Detection of the propagation of defects in pressurised pipes by means of the acoustic emission technique using artificial neural networks

The acoustic emission test has distinguished relevance among non-destructive testing and, therefore, research abounds at present aimed at the improvement of the reliability of results. In this work, the methodologies and the results obtained in a study performed are presented to implement pattern classifiers by using aritificial neural networks, aimed at the detection of propagation of existing defects in pressurised pipes by means of the Acoustic Emission testing (AE). Parameters that are characteristic of the AE signals were used as input data for the classifiers. Several tests were performed and the classification performances were in the range of 92% for most of the instances analysed. Studies of parameter relevance were also performed and showed that only a few of the parameters are actually important for the separation of the classes of signals corresponding to No Propagation (NP) of defects and Propagation (P) of defects. The results obtained are pioneering in this type of research and encouraged the present publication.