Estimating current derivatives for sensorless motor drive applications

Abstract

The PWM current derivative technique for sensorless control of AC machines requires current derivative measurements under certain PWM vectors. This is often not possible under narrow PWM vectors due to high frequency (HF) oscillations which affect the current and current derivative responses. In previous work, researchers extended the time that PWM vectors were applied to the machine for to a threshold known as the minimum pulse width (tmin), in order to allow the HF oscillations to decay and a derivative measurement to be obtained. This resulted in additional distortion to the motor current New experimental results demonstrate that an artificial neural network (ANN) can be used to estimate derivatives using measurements from a standard current sensor before the HF oscillations have fully decayed. This reduces the minimum pulse width required and can significantly reduce the additional current distortion and torque ripple.

Improvement of low‐speed sensorless control with multi‐layer neural network

“…The use of neural networks (NNs) for position sensorless control to deal with nonlinear elements and to improve control performance has been explored 20–33 . In References 20,21 dealing with induction motors, magnetic flux observers combined with NNs are used for speed estimation with regard to temperature changes and parameter fluctuations, and improvement of control performance is reported.…”

“…In addition, References 32,33 consider the use of current slope (derivative) during voltage vector application in low‐speed sensorless control based on saliency propose; specifically, a method is proposed to use NN for removal of detection noise in standard Hall current sensors caused by switching. High‐frequency noise occurs during inverter switching, and this noise superimposes on current sensor signals as well.…”

“…The present paper explores application of noise removal method proposed in references 32,33 to three‐shunt current sensing widely employed in home appliances and other applications, and proposes a method to remove switching noise from a three‐shunt current detector by means of a NN, without any special modulation. Some problems arise when the mentioned noise removal technique using NNs is applied to three‐shunt current sensing.…”

“…Some problems arise when the mentioned noise removal technique using NNs is applied to three‐shunt current sensing. References 32,33 assume a Hall CT; thus, current noise contains mainly high‐frequency ringing components determined by the circuit pattern, magnitude of stray components in IGBT or other power devices, and dv/dt in switching. On the other hand, with three‐shunt current sensing, detected current is a pulsed signal that switches between zero and current magnitude depending on whether a power device is on or off.…”

“…Shifting current detection away from a point immediately after switching is helpful to avoid the influence of slope change, but this results leads to increase of voltage vector period needed for sensing. Besides, in References, 32,33 three current derivatives are necessary for pole position estimation, and two kinds of active voltage vectors must be provided in one carrier cycle; thus a region in which detection is impossible occurs in case of operation at low or high modulation factor, same as in the current detection methods shown in References 34–36 . This problem, too, can be handled by using a peculiar modulation method, but the problem of increase in high‐frequency noise remains after all.…”