Research On Observer Design For Nonlinear Actuator Systems

In a system, the input and output signals are external variables and the internal status is thevariable of system movement reaction.In order to solve the control system problems, we need toobserve its internal variable .Since the system status can’t be observed directly, it’s not easy to getits internal information.So proper status feedback is introduced, reconstructing the status to replacethe real system.Researching the system status reconstruction problem is the the observer design problem,reconstructing a new system in which the original inpout and output signals measured directly workas the input to make the new system output gradually convergence to the origianal system withsome special conditions.Based on the principle of pneumatic film actuator modeling, the Matlab/Simulink works as theplatform .And a way of designing the the observer on the basis of RBF neural network adaptive andElman neural network nonlinear in order to get an insight into the interior of the actuator systemstate based on the study of both partial and the whole characteristics.First establish observer model for pneumatic actuator on neural network, in which there aretwo neural networks approaching the unknown non-linear parts. And then build MATLAB/Simulinksimulation model, using standard error feedback algorithm to adjust the network weights online.Finally simulation faults on the simulation platform, from the most common fault, such as the faultof film perforation and spring faults, experiments show that network model can be effective to thesystem, so as to achieve the objective of fault detection.Second based on the dynamic and complex nonlinear system, we use the Elman dynamicfeedback neural network to build observer model. The main steps as follows: first, constructnetwork model, including determine the input/output layer nodes and hidden nodes and selectionsamples. And then we chose LM algorithm after comparing several improved algorithms. Finally,after multiple tests include determine learning rate and momentum constant, we get the bestnetwork model by adjusting relevance weights.The last major is to validate the Elman of neural network constructed. Through observing theinternal state variable of pneumatic actuators include gas pressure of thin film , gas mass flow fromrelay to thin film, output back- pressure of torque motor and Stem displacement. The comparisonresult between the system actual value and the model observation value, which show that Elmannetwork observer model can reflect the system state truly, and the error analysis of the observationalcomparison proved the feasibility and practicability of the model.