Date of Award


Degree Name

Master of Science in Engineering


Electrical and Computer Engineering

First Advisor

Dr. Johnson Asumadu

Second Advisor

Dr. Ralph Tanner

Third Advisor

Dr. Massood Zandi Atashbar


Electric drives, direct torque control, neural network, inverters, Simuliink model

Access Setting

Masters Thesis-Open Access


In conventional direct torque control (DTC) scheme of induction motor (lM), Proportional Integral Controller (PI) is used as the speed controller. PI controller is more suitable in steady state condition and for linear system but both DTC and IM are mostly nonlinear. Multi-Layer perceptron neural network. (MLPNN) controller is more suitable and performs better than PI controller. The switching table of the conventional DTC is replaced by the MLPNN controller. The MLPNN inputs are, the magnitude of the stator flux, torque and the Voltage sectors. Levenberg-Marquardt back propagation technique has been used to train the MLPNN. The output of the MLPNN are the gate signals to the inverter, which switch to produce the required voltage vectors. The Hysteresis band for torque is l% and the hysteresis band for the flux is 0.5%.The advantage of using MLPNN is that, in classical DTC look up table is used to select switching states, thus size requirement will increase with some advanced control. MLPNN willtake less memory and is more reliable. Despite the fact that the proposed scheme does not eliminate the torque, flux and current ripples to a great extent, the dropping of the stator flux due to the sector changes is eliminated. Also a faster stator flux and Torque response is achieved in the transient state.