Optimal machine learning for detecting lathe machining parameters

Remote manufacturing process monitoring is becoming more popular in industry for many reasons including increased production, reduced production costs, and better product quality. However, a downside to remote monitoring is that there is minimal human interaction with the machine to ensure optimal performance. Many types of sensors have been used to monitor the manufacturing process parameters of a machine, but these types of sensors are typically invasive to the machine, costly to install, and can interfere with the machining operation. A solution to this problem is to use a non-invasive sensor such as a microphone array to remotely detect manufacturing process parameters. Acoustic signals produced by a lathe are used to monitor spindle speed, depth of cut, and feed rate to ensure optimal performance. Based on these acoustic signals, machine learning algorithms are used to predict manufacturing process parameters. Two common learner algorithms were discussed and compared, such as k-nearest neighbor (kNN) and Gaussian support vector machine (Gaussian SVM). Each algorithm was classified as either “fine” or “medium.” Various pre-processing algorithms were also discussed and compared, such as autocorrelation and/or cross-correlation in both time domain and frequency domain. This study found that autocorrelation in frequency domain yielded the best overall results for all three machining parameters, with learner algorithms of fine kNN or fine Gaussian SVM for depth of cut, fine Gaussian SVM for feed rate, and either fine kNN or fine Gaussian SVM for spindle speed. For prediction of machining parameters, an accuracy range of 90–100% was obtained.