Sensorless Parameter and Current Estimation with Adaptive Control for PMDC Motor Deficiency Compensation

Abstract

Accurate mathematical modeling of permanent magnet direct-current (PMDC) motors is limited by incomplete motor data, aging effects, and unit-to-unit parameter deviation, motivating the need for a reliable and computationally-efficient estimation framework. Conventional schemes rely on current sensors, which increase cost and complexity, or utilize joint nonlinear observers that become rank-deficient under speed-only measurements. This thesis proposes a currentless dual online estimation framework, which couples a Kalman filter (KF) with a modified instrumental variable recursive least squares (MIV-RLS) algorithm, to simultaneously estimate individual motor parameters and the electric current using \textit{only} speed measurements. During motor operation, the model is continuously updated. The framework compensates for parameter drift, maintains optimal performance under varying conditions, minimizes controller effort, adapts to system degradation, and provides early warnings of potential failures. Furthermore, the estimator is integrated with a modified indirect model-reference adaptive controller (IMRAC), which computes control actions in real time to achieve robust trajectory tracking despite parametric uncertainties. The proposed framework is tested on a low-cost test rig comprising eight identical motors and a mockup two-degree-of-freedom (2-DOF) robotic manipulator arm to demonstrate trajectory tracking. Simulation and experimental results demonstrate fast (<1.0 s) and accurate convergence of the estimation frameworks, and that the proposed adaptive controller outperforms the traditional baseline IMRAC, achieving a 40.1 % reduction in tracking RMSE (4.02◦ vs. 6.71◦) under a 90◦ sinusoidal reference with a 2.5 s period. This work presents a complete framework for PMDC motor state monitoring and position control. Future work could integrate fault-tolerant control (FTC) or explore Extended-RLS (ERLS) to integrate nonlinearities in the estimation problem.