Detection of Fault from Acoustic Signals in Automobile Engines using Deep Learning Techniques

Detecting faults in automobile engines from sound signal is a challenging task at the production phase of automobiles. That is why it attracts engineers and researchers to handle this issue thereby applying various solutions. In this work, we propose deep learning-based fault detection mechanism in automobile engines from different sound resources. In the dataset collection phase, various vehicle breakdown sounds are gathered from social media environments by constructing our own customized crawler. Moreover, noise addition is applied to increase the amount of data. Subsequently, raw audio files are processed at the feature exraction step employing mel-frequency cepstral coefficients. To detect the vehicle breakdown sounds, 1-D and 2-D convolutional neural networks, long short-term memory networks, artificial neural networks, and support vector machine are modeled. Experiment results show that the usage of 1-D convolutional neural network is transcendent with 99% of accuracy compared to the other techniques, especially, state-of-the-art studies are considered.