Proceedings of the 6th International Conference on Mechanics and Materials in Design, Editors: J.F. Silva Gomes & S.A. Meguid, P.Delgada/Azores, 26-30 July 2015 -1843- PAPER REF: 5613 MULTIFRACTAL ANALYSES OF WELD DEFECTS PATTERNS OBTAINED FROM D-SCAN IMAGES Lindberg L. Gonçalves (*) , Elineudo P. Moura Department of Metallurgical Engineering and Materials, Federal University of Ceará, Fortaleza, Ceará, Brazil (*) Email: lindberg@fisica.ufc.br ABSTRACT In this paper are presented multifractal analyses of D-scan patterns constructed from TOFD ultrasonic signals obtained in specimens with three welding defects: lack of fusion, incomplete penetration and porosity. The analyses of the patterns were performed by means of Hurst and detrended-fluctuation (DFA) analyses. The curves obtained from these analyses, were processed by using two pattern classification techniques, namely, principal-component analysis and Karhunen-Loève expansion. For an appropriate combination of the analyses at different multifractal orders by using principal-component analysis a 100% success rate has been reached for the classification of the different defects. A similar result has been obtained by using the Karhunen-Loève expansion and by considering all events available as the training set. Keywords: Multifractal analysis, Hurst analysis, DFA analysis, time of flight diffraction (TOFD), welding defect, principal-component analysis, Karhunen-Loève expansion. INTRODUCTION Ultrasonic tests can serve as an useful tool for evaluating the integrity of metallic structures, and specially of weld joints. By inspecting the scattering pattern of ultrasonic waves propagating in the material, it is possible to identify the presence of defects, and to estimate their dimensions. However, it is often desirable to have precise information about the nature of the defects, and recently a new approach, based on fractal analysis has been proposed (Silva, 2012). These main features of the method have been introduced in our previous works, where welding defects have been studied by analysing TOFD ultrasonic signals (Vieira, 2008) and radiographic images (Tesser, 2007). In the present paper, we extend our previous result by analysing the problem within the context of multifractal (Oswiecimka, 2014) properties of the patterns built with the ultrasonic TOFD signals. By considering the new approach, we analyze three weld defect patterns (lack of fusion, lack of penetration and porosity), together with a pattern without defect, obtained from the D-scan TOFD ultrasonic signals associated with two well known statistical pattern- classification, namely, principal component analysis (PCA) and Karhunen-Loève (KL) linear transformation (Webb, 2002). We used 48 D-Scan images obtained by the TOFD technique, with 9 images corresponding to lack of fusion, 13 images corresponding to lack of penetration, 11 images corresponding to porosity, and 15 images corresponding to regions without defects. In order to classify the