Comparison of linear observation techniques for robust load torque estimation in actuators

TL;DR

The paper tackles robust estimation of external load torque in rotary actuators when only motor drive torque and shaft displacement are available. It develops and compares two linear observers—the reduced-order Luenberger observer and the disturbance observer (DOB)—using a common PMSM-based actuator model and validates them through a detailed experimental setup with a torquemeter as reference. By analyzing the observer open-loop transfer functions and stability margins under parametric uncertainties, it finds that DOB can be more sensitive to model perturbations and may suffer reduced gain margins, whereas the reduced-order Luenberger observer exhibits greater robustness with a high phase margin. The experimental results corroborate the theoretical findings, showing that both observers can track load torque effectively, with the Luenberger observer providing slightly better estimation accuracy in the reported case, guiding practical observer selection for load-torque compensation in actuator control.

Abstract

The paper addresses the problem of estimating robustly the external load torque in rotary actuator systems, when only the generated motor drive torque and angular displacement are the available input and output. We compare, theoretically and experimentally, two sufficiently established linear observation techniques (i) reduced-order Luenberger observer and (ii) disturbance observer, both using the same identified model of a permanent magnet synchronous motor (PMSM)-based actuator. Our goal is to highlight several aspects related to the implementation, relative degree of the input-torque to estimated-load-torque transfer characteristics, observer open-loop transfer function, and the associated sensitivity (respectively stability margins) with respect to inherently uncertain system plants. Apart from the developed analysis, a detailed experimental case study is demonstrated where the load torque sensor provides reference measurements and allows for evaluation of both observers.

Figures (5)

• Figure 1: Experimental setup (laboratory view).
• Figure 2: No-load test for friction identification: (a)-(b) measured rotor speed and electromagnetic torque during the test, (c) data fitting results.
• Figure 3: Block diagram of the load torque observer in an open-loop.
• Figure 4: Measured and estimated (with $\Omega_L,\, \Omega_D$ observers) load torque from the experiments with the controlled speed about 10.5 rad/sec.
• Figure 5: Zoom-in of the measured and estimated (with $\Omega_{\{L,D\}}$ observers) load torque from the experiments as depicted in Fig. \ref{['fig:TorqueExp']}.