Model identification of a cable-based mechanical transmission for robotics using a Best Linear Approximation approach

This article presents an experimental approach for frequency-based identification and excitation signal design to be used for identifying the mechanical motor-to-joint cable-based transmission of a robotic actuation system. By using the Best Linear Approximation (BLA) theory, frequency responses are estimated and nonlinear distortions are mitigated. The identification process is performed in closed-loop to ensure system stability, with multisine excitation signals designed to respect system constraints. The paper presents a technique for enhancing the accuracy of estimated frequency responses when dealing with signal saturation in a closed-loop system. This approach involves shaping the multisine excitation to enhance the Crest Factor of the control signal. The study includes experimental results from non-parametric frequency identification conducted on a test bench, and a parametric model is derived using a non-convex optimization method. • Frequency-domain identification catches the flexible modes of a flexible transmission. • A Best Linear Approximation is obtained despite of noise and nonlinear distortions. • Multisine signal is an efficient excitation in a closed-loop context. • Physical parameters are precisely obtained by minimizing a H-infinity criterion.