M.S. (Master of Science)
Department of Mechanical Engineering
Robotics||Artificial intelligence||Mechanical engineering
Controller design of a nonlinear system is in general very difficult. One way to avoid such complexity is using a simplified model so that certain nonlinear control techniques can be easily applied. Using a linearized model could make the controller design even simpler. However, some control error is inevitable with a simplified model. Therefore, in this thesis, a neural network-based approach is proposed in order to compensate for the errors caused by using a simplified dynamic model. The base controller which is designed by using the simplified dynamic model will be compensated by a PID controller with adjustable gains. A neural network is used to update the PID gains during control process. Finally, the outputs of the NN-based PID compensator and the base controller are added together to control the actual nonlinear system. This way, the NN-based PID compensator tries to compensate for the effects of the ignored nonlinear terms of the dynamic model. The performance of the proposed control method is verified on the ball-on-plate system that is built for this study. Approximate feedback linearization is applied as the base controller on a simplified decoupled dynamic model. A NN-based PID compensator is added to each decoupled ball-on-beam system. Experimental results that show better stabilization and trajectory tracking performance are provided and discussed in the thesis.
Mohammadi, A., "Neural network-based PID compensation for nonlinear systems ball-on-plate example" (2018). Graduate Research Theses & Dissertations. 4107.
Northern Illinois University
Rights Statement 2