PU-Foam 예측모델과 LLM 연계 사례를 통한, 제조업 AX를 위한 레거시 시스템과 LLM 간의 통신 메커니즘 연구 = Research on Communication Mechanisms between Legacy Systems and LLMs for Manufacturing Digital Transformation: A Case Study of PU-Foam Prediction Model and LLM Integration

다국어 초록 (Multilingual Abstract)

Establishing an efficient communication mechanism between Large Language Models (LLMs) and legacy manufacturing systems has emerged as a critical challenge in the digital transformation of modern manufacturing. Traditional Inter-Process Communication (IPC) approaches such as Socket, Pipe, and Shared Memory impose significant constraints on integration with heterogeneous legacy systems due to high implementation complexity and technology dependency. This study proposes a three-layer hybrid architecture utilizing Model Context Protocol (MCP) and quantitatively compares development efficiency and system performance by implementing two IPC approaches—file-based IPC and Qt Signal/Slot-based IPC—on an identical PU foam property prediction system. File-based IPC provides data integrity and complete process isolation through JSON format and atomic file operations, with McCabe cyclomatic complexity analysis revealing a value of 3.3 for core communication functions, approximately 1.5 times simpler than the Signal/Slot approach’s 5.0. The Signal/Slot approach achieved an average response time of 51.72ms and throughput of 33.85 req/s through memory-based communication, demonstrating 1.46 times faster performance compared to the file-based approach (75.72ms, 21.97 req/s). Experimental results confirm that file-based IPC excels in technology independence, process isolation, and maintainability, while the Signal/Slot approach shows advantages in real-time processing performance, indicating that the two approaches possess complementary characteristics. This study presents selection criteria for communication methods based on manufacturing environment requirements and provides a practical solution for communication between LLMs and legacy systems.