Implementation of Low Speed Sensorless Control using ANN on General-Purpose MCU

In sensorless control of PMSM(Permanent magnet synchronous motor), the INFORM(Indirect Flux detection by On-line Reactance Measurement) method detects the current slope amount during PWM based on the active voltage vector and can realize position estimation at a low speed without applying a high-frequency voltage and is therefore effective for low noise. However, when the frequency of the PWM carrier is high with the INFORM method, the detected current difference will contain a lot of switching and sampling noise and outliers. This paper reports on the current sampling method, the configuration of the experimental system, the principle of the magnetic pole position estimation method using ANNs(Artificial Neural Networks) for noise suppression, and results of validation using an actual machine.