Improved gray-box modeling of electric drives using neural networks

Electric drives are used in many industrial and commercial applications. High performance control of electric drives requires the accurate modeling of the motor and mechanical load. In many industrial applications, it is desirable that the electric drive has the capability of self-tuning controller parameters to be able to drive different mechanical loads. One way to achieve this flexibility is by direct identification of the drive and mechanical load. Modeling and identification of Electric drive coupled to a load can be a challenging task. This research investigates the use of gray box models to identify electric drive systems connected to an unknown load.; In the proposed model, the electrical subsystem of the machine is modeled using physical principles while the mechanical subsystem is modeled using a black box model based on neural networks. A two-stage identification approach that separates electrical subsystem parameter estimation from mechanical subsystem identification is presented. At each stage the parameters are estimated using the linear least squares approach. Simulation results are presented to demonstrate the feasibility of the approach.