Sensorless DTC of induction motor using improved neural network switching state selector controller

Abstract

The paper deals with development of sensorless Direct Torque Control (DTC) system based on neural network. This network is built to solve the task of proper switching states selection based on information about electromagnetic torque and stator flux (position and magnitude) of induction motor. In fact, this technique which uses conventional switching table is not convenient for one-line and real time control for its high computation time. In order to avoid this problem a solution based on neural network is proposed. Well trained Artificial Neural Network structure can replace successfully the switching table. However, in the Neutral-Point-Clamped topology, it has an inherent problem of Neutral Point Potential (NPP) variation. In this way, a Neural Network-Direct Torque Control technique has been applied and the estimated value of the Neutral Point Potential is used, which is calculated by motor currents. This control strategy offers the possibility of selecting appropriate switching state to achieve the control of Neutral Point Potential. Simulation results verify the validity of the proposed method.

Design and Analysis of the Speed and Torque Control of IM with DTC Based ANN Strategy for Electric Vehicle Application

“…The conventional PI speed controller was replaced by an Artificial Neural Network with the goal to get optimal performances for the closed loop control schema ( Figure 5). This regulator has one input which is the error between the reference speed * and process output speed [20,21]. The output of ANN controller is applied to the entry of the process.…”

“…The output of ANN controller is applied to the entry of the process. After several tests, the architecture with two hidden layers was adopted [20,22].…”