Automatic defect recognition of single-v welds using full matrix capture data, computer vision and multi-layer perceptron artificial neural networks

摘要

This paper describes the development of an automatic defect recognition system applicable to full matrix capture (FMC) imaged data. Computer vision principles were used on FMC-reconstructed images for feature extraction and combined with a multi-layer perceptron artificial neural network for classification. A wide variety of single-v weld training samples were used to train the artificial neural network, which was then tested to determine its accuracy. Automatic defect classification of real single-v weld inspections achieved a high level of success. By training an artificial neural network with information relating to defect orientation, size and location, extracted automatically using computer vision techniques, it was shown that automated defect classification was possible, with little or no user intervention. The ability to automatically determine defect characteristics offers significant advantages for NDT.