Using Feature Engineering and Principal Component Analysis for Monitoring Spindle Speed Change Based on Kullback–Leibler Divergence with a Gaussian Mixture Model

Abstract

Machining is a crucial constituent of the manufacturing industry, which has begun to transition from precision machinery to smart machinery. Particularly, the introduction of artificial intelligence into computer numerically controlled (CNC) machine tools will enable machine tools to self-diagnose during operation, improving the quality of finished products. In this study, feature engineering and principal component analysis were combined with the online and real-time Gaussian mixture model (GMM) based on the Kullback–Leibler divergence’s measure to achieve the real-time monitoring of changes in manufacturing parameters. Based on the attached accelerometer device’s vibration signals and current sensing of the spindle, the developed GMM unsupervised learning was successfully used to diagnose the spindle speed changes of a CNC machine tool during milling. The F1-scores with improved experimental results for X, Y, and Z axes were 0.95, 0.88, and 0.93, respectively. The established FE-PCA-GMM/KLD method can be applied to issue warnings when it predicts a change in the manufacturing process parameter. A smart sensing device for diagnosing the machining status can be fabricated for implementation. The effectiveness of the developed method for determining the manufacturing parameter changes was successfully verified by experiments.

Keywords

• machine tool
• Gaussian mixture model
• principal component analysis
• feature engineering