Fault classification of cooling fans using a CNN-based approach

Abstract

In industries, cooling fans are vital in a wide range of machines to ensure a tolerable temperature for their intricate electronic components. Therefore, to avoid machine failure, a fault condition monitoring (FCM) system for cooling fans can be highly valuable. One way to monitor defects in rotational equipment is to analyze the machine vibration, which varies as the components wear off. Hence, this paper presents a technique to diagnose faults in cooling fans by analyzing the vibration data. In this conference paper, convolutional neural networks (CNNs) are used to classify the faults based on the vibration. The vibration data are collected from an experimental test bench that consists of a fan, an accelerometer, and a microcontroller, among others. The data were used to train the VGG16 and ResNet50 CNN architectures. The accuracy and effectiveness of these two architectures for vibration analysis are compared in this paper.