Neural network based electromagnetic torque identification of squirrel cage induction motor

In this paper, an electromagnetic (EM) torque identification scheme of squirrel cage induction motor (SCIM) using recurrent neural network (NN) model is developed. The searched parameters are estimated by using Levenberg-Maquart (LM) training algorithms, meanwhile its structure is optimized involving the Akaike final prediction error (AIC-FPE) criterion. To this end, a simulation model is elaborated to collect the required training input-output data. The benefits of this identification method are manifold: it is online, it is based on the measured data only, it takes into account the saturation and transient behavior, and it is very suitable for predictive control. The usefulness of the developed method is examined by simulation within an automotive application.