Neural Network Approaches to Data Fusion

The challenge of meeting the increasingly sophisticated industrial inspection needs has led to the development of a number of nondestructive evaluation (NDE) methodologies. NDE techniques rely largely on the interaction of some form of energy and the test specimen to provide information relating to the condition of the material. The choice of energy utilized in the inspection process is dictated by the type of material under inspection as well as by the nature and location of the flaw. A variety of energy sources have been employed to interrogate materials. These include acoustic, electromagnetic, optical and x-ray energy sources [1]. Each of these methods brings its own set of advantages and disadvantages and often no single technique offers a full solution to the inspection problem. As an example, ultrasonic imaging techniques offer excellent resolution and sensitivity to both surface breaking as well as subsurface cracks. However, the method is also sensitive to a wide variety of measurement conditions including surface roughness and coupling. In contrast, eddy current techniques do not require contact with the test specimen and are relatively insensitive to surface roughness conditions. The disadvantages associated with the eddy current method lies its insensitivity to defects that lie in the recesses of the material, its poor resolution capabilities, and its sensitivity to variations in liftoff. The energy / material interaction process is also fundamentally different in the two cases. Unlike the ultrasonic method which relies on wave propagation of energy, the eddy current process is essentially diffusive in nature. It can therefore be argued that one could profit from integrating information obtained from the two tests. The challenge lies in isolating components of information that are either complementary or redundant. The complementary segments of information can be utilized to improve the quality of characterization while at the same time using redundant information to improve the signal-to-noise ratio(SNR).