augmented artificial intelligence for smart manufacturing
peng (edward) wang, ph.d.
assistant professor, endowed robley d. evans faculty fellowship
institute for sustainable manufacturing
department of electrical and computer engineering
department of mechanical engineering
university of kentucky
over the past decade, the manufacturing community has been enthusiastically embracing industry 4.0 techniques, such as the internet of things (iot), edge and cloud computing, and artificial intelligence (ai), toward improved production visibility, planning, productivity, quality control, and safety. however, most ml solutions and tools are developed in lab environments and lack credibility for practical implementation in manufacturing plants. this presentation highlights some preliminary works on developing generalizable and applicable machine learning (ml) solutions on the manufacturing shop floor. two case studies are demonstrated in the context of robotic automation of welding processes and machine condition monitoring.
in the first case study, a hybrid ml framework that integrates visual welding sensing with pixel-level image segmentation-enabled weld pool quantification, efficient perceptron-based process-quality modeling, and advanced gradient descent controller for adaptive process parameter adjustment, is demonstrated to realize a closed-loop, real-time, adaptive robotic control and automation of arc welding. in the second case study, to address the data discrepancy and model generalizability issue, a novel approach is demonstrated to automatically learn from unlabeled data streaming and continuously update ml models to adapt to new machine conditions without prior knowledge of condition change points, through the integration of self-supervised learning (ssl) with continuous learning (cl). these studies set the foundation for developing augmented artificial intelligence for generalizable and applicable data learning in smart manufacturing.