A Study on Classification of Faulty Motor Sound Using Convolutional Neural Networks

Jamil Md Shafayet, Gunaratne Praveen, Tamura Hiroki

doi:10.5954/icarob.2024.gs1-2

Classification of sound has its usage in various fields in today's world. In this paper we will go through the sound classification techniques for the detection of faulty machines with the help of the sound data produced by the machine. The focus is towards determining the pertinency of audio classification methods to detect faulty motors by their sounds; both in noisy and noise-free scenarios; so that the requirement of human inspection can be reduced in factories and industries. Noise reduction plays such an important role in improving accuracy of detection some researchers simulated data by adding noise for benchmarking their models. Hence noise reduction is widely used in audio classification tasks. Among various available methods, we have implemented an autoencoder for noise reduction. We have conducted the classification tasks on both noisy and denoised data using Convolutional Neural Network. Accuracy of classification on the data denoised using autoencoder is compared with the noisy ones. For classification, we used spectrogram, Mel-frequency cepstral co-efficient (MFCC) and Mel-spectrogram images. These processes yield promising results in distinguishing faulty motors by their sound.

A Study on Classification of Faulty Motor Sound Using Convolutional Neural Networks

Description

Journal

Keywords

Details 詳細情報について

Export

Report a problem