In modern competitive industrial environments, maintaining optimal efficiency and quality is of crucial importance. Traditional methods often rely on visual inspections or sensor-based monitoring, which may not detect subtle changes or provide real-time insights. In this thesis, a methodology was developed to improve real-time monitoring and evaluation of manufacturing processes using unsupervised machine learning algorithms, specifically the k-means algorithm, for the classification and analysis of acoustic data from production environments. The methodology includes identifying key acoustic features such as directionality, frequency spectrum properties, and statistical signal descriptors at multiple time constants, and using these features to train the k-means algorithm. The classification results enabled reliable categorization of acoustic signals, allowing for the identification of machine operating states and the detection of process deviations. This research confirms that clustering acoustic data using unsupervised machine learning algorithms has the potential to enhance real-time monitoring and optimize manufacturing processes, which can significantly contribute to increased efficiency and reduced production costs.
|
