A framework for reconfigurable production line changeover task planning based on large language model

Addressing the issues of manual intervention, low efficiency, and high error rates in the changeover operations of reconfigurable production lines, this paper proposes an automatic changeover task planning framework for reconfigurable production lines based on Large Language Model(LLM). Utilizing this framework, the instructions input by the user can be interpreted, and the changeover robot can be controlled to automatically complete the changeover of the production line.The framework is primarily composed of five essential components: knowledge graph, high-level task planner, low-level task planner, changeover robot, and reconfigurable production line. The knowledge graph is utilized to depict the attributes and interrelationships of products, processes, tool & fixture tool & fixture resources, and programs of the reconfigurable production line.The semantic instructions input by the user are interpreted by the high-level task planner, which then decomposes the task into changeover subtasks, matches robot skills to each subtask, and generates a sequence of skills. With the low-level task planner, the sequence of skills is translated into executable code for the changeover robot, which controls the robot to complete the automatic changeover task. A low-code industrial software is designed to implement skill-based modular programming, and the four-layer control architecture including the physical, control, skill, and application layers has been implemented. The changeover robot was designed and fabricated. Through single-robot task planning experiment and actual production line changeover task planning experiment, the correctness of the proposed changeover task planning framework is demonstrated, automatic changeover of the production line can be realized.