Virtual speed sensor for DC motor using back-propagation artificial neural networks

MONTESDEOCA, JULIO; ZAMBRANO, JULIO; MORALES, JOHN; CABRAL, R.J.

Conferencia; 2014 IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC); 2014

The use of artificial neural networks has enabledapplications that would be impossible to achieve with conventionalelectronics, through to versatility that them have, they canbe configured as needed and use as required such as classifiers,adaptive filters, controllers and predictors.This paper describes a experimental implementation of virtualspeed sensor for DC motor using back-propagation artificialneural networks through voltage and current acquisition. Inimplementation process a comparative between both logsig andtansig activation function in hidden layer is made.The training of the neural network is performed by acquiringsignals of voltage, current and speed sensor, later the responseefficiency of the network is analyzed by comparing the resultproduced by physical sensor and virtual sensor implementedusing neural networks. In this experiment training pattern valueshas a large range, a several process training with both small andlarge section range is made.Results show a good performance and a minimum error withboth logsig and tansig activation function in hidden layer whenpattern signal is full training.