Practical Offset-Free Model Predictive Control and Its Embedded Application to Aeroengines

Model predictive control (MPC) is popular in applications with slow dynamics because of its advantages in handling constraints and multivariable optimization. But for aeroengines, it is difficult to obtain an exact prediction model, which will lead to offsets in tracking. Besides, deploying MPC to embedded controllers for real-time control is a well-known challenge. Therefore, this paper presents a switched linear MPC, which incorporates the augmented prediction models with error integrator for offset-free tracking, the sparse-based quadratic programming formula for solving MPC, and a reset strategy for achieving bumpless transfer at the switching instant. Further, on the hardware board we developed, six hardware-related acceleration strategies are explored and evaluated for real-time performance. Then, eight cases of five objects are tested, whose results indicate a significant speedup of around 50 times. At last, the hardware-in-the-loop tests of the turbofan engine and the real bench tests of the micro-turbojet engine are performed, which verifies the superiority, real-time performance, and potential for practical applications.
Note to Practitioners
—This paper was motivated by the problem of applying the offset-free MPC to the embedded control system for aeroengines. The difficulty of accurate modelling and the requirement to compute the embedded MPC in a limited time are two major challenges. Accordingly, we explore six hardware acceleration strategies that are often ignored to ensure the real-time performance. Moreover, we investigate the switched MPC to obtain the desired dynamic response. However, the controller switching tends to induce fluctuations that are harmful to the engine. Hence, we present a reset strategy to ensure bumpless transfer performance. Preliminary physical experiments suggest that the proposed approaches are feasible to achieve the predetermined objectives. Further, the comparisons with the previous methods highlight our superiority. Overall, this paper contributes to optimize the computational performance of microcontrollers and enhance bumpless transfer performance for MPC. In our future work, the proposed approaches will be applied to flight experiments and more industrial devices.