Neural Network Based Rail Flaw Detection Using Unprocessed Ultrasonic Data

In the current practice in rail flaw detection the raw (unprocessed) data are processed to generate the data for simple visual displays that can be produced for the operator. Useful information gets discarded in processing the ultrasonic data. This Innovations Deserving Exploratory Analysis (IDEA) project aimed to develop methods, employing appropriately designed and trained neural networks, that can use the unprocessed ultrasonic data in railroad rail flaw detection. Sperry Rail Service, Danbury, Connecticut, was to be the co-funder and participant in this project. When Sperry Rail Service had to withdraw from the project for internal reasons, the project was terminated. Even though this project could not be completed, the potential advantage of using unprocessed ultrasonic data remains. A vision for future research is provided in this final report.