CAPP-GPT: A computer-aided process planning-generative pretrained transformer framework for smart manufacturing

Abstract

Smart manufacturing (SM) constitutes the backbone of Industry 4.0 (I4.0), allowing for heightened autonomy of the various interacting cyber-physical systems, making the various entities on the production floor. Connectivity, a vital enabler, plays a crucial role through state-of-the-art Digital Twinning (DT) technologies driven by underlying innovations like the industrial Internet of Things, Cloud Computing, and advancements in sensory devices. DT, which plays a vital role in the various planning functions under the production and operations management umbrella, is being used in the developed combined CAPP-GPT (Computer-Aided Process Planning-Generative Pretrained Transformer) and production scheduling approach to address disruptions on the shopfloor and in self-healing of the manufacturing processes at a micro-CAPP level by optimally adapting the process parameters and the developed toolpath on the fly based on online process signature measurements. In a leap commensurate with that which has taken place in Natural Language Processing-Large Language Models (Chat-GPT), similar efforts are currently being undertaken to parse CAD data structures and blueprints, fusing operations research and predictive analytics algorithms to carry out setup planning as well as sequencing and grouping manufacturing sub-operations. A hybridized Optimization and Machine Learning (ML) approach is employed where Logical Analysis of Data is used to solve the problem heuristically, exploiting various generative and variant methods at heart. Another extension of this macro-CAPP problem is being tackled by integrating the problem with delayed product differentiation, lot-sizing, and transfer line balance for futuristic batch-production shops employing Hybrid Manufacturing (HM) and Smart Assembly. At the micro-CAPP level, HM process parameters are optimized using a comprehensive approach employing the Taguchi loss function to assess surface roughness, internal failure costs, and other criteria, including greenhouse gas emissions and expended energy. Online measurements of the process signatures are also employed to adapt the initial set of process parameters using different automatic control schemes. ML is used to identify the process parameters carrying simulations on Simulink before the system is deployed.