Development of predictive model for vibro-acoustic condition monitoring of lathe

Present day requirements for enhanced reliability of rotating machinery have become critical for the manufacturing sector. Every rotating machine exhibits a unique characteristic vibration and acoustic signature. This can be used to identify the defective parts and estimate the present severity of the problem; most importantly, without opening the machine for inspection. Moreover, it aids in the reduction of unscheduled down time, turnaround time and existing noise levels. The paper deals with the vibro-acoustic condition monitoring of metal lathe and development of predictive models for the detection of probable faults using Machine Learning. Experiments were conducted to obtain vibration signatures using accelerometers and the data was further processed while the acoustic signatures were obtained using noise level meters. Results were obtained for idle running, turning and facing operations using a single point cutting tool for constant spindle speeds, feed and depth of cut. The vibro-acoustic signatures of six metal lathe machines were collected over a period of 5 months and the trends obtained were analyzed. The filtered acceleration (g-peak) signatures were compared with the General Machinery Vibration Severity Chart and based on the velocity classification results, the best machine was chosen for the development of predictive models. Vibration as well as acoustic signatures were isolated using filters, empirical relations and manufacturing data. Predictive models were made using machine learning algorithms to predict the failure of the lathe based on its historical data. These models can be used by industries to detect unhealthy trends and identify troublesome parts in the machine which can be then scheduled for maintenance thereby decreasing production downtimes