Hybrid Stochastic and Neural Network Approach for Efficient FPGA Implementation of a Field-oriented Induction Motor Drive Controller

The FPGA (field programmable gate arrays) is concurrent, executing all its logic in parallel, therefore is good for neural network applications that are characterized as heavy parallel calculation algorithms. The stochastic arithmetic can simplify the computation elements and is compatible with modern VLSI design. This paper presents an efficiency approach for a single FPGA to implement the field-oriented control of induction motor drive based on stochastic theory and neural network algorithm. A stochastic neural network structure is proposed for a feedforward neural network to estimate the feedback signals in an induction motor drive. A new stochastic PI speed controller is developed with anti-windup function to improve the speed control performance. By applying the stochastic theory and neural network structure, the proposed algorithms enhance the arithmetic operations of the FPGA, save digital resources, simplify the algorithms, significantly reduce the cost and provide design flexibility and extra fault tolerance for the system. A hardware-in-the-loop test platform using real time digital simulator (RTDS) is built in the laboratory. The experimental results are provided to verify the proposed FPGA controller