Uso de inteligência artificial para identificação de condições de usinagem a partir de sinais de áudio

Abstract

Machining operations are crucial to global industry, so the quality and efficiency of the machines that perform these processes are crucial to a company's bottom line. Monitoring these machines thus gains relevance by preventing failures and ensuring process quality. Although most sensors used for monitoring are accelerometers, and vibration analysis is essential, the use of microphones emerges as a powerful and low-cost option, coupled with the use of artificial intelligence (AI) algorithms to analyze sound signals. This work uses AI techniques to classify audio signals from computer numerical controlled (CNC) lathes during the turning process of a titanium alloy cylinder (Ti-6Al-4V ELI), captured from a smartphone. The sound signal obtained from nine tests with different machining parameters and two background noise files was processed, and the Random Forest, Logistic Regression, k-nearest neighbours (kNN), and support vector machine (SVM) models were tested to identify the most capable of predicting the tests and, consequently, the machining parameters of each signal. All models performed well, with area under the receiver-operating curve (AUC) > 93%. However, the logistic regression model stood out with 96.2% accuracy and 99.75% AUC. These results demonstrate great potential for this monitoring technique.