Toward smart manufacturing: Analysis and classification of cutting parameters and energy consumption patterns in turning processes

Abstract

Advanced monitoring technologies with embedded devices, sensors, and wireless data communication have been developed to capture machine, process, tool dconditions, and energy consumption and become essential to guarantee industrial intelligence in modern manufacturing. This paper studies the correlation of turning process parameters of a CNC machine with the tool vibration, acoustic signal, and energy consumption using preliminary data analysis and machine learning methods. A turning process is being monitored through three different sensors, namely vibration, acoustic and current. A vibration sensor was attached to the tool to measure the variation in tool vibration with respect to machining conditions. Additionally, a microphone was attached to the tool holder to capture the acoustic signature of the process. Further, current clamps were attached to the power leads going into the machine’s main circuit breaker, allowing to monitor the power consumption during operation. Machining parameters consisted of three different spindle speeds, three feed rates, and the depth of cut was kept constant throughout all experiments. Signals from each sensor were independently analyzed to examine the effect of process parameters on the change of vibration, acoustic signature, and power consumption. Additionally, low-cost machine learning models were applied to classify the cutting conditions using the measured acoustic signals as training, validation, and test samples. It was found that vibration signals are more sensitive to the change in spindle speed while power response is more sensitive to the change in feed rates. On the other hand, speeds and feeds have combined effects on the acoustic signal. Finally, the low-cost machine learning model using the Quadratic Discriminant Analysis (QDA) classifier could correctly correlate the acoustic signals to nine different operating conditions with an accuracy of 86.9%