Neural network analysis of nondestructive evaluation patterns

Nondeabuctive evaluation (NDE) teelmiques are used to verify the integrity of aircraft parts prior to assembly, during assembly, and during maintenance operations. These tec.hniqueaare used to ensure that matedal flaws do not exist in critical structural parts. Current NDE technologies are being challenged by demands for increased aeeuracy, sreliability, and cost effectiveness. The objective of this reaearch was to develop techniques for the automated analysis of NDE test patterns to assist human technicians in performing reliable and efficient testing. This work was a cooperative effort between the Artificial Intelligence groups at Douglas Aircraft Company (DAC) and McDonnell Douglas Research Laboratories (MDRL) and tho NDE group at DAC. The DAC/MDRL NDE research team utilized real NDE tat data and two different neural network machine learning algorithms to construct knowledge bases which accurately analyzed NDE outputs at the human expert level. This work describes the reaearoh performed along with the results of experiments on both metallic and composite parts. This approach appears to be applicable to other similar domains.