Recursive neural network as speed controller for two-sided electrical drive with complex mechanical structure

In the paper, an adaptive structure for two-sided electrical drive is presented. The control object has a multiple-input multiple output structure, since two input signals (motor torques) are available for the controller. The assumed model of the plant is non-linear (presence of backlash) and reveals resonant nature (elasticity in the mechanical structure). The controller is a two-layer recursive neural network, which outputs are reference torques of motors of the electromechanical plant. The conducted research investigation included variety of structures of the neural controller, with different numbers of recursive feedbacks, different horizon of former measurements. Operation of this control structure with and without external anti-resonant filtration was compared as well.