摘要：AbstractThis paper presents a study on the practical utilization of various microcontroller units (MCU) for model predictive vibration control, using an online active-set quadratic programming solver. The implementation properties of various 32bit MCUs are assessed in experiment. An analysis of memory requirements is made on all investigated MCUs, along with evaluating timing requirements for a simple vibration control problem. These tests show execution time limits, memory footprint, and convergence under different circumstances. The experiments investigate the effects of hot-starting, numerical precision and the use of the floating-point unit (FPU) of the MCU. The results presented in this paper show that the hot-starting technique used with double precision data format is preferable, since computation using single precision numeric format fails to converge on many instances. Although it is tempting to reduce numerical precision to extend the maximal horizon stored in the volatile and nonvolatile memories and to cut execution times, the numerical stability of the investigated online active-set quadratic programming solver is heavily affected. Moreover, it has been found that a built-in FPU does not have significant impact on performance; the presence of this hardware feature does not significantly aid the implementation of model predictive control for the case study presented here.
