Accurate monitoring of machining process of Ti-6Al-4V using deep multi-task learning

Accurate machining process monitoring (MPM) is crucial to ensure manufacturing process stability and efficiency. In particular, tool wear, surface roughness, and cutting force comprise primary elements of MPM, which are physically interconnected and significantly affect machining quality and productivity. Hence, MPM should be performed through comprehensive consideration of the primary elements and their interactions. Yet, there is a lack of a unified approach to MPM that holistically predicts, which limits the practical applicability of MPM in real-world machining processes. Therefore, this work proposes a novel deep multi-task learning (MTL)-based method that integrates primary MPM elements by simultaneously predicting tool wear, surface roughness, and cutting force. To this end, the proposed method develops a probabilistic multitask likelihood that considers task uncertainty for seamless end-to-end training. The proposed method’s efficacy in MPM is validated with real-world datasets from an end-milling process of Ti-6Al-4V. The comprehensive experimental results demonstrate that the proposed method consistently shows superior prediction performance compared to existing single-task and MTL methods, indicating its effectiveness in real-world MPM.