Large Language Model-Based Tuning Assistant for Variable Speed PMSM Drive with Cascade Control Structure

A cascade control structure (CCS) is still the most commonly used control scheme in variable speed control (VSC) electrical drives with alternating current (AC) motors. Several tuning methods are used to select the coefficients of controllers applied in CCS. These approaches can be divided into analytical, empirical, and heuristic ones. Regardless of the tuning method used, there is still a question of whether the CCS is tuned optimally in terms of considered performance indicators to provide high-performance behavior of the electrical drive. Recently, artificial intelligence-based methods, e.g., swarm-based metaheuristic algorithms (SBMA), have been extensively examined in this field, giving promising results. Moreover, the intensive development of artificial intelligence (AI) assistants based on large language models (LLMs) supporting decision-making processes is observed. Therefore, it is worth examining the ability of LLMs to tune the CCS in the VSC electrical drive. This paper investigates tuning methods for the cascade control structure equipped with PI-type current and angular velocity controllers for PMSM drive. Sets of CCS parameters from electrical engineers with different experiences are compared with reference solutions obtained by using the SBMA approach and LLMs. The novel LLM-based Tuning Assistant (TA) is developed and trained to improve the quality of responses. Obtained results are assessed regarding the drive performance, number of attempts, and time required to accomplish the considered task. A quantitative analysis of LLM-based solutions is also presented. The results indicate that AI-based tuning methods and the properly trained Tuning Assistant can significantly improve the performance of VSC electrical drives, while state-of-the-art LLMs do not guarantee high-performance drive operation.