Optimal transport-based transfer learning for smart manufacturing: Tool wear prediction using out-of-domain data

Abstract Various machine learning-based predictive modeling approaches to tool wear prediction have been introduced over the past few years. However, predicting tool wear under different operating conditions (e.g., depth of cut, feed rate, and workpiece material) with small datasets remains a challenge due to complex tool wear mechanisms. To address this issue, an optimal transport (OT)-based transfer learning algorithm is developed to transfer knowledge on tool wear from one operating condition to another. The OT-based transfer learning model has been demonstrated on a small dataset collected under different operating conditions. Experimental results have shown that the OT-based transfer learning method significantly improved tool wear prediction accuracy.