FPGA implementation of linear model predictive controller for real-time position control of DC motor

Abstract

Model predictive control (MPC) is a predictive class of control algorithm which has been widely adopted for slow dynamic, constrained, multi-variable process control problems. The applicability of MPC for real-time control is restricted due to the associated computational complexity of quadratic programming (QP) problem and the choice of appropriate computing platforms. MPC can be formulated as a standard QP problem and according to receding horizon control strategy, in order to get optimal control signal to apply to the system, in each discrete time step, a QP problem is to be solved online. In order to meet high sampling frequency demand of real-time control applications, in this paper, we propose a novel accelerated field programmable gate array (FPGA)-based customised architecture for active set method (ASM)-based QP solver to implement linear MPC. The approach is implemented successfully on Xilinx's Virtex-4 FPGA and demonstrated practically by applying MPC to Quansers QET DC servomotor position control plant. From experimentation; it is observed that, FPGA-based implementation occupies 35% of Virtex-4 area and achieves 500 µs sample time.