Utilizing LLM models for advanced automation, manufacturing operations

Abstract

With the high rate of digitalization in manufacturing operations, extensive data gathering and local optimization have been made possible; however, the fragmented nature of the decision-making process in manufacturing operations, in the domains of production, quality, and maintenance, still persists. The state-of-the-art automation and artificial intelligence solutions that are currently available in the market are largely specific to the task and do not possess the ability to understand unstructured knowledge, reason across diverse contexts, and provide transparent support for the decision-making process to engineers and operators. The recent advances in large language models (LLMs) open new opportunities to provide cognitive and semantic reasoning capabilities in manufacturing operations. In this paper, a conceptual and design-oriented framework is proposed to utilize large language models (LLMs) to facilitate sophisticated automation in manufacturing operations. The proposed architecture for utilizing LLMs in manufacturing operations considers LLMs as a cognitive orchestration and decision support layer, where enterprise and shop-floor systems, operational context knowledge, and analytical and optimization tools are integrated. The proposed framework has clearly defined functional operational roles for LLMs in manufacturing operations, such as production, quality, and maintenance, and a formalized human-in-the-loop approach to ensure accountability, safety, and regulatory compliance in manufacturing operations, along with governance and validation mechanisms to address the risk of unreliable results from LLMs. The research provides a manufacturing-related architectural perspective for the automation enabled by large language model (LLM) technology, highlighting the role of LLM technology in the facilitation of semantic interoperability, cross-functional coordination, and explicable decision-making. Although the research does not include any validation, the suggested framework provides a basis for the upcoming industrial pilot research and supports the move towards robust and human-centric manufacturing systems, as suggested by the industry 5.0 concept.