Nondestructive Testing from Fuzed Data

Nondestructive inspection using fused data from diverse inspection sensors has the potential fo characterizing defects within ordnance items which may not be characteriazed properly by one sensor only. Data from different NDI methods are often complementary, therby ading diagnostic information to help lead the inspection to the correct decision about the disposition of a defect. The objectives of the Phase I proposed effort are to design and implement a technique for fusing nondestructive sensor data. We will develop the fusion system for a specific problem that confronts Picatinny Arsenal in the analysis of ordnance items. In addition, we will design the fusion system to be general enough to fuse any type of sensor data. The general objectives of the Phase I project are to lay down a foundation for developing a robust image-encoding scheme that will allow a neral network-based classifier to reliable distinguish getween various defect classes. We will also design a decision-level fusion methodology that is able to combine different forms of evidence relating to the nature of particular defects found in ordnance.