Recursive Identification of Electric Drives using Gray-Box Models

Universal electric motor drives should have the capability to tune the drive control system to drive different loads and maintain system performance for a wide range of loads. Gray-box modeling using neural networks is presented here as a possible solution for the identification of electric motor drive systems. In the proposed gray-box modeling, the drive system is divided into the known part governed by the physical laws, which in our case is the electrical subsystem, and an unknown part, which in our case is the mechanical subsystem. The mechanical part is modeled with a black box model using a radial basis neural network. Linear regression models are proposed for the mechanical and electrical subsystems and used to design a two-stage parameter estimation algorithm based on linear least squares. Simulation and experimental results are presented showing the potential of the approach.