Neural networks in automatic testing of diode protection circuits

Limiting Zener diode circuits are often used in ground support equipment for the Minuteman Missile. These circuits protect sensitive transistor and resistor components from electrical surges. Data were collected for 110 waveforms for a combination of good circuits and the most frequently encountered failure modes, including shorted diodes, open diodes and bad amplifiers. The data were then used to train a neural network pattern recognition system to see if neural network technology could correctly identify good versus bad protection circuits. When trained using all of the diodes, the neural network was able to identify correctly all of the circuits and associated failure models. To validate the neural network model, a subset of 59 samples was randomly selected for training of the neural network, and the remaining 51 samples were used for testing. In both instances, the network did an excellent job (100%) of identifying failure.<>