Sensorless Control of Induction Motors by Artificial Neural Networks

1038

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 48, NO. 5, OCTOBER 2001

Letters to the Editor______________________________________________________
__________
Sensorless Control of Induction Motors by Artificial Neural Networks
J. R. Heredia, F. Perez Hidalgo, and J. L. Duran Paz putation and fault tolerance characteristics due to distributed network intelligence. Many neurons or processing elements are interconnected to form a parallel neurocomputing network. Each element or neuron is a very simple processor that carries out the pondered sum of its inputs and applies them to a function (linear, sigmoidea) to generate an output that is sent to another neuron. The most usual type of ANN is the feedforward multilayer one, where no information is fed back during the recall process. Feedback signals are used only during the training of the neural network. Generally, the backpropagation method is used for adjusting the neural network weights during the training. This process requires a high consumption of time since the algorithm takes a long time to converge to the desired error, but this phase is usually made offline. By using an ANN, it is possible not to depend on aproximate models since the net learns with the complete model [3]. The use of a neural network to estimate the speed consists of an association of some inputs (stator currents, voltage, and frequency) with some outputs (speed and torque). In this case, for each set of inputs there is a set of outputs. To accomplish this operation, the net will have to be trained in a first phase. It is not necessary to carry out this phase in real time and to give the net all the possible inputs–outputs combinations since it has the capacity to generalize results starting from a limited set of inputs–outputs. Once the phase of training has been accomplished, the net is prepared to estimate the speed for any set of inputs. During the training, we have experimented with different architectures and learning... [continues]

Read full essay