Advances in Classification of AE Sources


COFREND Conference 2001, CND & Corrosion, Reims, France, 24-26 April 2001

Dimitrios Kouroussis, Dr. Athanassios Anastassopoulos, Dr. Jean-Claude Lenain, Dr. Alain Proust


During the last few decades, the Acoustic Emission NDT technique has experienced considerable growth both in terms of sheer number of inspections and users, and in terms of range of applications. The capabilities of the technique have been proven or are under investigation for a vast number of materials, processes, applications and structures. The ever-increasing demand for the analysis, characterization and understanding of the Acoustic Emission (AE) sources, which are detected during loading of structures, has led researchers towards the development of methodologies for the classification of the corresponding AE data. In this context, Unsupervised Pattern Recognition (UPR) has recently been applied for the segregation of AE data obtained during various different applications. The present paper outlines the basic features of Unsupervised Pattern Recognition for AE data and reports on some successful application examples of the technique, with the use of specialized Pattern Recognition software. It is shown that, upon proper selection of the UPR algorithms and parameters, the technique can successfully identify and separate noise-related AE (EMI, friction, mechanical impacts) from legitimate AE. Additionally, application of the UPR technique on AE data obtained during fatigue testing of a composite wind turbine blade managed to identify the various co-existing failure mechanisms and to assess their accumulation on the blade with increasing fatigue cycles up to final failure. Furthermore, UPR has been applied for the identification of AE signals arising from hydraulic noise during loading of an aerial man-lift device. The corresponding results of UPR were applied by means of Supervised Pattern Recognition on further AE data obtained during manipulation of the device’s arm, and hydraulic noise was very efficiently separated. In overall, it is now evident that combination of traditional AE analysis techniques (AE location, AE activity with load etc.) with UPR analysis can be a powerful tool towards the evaluation and physical interpretation of AE data. Finally, future work in the area of the definition of AE signatures for the existing damage and noise types of Acoustic Emission in various applications, and application of SPR could, ultimately, lead to the automation of the classification and noise elimination procedure and the establishment of pass/fail criteria for future tests, based on structurally significant classes.