Control and diagnosis of electrical drives: some applications by using neural networks

Some applications of neural networks to the control and diagnosis of electrical drives are presented. In the first part a direct inverse control scheme is presented for controlling a DC motor, which is based on a clustering neural network, called the progressive learning network (PLN) because of its inherent capacity of learning online. This approach can control the whole system without having to use a very rich training set; moreover it is able to adapt itself online to new working conditions by varying the number of neurons. In the second part of the paper some applications of self-organising neural networks are described for the diagnosis of AC drives. In particular it is shown that the vector quantisation projection algorithm can be useful for diagnosis purposes since it permits an easier representation of the output space than that available with the Kohonen's map.