Emulation of nonlinear mechanical loads using multi-layer neural networks

Abstract

This study describes the torque control of a vector controlled load machine (dynamometer) mechanically coupled to a drive machine for the emulation of nonlinear loads. Proposed dynamometer control strategy is based on model reference control using an on-line trained Multi-layer Neural Networks (MNN). The emulation is involved in the closed loop speed control of the drive machine. After the training of the neuro-controller, the drive machine will see the desired nonlinear mechanical load. An integral compensator supporting the trained MNN is used for eliminating or reducing the model tracking steady state errors. Training problems of the MNN in drive systems are also discussed. Variety of load models which are the nonlinear function of the speed, friction and inertia are successfully emulated and the generalization capability of the trained MNN is tested for various reference inputs. Simulation results showing the excellent dynamometer control performance are presented.