An Efficient Intelligent Edge Diagnosis System Based on WDCNN in the Application of Equipment Fault Classification

Abstract

Edge computing has been extensively used in industries to solve problems such as data explosion. However, in the field of mechanical equipment fault diagnosis, there are still problems such as the large amount of data transmission and storage and the difficulty of applying intelligent algorithms to equipment edge fault diagnosis. Therefore, an intelligent edge diagnosis system (IEDS) is designed and applied to mechanical equipment fault diagnosis, which includes an edge diagnosis unit (EDU) and an edge network model. The EDU uses the STM32F405 chip as the core and acquires vibration signals by MEMS acceleration sensor. Using Deep convolutional neural networks with wide first-layer kernels (WDCNN) as the edge network model. Due to the limitation of hardware resources, this paper uses a fixed-point quantification method to minimize the hardware resource usage of WDCNN. The proposed method can efficiently implement the edge fault diagnosis of mechanical equipment, which is reflected by the fact that IEDS can be deployed directly at the edge of equipment to accomplish efficient data reduction and accurate fault diagnosis results output. The proposed method provides a new solution for intelligent fault diagnosis of mechanical equipment.