Neural Speed Estimation Applied to Stator Flux-Oriented Control Drives

The induction motor speed is an important quantity in an industrial process and can be used indirectly in flow, pressure, and drive control. However, the direct measurement of speed compromises the driver system and control, increasing the implementation cost. Speed estimators are usually based on a mathematical model of the induction motor, but it is typically necessary to obtain the parameters of the motors. Thus, this work proposes an artificial neural network approach to estimate the mechanical speed of induction motors in a stator flux-oriented vector control by direct current control and direct torque control. In this proposed strategy, no machine parameters adaptation is needed. The neural speed estimators, without weight change, are tested in two different motors to evaluate its robustness. First, by simulation, the neural networks are trained with constant machine parameters and then the estimator performance is evaluated under stator and rotor resistance variation. After that, the same neural estimators are experimentally tested, with another machine, under the variation of motor load torque and speed reference operating point.