Abstract

The cognitive computation of a digital-twin becomes time-intensive beyond the requirement of a smart system when it uses simulation tools that solve governing constitutive equations in the form of partial differential equations (PDE). However, such physics-based digital twins offer a critical core-competency that enables the smarting through limited data. The current data-driven digital twins that use machine learning (ML) algorithms take significant initial data to mature for smarting. In most manufacturing processes, there is no such data. Therefore, cyber-manufacturing systems work with limited data and adaptive learning. For example, when fabrication deals with overlay metal deposition (also known as cladding), the digital twin can manage the adverse effects such as distortion and residual stress during the deposition. However, the process of finding an effective deposition pattern is a challenging task given a large number of possible deposition scenarios. We build a hybrid digital-twin that takes advantage of ML algorithms for quick response while gaining fidelity through adaptive learning with FEA simulation tools. Our digital twin consists of a quick learner that is data-driven to be responsive, and it uses an active learning algorithm to wisely navigate the data selection toward a higher training rate and gain fidelity using critical data points rather than an aggregated data set. We use our hybrid digital-twin to explore various overlay scenarios in real-time to form a platform for smart cladding. We present our platform with an actual overlay deposition application on panel structures, including countless depositions scenarios. This digital twin has gained an acceptable fidelity with 100 interactive and iterative labelling queries to FEA. The learner uses a physics-guided machine learning approach to become more informed and reducing data dependency. This hybrid digital-twin is packaged as a cloud-based tool that enables engineers to analyze and compare different patterns to assess fabrication scenarios without computational time delay. Smart systems are not explicit programming; they are architectured to learn continually. Our hybrid digital twin learns from using. The more to use, the higher fidelity evolves in the cloud.