U-Model Based Robustness Analysis And Control Research For Nonlinear Plant

In modern industrial processes, for higher products quality and production rate, thesystems usually exhibit nonlinearities, time-varying, unpredictable parameters and othercomplexities. In general, dynamic characteristics of the systems cannot be describedexactly only by linear system methodologies. The research on model based control of thenonlinear dynamic plants has become an important research field. However, the primaryproblem in the research is that there is not a general model structure describing thenonlinear dynamic plant accurately, or the model structure and parameters are toocomplex. In this paper, a class of random nonlinear dynamic plant is analyzed, identifiedand controlled base on the control-oriented model (U-model) as follows:The U-model based control-oriented structure is presented firstly with correspondingterms defined. The Newton–Raphson algorithm is used and modified to find the solutionof U-model. System robustness is analyzed: U-model learning arithmetic is studied baseon feedback frame, the stability of which is analyzed using small gaim theory; appropriatelearning rate is set for system robustness under noise; the range of optimal learning rate isgiven to speed up the convergence.Sencondly, for single input single output (SISO) random nonlinear dynamic plant, aU-model based adaptive ploe placement controller is designed using adaptive law andpole placement method. Then U-model adaptive PID controller is designed using the ploeplacement criterion. And a U-model based radial basis function neural network (RBFNN)adaptive-tracking controller is designed using adaptive law and RBFNN. Simulations aredone finally.For many input many output (MIMO) random nonlinear dynamic plant, a newlearning feedforward controller is designed as follows: MIMO U-model structure isgotten from the SISO one; for unknown pamameters the U-model are updated onlineusing RBFNN; A U-model based controller is designed using Newton–Raphsonalgorithm. Simulations are done with constant or adaptive learning rate.Finally, a class of typical random nonlinear dynamic plant-double-wall welded pipe welding temperature process is presented. The steady-state model is built based onwelding temperature analysis. Influenced by kinds of unkown factors, the plant can besketched by U-model exactly and easily. So the steady-state model can be transformed tothe U-model frame. Then the U-model based adaptive controller is designed and pipewelding temperature control platform is built. The communication is implementedbetween Matlab and Kingview base on DDE technology. Finally, the validity of controlmethod is proved by pipeline welding temperature control experiment.