Abstract
Induction motors exhibit non-linear behaviour and are difficult to model. Furthermore, load disparities cause speed variations and therefore predicting the motor’s response is challenging. In this paper, feedforward neural networks (FNNs) are used to model induction motors at two load levels (i.e. no-load and full-load). The two FNN models are then transformed into auto-regressive moving-average (ARMA) models using a new NN2TF algorithm. A weighted parametric model is then formulated by combining both ARMA models to provide appropriate transfer functions at different loads ranging from no-load to full-load. In order to validate the developed model, experimental data (with voltage/speed as input/output) is collected from an induction motor plant at five different load levels and used to test the proposed model. Simulation results show that the estimated model produced dynamic responses that follow the experimental data with good accuracy, regardless of the load.
Original language | English |
---|---|
Pages (from-to) | 1645-1658 |
Number of pages | 14 |
Journal | Transactions of the Institute of Measurement and Control |
Volume | 40 |
Issue number | 5 |
DOIs | |
Publication status | Published - Mar 1 2018 |
Keywords
- ARMA models
- induction motors
- neural networks
- robust design
- system identification
ASJC Scopus subject areas
- Instrumentation