Automatic detection of defects in industrial ultrasound images using a neural network

Time-of-Flight Diffraction (TOFD) is a relatively new method of ultrasonic inspection and is well suited to semiautomation using methods such as robotic scanning, computer conditioned data acquisition and signal and image enhancement. However very little work has been documented on the full computer understanding of such scans. Instead, most work has been directed at aiding the manual interpretation process to determine defect characteristics. This paper describes the application of image processing and neural networks (ANNs) to the task of completely automating the decision making process involved in the interpretation of TOFD scans. Local area analysis is used to derive a feature vector which contains two-dimensional information on defect/component and non-defect areas. These vectors are then classified using an ANN trained with the backpropagation algorithm. The labelled image is then further segmented using binary shape analysis to discriminate between component echoes, or defect signals. Time-of-Flight correction techniques may be then used in order to determine the location of defects within a scanned weld.