INTERNATIONAL CUMHURIYET ARTIFICIAL INTELLIGENCE APPLICATIONS CONFERENCE CONFERENCE PROCEEDINGS 2021, Sivas, Türkiye, 03 Aralık 2021, ss.6-11
Abstract— It is very difficult to completely define a nonlinear electric motor such as Switched Reluctance
Motor (SRM) by its mathematical model. For a threephase 6/4 SRM, current data varying according to the
rotor position were obtained from the phase windings by the pulse injection method applied to the sensored
driven motor. By measuring the phase currents changing according to the rotor position, the input data
of the model and the active phase output data were obtained. The relationship between these current values
and the rotor position was analyzed by machine learning algorithms and the rotor position was predicted. Tuned
Fine Tree and Ensemble Bagged Tree algorithms were determined as the most successful algorithms in the
study, in which time series approach was used to increase model success. The motor was driven at
constant speed and load to keep the model size small. Data is not preprocessed to reduce microcontroller load,
shorten processing time, and facilitate real-time operation. This approach eliminated the premathematical processing load and contributed to the
estimation speed as high as possible. In this study, it has been shown that machine learning algorithms used in
modeling nonlinear systems can be used to model electric motors with similar structures and can increase
model performance with time series approach.
