Experimental validation of MRAS sensorless direct torque control using ANN for induction motors in a pumping system

The use of direct torque control (DTC) for electric motor applications provides many benefits, such as quick dynamic response and efficient operation at low speeds. However, it does come with some disadvantages (i.e., torque ripple and an additional expense with speed sensors) that could complicate the system and add to the overall costs associated with the system. Here, we present an approach that uses an innovative method to eliminate the need for speed measurement to employ DTC on an electric pump driven by an induction machine using an asynchronous squirrel cage method. We achieved this through utilizing a model reference adaptive system (MRAS) in conjunction with artificial neural networks (ANN). Using MRAS enabled real-time measurements of motor speed, reducing complexity in the motor model and improving the computational load. However, adding MRAS to DTC does introduce some level of estimation error and incurs additional torque ripples; we have incorporated the use of an ANN to assist in optimizing the switching schemes that mitigate these adverse effects. In our experimental study, we utilized MATLAB/Simulink for rigorous testing and validated our results on a dSPACE 1104 board. The improvements in response time from 0.8 to 0.58 s (DTC to DTC-MRAS + ANN) and a reduction in total harmonic distortion (total harmonic distortion reduced from 9,76% to 7,80%) provide an innovative new perspective to the current literature, demonstrating both the repeatability and effectiveness of this approach while producing significant improvements in control methods compared with both DTC-MRAS and DTC-ANN techniques by numerous control indices.