Machine learning based electrical imbalance detection of ball screw feed drive in machine tools

The performance of ball screw feed drives determines the precision and effectiveness of modern machine tools. However, drive motor electrical connection faults can result in current imbalance and pulsating torque, leading to excessive vibration and loss of accuracy, but can be hard to diagnose. This paper proposes a low-cost current-based predictive maintenance framework using three-phase motor current signals. A supervised machine learning approach is employed for both binary fault detection and multiclass imbalance severity classification. Time domain features, including root mean square (RMS), skewness, kurtosis, and cross-correlation, are extracted from sliding windows of the current signals and used to train a Support Vector Machine (SVM) classifier. Experimental results show that the proposed SVM achieves 94% accuracy in binary classification for the severe fault condition and 78% accuracy for multiclass severity classification up to 41% of motor impedance. Comparative results with Random Forest and Convolutional Neural Network (CNN) models demonstrate that SVM provides the best balance between accuracy and computational efficiency.