36 November/December 2023 | E-Mobility Engineering typical range of the variable switching frequency is limited to 6-10 kHz. If the optimal switching frequency is selected for each operating range, a major improvement can be achieved in terms of the switching loss. There is active research on the VSFPWM strategy to improve inverter efficiency. The voltage ripple is a key design factor of the capacitor, as this has an impact on the control stability of the inverter and the voltage utilisation ratio. The problem of voltage ripples also affects the battery’s state of health. However, with this approach a separate current sensor needs to be added to the input side of the inverter, and the voltage ripple is observed through the integration of the sensed current. This involves an increase in the inverter cost. Also, it is not easy to apply this method to SiC MOSFET-based high switching frequency PWM inverters, because of the increase in the calculation time required for analogue-to-digital (ADC) conversion of the sensed current and the related signal processing. To address these limitations, the VSFPWM strategy is based on a mathematical model of voltage ripple and does not require additional hardware, such as sensors or circuits. The voltage ripple factor is extracted in advance and incorporated into the algorithm. In operation, the switching frequency is determined through the minimum calculation between the pre-calculated ripple factor and the load phase current. Because the real-time calculation is thus minimised, the calculation time is short, which makes it suitable for application to the SiC MOSFET-based PWM inverter, which operates at a high switching frequency. Machine learning One emerging technique for producing that perfect sine wave is to use machine learning (ML) for soft-switching. Softswitching minimises switching losses but it has never been successfully implemented for DC-AC systems with varying input voltage, temperature and load conditions. The ML framework has been trained on signals for motors and drivers with varying input voltage, temperature and load conditions. Once trained, the framework runs on a controller board and can double the power output for a typical inverter, or provides an increase in switching speed by a factor of up to 20. The ML constantly adjusts the relative timing of elements within the switching system required to force a resonance to offset the current and voltage waveforms to minimise switching losses. This forced-resonant soft-switching topology replaces the traditional IGBT or SiC driver with a common intelligent controller board and a specific resonant power gate module optimised for SiC or IGBT transistors. The topology is a variation of the Auxiliary Resonant Commutated Pole (ARCP) soft-switching converter design with embedded AI to solve complex switching system timing calculations dynamically to ensure accurate softswitching under changing input voltage, output load, device tolerances, and temperature changes. Adaptations are made on a cycle-by-cycle basis to minimise losses and maximise efficiency. This can reduce the losses in the iron core of the electric motors at cruising torques to increase an EV’s range by up to 12%. A 200 kW inverter reference design shows the technique reduces total system switching losses by 90% or more. That then enables switching frequencies four to five times faster than hard-switched IGBT systems and 35 times faster than hard-switched SiC and GaN systems, and requires only half the number of transistors. The higher frequency can also improve the efficiency and reduce the size of the magnetics, reducing the weight of the powertrain. In the case of an SiC-based EV inverter, increasing the switching frequency from 10 to 100 or 300 kHz creates a near-perfect sine wave without any output filter, and achieves 99% efficiency. The result is the elimination of unnecessary motor iron losses and an increased motor efficiency at low torque and low rpm. Higher switching frequencies also enable higher rpm motors that are lighter and cost less. The inverter exceeds 99.3% at 100 kHz using only three low-cost 35 mΩ SiC MOSFETs, one for each phase. The ML algorithm constantly predicts the zero voltage switching point. The controller analyses multiple Focus | Motor control A 200 kW inverter using machine learning motor control (Courtesy of Pre-Switch)
RkJQdWJsaXNoZXIy MjI2Mzk4