Application of large language models in the motion control of cartesian tooth belt drive system
Edeh, Christian Nnadike (2026)
Diplomityö
2026
School of Energy Systems, Sähkötekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2026052050670
https://urn.fi/URN:NBN:fi-fe2026052050670
Tiivistelmä
This master’s thesis investigates the application of large language models (LLMs) for motion profile generation in a Cartesian tooth-belt drive system and evaluates their performance against conventional trajectory planning methods such as trapezoidal and S-curve profiles. The primary objective is to determine whether LLM-generated motion trajectories can improve tracking accuracy, motion smoothness, and energy efficiency under identical mechanical constraints.
To achieve this, an LLM-based motion generation framework was developed using structured prompt engineering with Anthropic Claude Sonnet 4.6. The generated trajectories were implemented in MATLAB/Simulink and compared with traditional motion profiles in a cascaded control architecture that consists of position and velocity loops with the dynamic model of the Cartesian tooth-belt drive system. The performance was evaluated using a detailed simulation model by analyzing position and velocity tracking errors, torque response, and energy consumption.
The simulation results showed that the LLM-based trajectory showed an improved performance compared to the traditional method. It achieved lower position root mean square (RMS) error with 10.25 % improvement, lower velocity RMS error (17.42 % improvement), reduced RMS torque (35.95 % improvement), and absolute peak torque improvement of 68 %. The total energy consumption was improved by 10.59 %, thereby demonstrating an improved energy efficiency of the system. The LLM-based profile also produces smoother motion with reduced dynamic excitation and improved transient behavior, particularly during acceleration, deceleration, and direction reversal phases. These results show that LLMs can generate feasible and performance-optimized motion trajectories that enhance both control accuracy and energy efficiency in Cartesian tooth-belt drive systems.
To achieve this, an LLM-based motion generation framework was developed using structured prompt engineering with Anthropic Claude Sonnet 4.6. The generated trajectories were implemented in MATLAB/Simulink and compared with traditional motion profiles in a cascaded control architecture that consists of position and velocity loops with the dynamic model of the Cartesian tooth-belt drive system. The performance was evaluated using a detailed simulation model by analyzing position and velocity tracking errors, torque response, and energy consumption.
The simulation results showed that the LLM-based trajectory showed an improved performance compared to the traditional method. It achieved lower position root mean square (RMS) error with 10.25 % improvement, lower velocity RMS error (17.42 % improvement), reduced RMS torque (35.95 % improvement), and absolute peak torque improvement of 68 %. The total energy consumption was improved by 10.59 %, thereby demonstrating an improved energy efficiency of the system. The LLM-based profile also produces smoother motion with reduced dynamic excitation and improved transient behavior, particularly during acceleration, deceleration, and direction reversal phases. These results show that LLMs can generate feasible and performance-optimized motion trajectories that enhance both control accuracy and energy efficiency in Cartesian tooth-belt drive systems.