TMS320 DSP based neural networks on fault diagnostic system of turbo-generator

Artificial neural networks (ANN) are massive parallel interconnections of simple neurons that function as a collective system. More and more people are paying attention to ANN on fault diagnostic of turbo-generator for its association of thought, recollection and study function. However, the disadvantage of ANN lies in its huge data computation and the low speed of convergence. If we realize ANN with common CPU, it needs so much time on computing the huge data, so that real-time fault diagnosis become impossible. In fact, most of the computation in ANN is multiplication and addition. While digital signal processors (DSP) has altitude advantage on multiplication and addition computation, it can perform parallel multiplication and addition in a single cycle clock. Consequently we design a master/slave system to solve the problem. The slave system was mainly made up of DSP, which perform high speed ANN calculation. The master system was made up of PC, which performs data communication and real-time fault diagnosis. In this paper we bring forward a practical system design and present particular design method of hardware and software by using back-propagation (BP) network often used in fault diagnosis.