolution Network for Equipment Intelligent Edge Diagnosis under Small Samples

Abstract

Data-driven intelligent diagnosis models need massive monitoring data to train themselves for desired performance. However, in many engineering scenarios, collecting fault data is often expensive and time-consuming, which leads to few-shot learning becoming a valuable research hotspot for intelligent diagnosis. Inspired by mode characteristics and feature enhancement learning, this study propose a Sinc-based multiplication-convolution network (SincMCN) for intelligent fault diagnosis under small samples. It works in frequency domain, and consists of only three layers, including a feature separator, a feature extractor and a classifier. In the feature separator, a series of Sinc-based multiplication filtering kernels (SincMFKs) are designed for improving the utilization of fault information of spectrum samples. The products between SincMFKs and spectrum samples are stacked into activated mode spectrum images (AMSIs) with rich fault-related features retained. Since AMSIs are concise enough, this study employs only a 2D convolutional layer and a fully connected layer as the feature extractor and the classifier for achieving a fast and precise pattern recognition. Experimental results show SincMCN has better diagnosis accuracy and stronger potentials for few-shot diagnosis compared other cutting-edge models. Specially, analytic filtering kernels not only cut down the model parameters for edge diagnosis and provide powerful application potentials and engineering value for online monitoring of rotating machinery.