Hybrid Analytics in Smart Manufacturing Systems—Addressing the Manufacturing Data Challenge

ABSTRACT Smart manufacturing has opened tremendous opportunities to access, collect, and analyze a plethora of process and product data. Simultaneously, the manufacturing domain presents unique challenges regarding data for current modeling approaches—be it machine learning or physics-based models. In this paper, we highlight the opportunity presented by combining data-driven and physics-based models in a hybrid approach to address these data challenges. The paper provides a depiction of the unique data challenges in the manufacturing domain, illustrates the different facets of data analytics in manufacturing (including physics-based, data-driven, and hybrid modeling), and provides a qualitative mapping of fit for the different modeling classes on the data challenge dimensions.