Smart Optimization Manufacturing in Lightning Arrester Production: An Integrated Approach Using Clustering, Large Language Models, Knowledge Graph and Digital Twin Model

This study aims to enhance quality control in the production of varistor discs used in lightning arresters at Hitachi Energy by integrating K-Means clustering, Knowledge Graphs (KGs), and Digital Twin Models (DTMs). It addresses the challenge of achieving production stability through a systematic, data-driven approach, motivated by the limitations of traditional engineering-based adjustments. K-Means clustering segments production data to identify defect patterns and enable precise parameter adjustments. Knowledge Graphs formalize and standardize tacit engineering knowledge into actionable rules, supporting knowledge retention and transfer. Digital Twin Models simulate the production line for real-time optimization, reducing material waste and accelerating process improvement. Experimental results validate improvements in quality and stability, proposing robust tolerance ranges. This M.Sc thesis contributes to intelligent manufacturing by demonstrating a novel, integrated framework that bridges data analytics, semantic modeling, and dynamic simulation, offering scalable insights and practical solutions for industrial applications.