Data-Driven Optimization of Hysteresis Compensation Techniques for Magnetostrictive Systems

Abstract

A range of sophisticated sensors, especially sensors measuring force or torque using magnetic principles exhibit undesirable "memory" effects called hysteresis. Hysteresis describes an effect in which measurements from the sensor depend not solely upon the current measurement of force or torque, but also on the history of measurements which can be confounding for its accuracy and reliability. This thesis proposes a practical method that can reduce hysteresis-induced errors in magnetostrictive torque sensors using a data-driven approach that allows for simple implementation on cheaper common microcontrollers at a real-time response. The core of the thesis involved the acquisition of an accurate mathematical model that captures the unique hysteresis properties of the system. This was achieved by conducting a thorough data analysis on experimental data acquired from the sensor under different conditions, followed by analyzing and refining the model parameters using computational optimization algorithms. Once the hysteresis was accurately modeled , a computationally efficient compensation algorithm for this use-case was implemented. This algorithm acts like a filter by taking in the raw, hysteretic sensor readings and correcting them in real-time to provide a more accurate output. The algorithm was implemented on a cheap ESP32 microcontroller and was validated to be able to operate in real-time requirements. Through testing, including operating the ESP32 in real-time feedback, we were able to validate that the technique proposed in this thesis was effective. The system with the compensation was able to reduce measurement error introduced by hysteresis by roughly 40%-50% indicating significant potential to improve the use of magnetostrictive sensors.While some complex material behaviors remain a challenge, this work provides a foundation for creating more accurate systems using data-driven compensation strategies.