Multilabel artificial intelligence model for online monitoring of electrical discharge turning by audio-based signals

In today’s industry, artificial intelligence (AI) methods are widely used for various techniques in process monitoring. They have demonstrated their reliability in monitoring tasks. These models offer fast, precise, and high-performance responses, making them ideal choices for manufacturing process monitoring. This is particularly significant in processes like electrical discharge turning (EDT), which involves numerous input parameters often prone to human errors. We developed a multilabel AI model to address the challenge of handling multiple input parameters in EDT. Our AI model allows for independent monitoring of individual input parameters. Monitoring the EDT with the current methods, such as acoustic emission, has some limitations, including low-quality signals in rotary parts, inappropriate response in high temperatures, and high cost. In our work, we proposed using the audio signals generated during the EDT by microphone. The signals were preprocessed using the fast Fourier transform (FFT). The performance of our AI multilabel model substantially improved by obtaining the best preprocessing parameters sequentially. The results demonstrated that our proposed AI model achieved high precision (89%), recall (86%), and F1 score (88%) for the micro average scores on the test dataset, implying that our AI model can detect the EDT monitoring factors accurately, which is incredibly difficult or almost impossible for humans by listening to the generated sounds. Our AI model can monitor complex industrial processes like EDT. This accurate monitoring model can detect independent monitoring factors using the generated sound and opens new doors for efficiently automating and monitoring other industrial processes.