Neural network speed controller for drive system with elastic joint

This paper presents application of neural network for speed control of drive system with elastic connection. Analyzed object is characteristic due to complexity of the mechanical part of the drive. Motor machine is connected with load using long elastic shaft. Such form of electromagnetic torque transmission leads to appearing of oscillation in state variables transients. In result precise control of the speed or position is difficult to obtain. In article application of neural network trained on-line is applied in speed control loop. Weights coefficients are recalculated according to backpropagation algorithm based on error from reference model placed in control structure. Several simulation tests are presented. Obtained results presents quality of control using described neural model. In addition robustness against parameter changes is shown. Moreover influence of introduction of nonlinear elements (backlash) on achieved results is analyzed. Simulations are verified experimentally on laboratory benchmark. Control algorithm is implemented in dSPACE 1104.