Development of an Abnormal Sound Detection System for Rotating Bodies Using Deep Learning

Abstract

Conveyor systems have a rotating body component that generates various sound signals when malfunction occurs. This work studies the development of a system that collects sound signal data generated during operation of a rotating body, applies an artificial intelligence (AI) algorithm over the data, and diagnoses if the motion of the rotating body is abnormal. The conveyor belt system is installed in various working environments such as mines and factories and consists of a number of rotors. Multiple connected data collection devices developed in this work are placed around the conveyor belt rotors to detect abnormalities in the rotors. Each collection device is equipped with the AI algorithm, which performs the data preprocessing on the sound data collected by each device. For the diagnosis of the rotor failure, sound data collected from the devices are converted to 2D spectrogram images, and a convolutional neural network model is applied to the converted images for determining the status of the rotating body which is transmitted as normal or failure to the manager via Ethernet TCP/IP protocol.