In modern industrial processes, efficient monitoring and analysis of acoustic phenomena are becoming crucial for ensuring traceability and optimizing performance. Traditional methods often rely on complex measurement systems or costly equipment, such as acoustic cameras, which creates the need for simpler and more energy-efficient approaches. In this study, a methodology for unsupervised classification of acoustic data using the k-means algorithm was developed, enabling the separation and categorization of acoustic events based on selected features. Special emphasis was placed on spectral, energetic, temporal, and cepstral characteristics, with one relatively unconventional feature emerging as particularly important for further implementation in FPGA circuits. The results demonstrated a high degree of consistency between the clustering outcomes and the actual recordings, confirming the reliability of the proposed method. This approach thus opens new opportunities for the development of cost-effective systems for real-time acoustic monitoring, which can significantly contribute to reducing production costs while enhancing anomaly detection and improving industrial process control.
|
