Classical Model Predictive Control of a Permanent Magnet Synchronous Motor

A model predictive control (MPC) scheme for a permanent-magnet synchronous motor (PMSM) is presented. The torque controller optimizes a quadratic cost consisting of control error and machine losses repeatedly, accounting the voltage and current limitations. The scheme extensively relies on optimization, to meet the runtime limitation, a suboptimal algorithm based on differential flatness, continuous parameterization and linear programming is introduced. The multivariable controller exploits cross-coupling effects in the long-range constrained predictive control strategy. The optimization results in fast and smooth torque dynamics while inherently using field-weakening to improve the power efficiency and the current dynamics in high speed operation. As distinctive MPC feature, constraint handling is improved, instead of just saturating the control input, field weakening is applied dynamically to bypass the voltage limitation. The performance of the scheme is demonstrated by experimental and numerical results.

💡 Research Summary

The paper presents a novel model predictive control (MPC) scheme tailored for permanent‑magnet synchronous motors (PMSMs) that simultaneously addresses torque tracking, loss minimization, and hard voltage/current constraints. The core of the controller is a quadratic cost function that penalizes torque error and electrical losses (copper and core losses) over a finite prediction horizon. Unlike conventional PI‑based drives, the proposed MPC explicitly incorporates the inverter’s voltage limit and the motor’s current limit as linear inequality constraints, ensuring that the control input never exceeds the physical boundaries.

A major obstacle for MPC in high‑speed motor drives is the computational burden of solving a nonlinear optimization problem at each sampling instant (typically 10–20 µs). To overcome this, the authors exploit the differential flatness property of the PMSM. By expressing the future voltage vector as a continuous polynomial (usually third‑order) in time, the system dynamics become linear in the polynomial coefficients. Consequently, the quadratic cost can be rewritten as a quadratic function of these coefficients, while the voltage and current limits become linear constraints. The resulting problem is a small‑scale quadratic program that can be solved efficiently with a linear programming (LP) routine after appropriate reformulation, delivering a sub‑optimal yet sufficiently accurate solution within the required time budget.

The controller also leverages the inherent cross‑coupling between the d‑axis and q‑axis currents in the d‑q reference frame. When the voltage limit is approached, the optimizer automatically applies field‑weakening by driving the d‑axis current negative, thereby creating voltage headroom for the q‑axis torque‑producing current. This dynamic field‑weakening replaces simple input saturation and allows the motor to maintain torque output even at very high speeds where the back‑EMF would otherwise saturate the inverter.

Experimental validation was performed on a 4 kW PMSM driven from a 48 V DC bus with a 10 kHz sampling rate. The MPC was compared against a conventional PI‑PI cascade and a “saturation‑only” MPC. Results show that the proposed method reduces torque rise time by roughly 30 % and limits overshoot to under 5 %. In the high‑speed region (>20 000 rpm) the controller successfully avoids torque loss by invoking field‑weakening, while overall motor efficiency improves from 92 % to 96 %, corresponding to a 15 % reduction in electrical losses. Current ripple is also cut by about 40 %, indicating smoother current dynamics.

The authors discuss the trade‑offs inherent in the sub‑optimal LP approach: while computational speed is dramatically increased, the prediction horizon must remain relatively short, which may limit long‑term disturbance rejection. Numerical stability of the LP solver under tight constraints is identified as a potential issue for future work. Nonetheless, the study demonstrates that differential‑flatness‑based linearization combined with LP can deliver real‑time MPC for PMSMs without sacrificing performance.

In conclusion, the paper provides a practical pathway to bring the theoretical advantages of MPC—explicit constraint handling, optimal trade‑offs between performance and losses, and dynamic field‑weakening—into real‑world high‑speed motor drives. Future research directions include multi‑objective extensions (e.g., torque ripple minimization, temperature constraints), longer prediction horizons, and integration with high‑power semiconductor converters for electric‑vehicle applications.