Preliminary Investigation on the Acoustic Characteristics of Turning Processes

Abstract

This research aims to characterize the turning process using acoustic signals (AS) for the purpose of remote condition monitoring. This will allow for non-invasive machine monitoring, reducing costs and interference in the machining operation. Various combinations of process parameters were investigated, including spindle speed, depth of cut, and feed rate. The machining parameters used herein were closely matched with those of a milling operation utilized in previous research. The intent is to investigate the use of AS to monitor and differentiate multiple machines around the shop floor, running simultaneously. The feed rates for the turning process were mapped to mimic those for the milling process. A spherical 32-microphone array was utilized for data collection with a sampling rate of 48 kHz. Frequency and time-domain characteristics were utilized to find distinguishing features of the AS. It was found that turning speeds produced noticeable differences in the observed peaks in the frequency content of the signal, providing a means of determining spindle speed from AS. Additionally, time-domain characteristics yielded discernible differences for both feed rate and depth of cut. An increase in the rms value was observed as the material removal rate (MRR) of the machining process increased. The results suggest that a combination of both frequency and time domain characteristics may be used to distinguish the process parameters. Feature extractions linked to MRR and the time/frequency domain can be used to expand AS monitoring to other process parameters and machines. Finally, a time-domain machine learning classifier was utilized for predicting the depth of cut. The Fine K-nearest neighbor (KNN) classifier was determined to provide the best results, with a prediction accuracy of approximately 62%.