Towards automation of non-destructive testing of welds

All welding processes can give rise to defects that will weaken the joint and can lead to failure of the welded structure. Because of this, non-destructive testing (NDT) of welds have become increasingly important to ensure the structural integrity when the material becomes thinner and stronger and welds become smaller; all to reduce weight in order to save material and reduce emissions due to lighter constructions.Several NDT methods exists for testing welds and they all have their advantages and disadvantages when it comes to the types and sizes of defects that are detectable, but also in the ability to automate the method. Several methods were compared using common weld defects to determine which method or methods were best suited for automated NDT of welds. The methods compared were radiography, phased array ultrasound, eddy current, thermography and shearography. Phased array ultrasound was deemed most suitable for detecting the weld defects used in the comparison and for automation and was therefore chosen to be used in the continuation of this work. Thermography was shown to be useful for detecting surface defects; something not easily detected using ultrasound. A combination of these techniques will be able to find most weld defects of interest. Automation of NDT can be split into two separate areas; mechanisation of the testing and automation of the analysis, both presenting their own difficulties. The problem of mechanising the testing has been solved for simple geometries but for more general welds it will require a more advance system using an industrial robot or similar. Automation of the analysis of phased array ultrasound data consists of detection, sizing, positioning and classification of defects. There are several problems to solve before a completely automatic analysis can be made, including positioning of the data, improving signal quality, segmenting the images and classifying the defects. As a step on the way towards positioning of the data, and thereby easing the analysis, the phase of the signal was studied. It was shown that the phase can be used for finding corners in the image and will also improve the ability to position the corner as compared to using the amplitude of the signal. Further work will have to be done to improve the signal in order to reliably analyse the data automatically