[Back]

T. Wolbank, M. Vogelsberger, R.H. Stumberger, S. Mohagheghi, T Habetler, R. Harley:
"Comparison of Neural Network Types and Learning Methods for Self Commissioning of Speed Sensorless Controlled Induction Machines";
Talk: IEEE PESC (Power Electronics Specialists Conference), Orlando, USA; 2007-06-17 - 2007-06-21; in: "Proceedings of IEEE Power Electronics Specialists Conference (PESC)", (2007), ISBN: 1-4244-0655-2; 1955 - 1960.

Speed sensorless control of ac machines at zero speed so far is only possible using signal injection methods. Especially when applied to induction machines the spatial saturation leads to a dependence of the resulting control signals on the flux/load level. Usually this dependence has to be identified on a special test stand during a commissioning procedure for each type of induction machine. In this paper an autonomous commissioning method based on a neural network approach is proposed that does neither depend on a speed sensor present as a reference nor on a load dynamometer coupled to the machine and guaranteeing constant speed. The training data for the neural network is obtained using only acceleration and deceleration measurements of the uncoupled machine. The reliability of the proposed autonomous commissioning method is proven by measurement results. When comparing the resulting sensorless control performance the proposed commissioning method reaches the same level as a manual identification using load dynamometer as well as speed sensor.