A neural network for the automatic diagnosis of the telephone switching systems

This paper reports the development of a neural network expert system for the fault diagnosis of telephone switching systems. By using fault diagnosis and maintenance records from experienced maintenance operators, the neural networks can be trained to diagnose faulty telephone switching system automatically. Binary type adaptive learning networks are selected for the implementation of the neural network diagnosis system. In addition, some modifications on the supervised adaptation learning algorithm are proposed to alleviate the local minimum problems in order to improve the performance. From the simulation results, the fault diagnostic rate of applying the neural networks expert system on a GTD-5EAX switching system is above 99%. To further enhance the retrieving performance of the neural network, the authors proposed a VLSI implementation of multiple (24) binary adaptive networks containing a total of 2¹⁴-1 nodes on a chip