Neural network application to particle impact noise detection

Particle impact noise detection (PIND) is a NASA-developed test to determine the presence of loose particles inside the cavities of electronic parts. These particles may cause failures during launch or in zero gravity. Because PIND testing is limited by the human senses, it is subjective and has an accuracy of less than 50%. In the research project reported here, the PIND test is standardized and automated by digitizing the analog noise signal obtained from a PIND tester and by analyzing the data using neural networks. Two types of neural networks were successfully trained and tested on the PIND data, achieving an accuracy within 1%. The results of this study indicate a significant improvement in the accuracy and reliability of PIND testing. This testing can help to eliminate the expense of false failures