Research On Online Estimation Of Disturbances And Intelligent Control For Linear Motor

Permanent magnet linear synchronous motor (PMLSM) is the specific type of linearmotors used to generate linear motion directly without any mechanical buffer units, such asgear boxes, chains and screw coupling. So compared with the rotating machine, thePMLSM is now receiving increased attention in the numerical control machine because ofthe advantages of simple structure, high speed, high precision and high stiffness. However,in the precision motion control of PMLSM，the presence of the force ripple and friction is ahighly nonlinear phenomenon, which will seriously influence the tracking performance ofthe PMLSM without any mechanical buffer units. So it is significant to erase thedisturbances in the PMLSM. Aiming at the force ripple and friction, the online estimationof the nonlinear disturbances and intelligent control for the PMLSM is taken as theresearch object. The major results achieved are as follows:(1) To erase the nonlinear force ripple in PMLSM, a method of the online estimationand the control for the force ripple based on wavelet neural network is proposed.According to the divide-and-conquer scheme, the control method is divided into two parts:On the one hand, a fixed parameter model is used as the velocity and accelerationcompensation for PMLSM. On the other hand, the wavelet neural network is used toestimate the force ripple and the imprecise model, which transforms the problem ofestimating disturbances and imprecise model into the learning of the coefficients in thewavelet neural network. By the theoretical analysis, it is proved that the selected waveletfunction can approximate the nonlinear force ripple with desired accuracy. The simulationresults demonstrate that the proposed control can improve the tracking performance androbustness without precise model of PMLSM.(2) By analyzing the characteristics between flexible sigmoid function and signfunction, the methods of modeling of non-smooth friction are proposed. Firstly, anapproximation model for static Stribeck friction model using flexible sigmoid function isproposed. On the basis of this, three modeling methods based on neural network areproposed, which are the hybrid model based on RBF neural network, the type-RBF neuralnetwork model and the hybrid neural network model, respectively. The simulation resultsshow that the hybrid neural network mdoel can not only approximate to the static Stribeckmodel effectively, but also verify the approximation and predication ability to theexperiment friction data.(3) The PMLSM control methods are carried on the PMLSM experiment platform. In the experiment, comparing the method of online estimation and control for the ripple forcebased on wavelet neural network with other control methods, such as the PID control, thecombined feedforward plus PID control and the neural network adaptive inverse control,the effectiveness of method of online estimation and control for force ripple based onwavelet neural network is demonstratedulteriorly.