DSP-based direct neural control for thrust active magnetic bearing learned through adaptive differential evolution

A digital signal processor (DSP)-based direct recurrent wavelet neural network (RWNN) controller is proposed to control the rotor position of a thrust active magnetic bearing (TAMB) system learned through adaptive differential evolution (ADE). First, the dynamic analysis of the TAMB with differential driving mode (DDM) is derived. Subsequently, due to the exact dynamic model of TAMB system is absent; a RWNN is adopted to deal with the highly nonlinear TAMB system for the tracking of reference trajectory. Moreover, Due to the gradient descent method is used in back propagation (BP) to derive the on-line learning algorithm for the RWNN; it may reach the local optimal solution due to the inappropriate initial values. Therefore, an ADE algorithm is adopted to optimize the initial network parameters including connective weights, translations and dilations for the RWNN controller. Finally, a DSP with PowerPC 440 processor and real time VxWorks OS is used for implementing the RWNN-ADE controller for TAMB system. Experimental results show the high-accuracy control performance of the proposed RWNN-ADE controlled TAMB system.