In this work, a novel predictive control scheme for the permanent magnet synchronous machine by using a pulse-width modulation technique of variable switching frequency and variable sampling time is proposed. The control scheme is based on the model-based predictive control theory and makes use of a flexible pulse width modulation stage that can be reconfigured during the operation of the system without stopping the execution.
The PI-controllers conventionally used in field-oriented control schemes are modified so that they can be used during the prediction stage of the control with different sampling times, without saturating the integral part of the controller. A finite set of switching frequencies is used for the optimization of a cost function, which aims to minimize not only the error in the torque and flux producing components of the currents of the space phasor of the PMSM. In addition, the cost function includes the reduction of the switching losses by directly reducing the switching frequency and maintaining a minimum level of desired performance mainly regulated by the maximum allowed torque ripple. The oversampling of the reference signals allows the adaption of the sampling frequency in order to avoid delays in the reaction of the system in case of transients.
The proposed control scheme demands a high degree of flexibility in the implementation and a fast signal processing. Therefore, a FPGA implementation was mandatory that allows the realization of most of the control algorithm in silicon and a parallel execution. The acquisition of analogue control signals is carried out by means of Delta-Sigma ADCs with a very short conversion time that permits an oversampling and a signal processing with a high time resolution. As a result, the control features not only an excellent dynamic behavior with high bandwidth allowing to consider the harmonics of the current i.e. the switching ripple.
The proposed predictive control approach, successfully overcomes the majority of the drawbacks of conventional field-oriented control and eliminates some of the drawbacks of conventional model predictive control, delivers good dynamics in the torque behavior and exploits the parallel processing capabilities and high computational power provided by an FPGA implementation. A sensorless implementation complements the proposed predictive control strategy. The faster processing capabilities of the FPGA allow the sensorless operation of the drive at very low speeds without the need of additional signal injection.