Noise removal by cluster analysis after long time AE corrosion monitoring of steel reinforcement in concrete

Acoustic Emission technique is gaining more and more appreciation in the field of structural health monitoring for reinforced concrete structures. Noise removal and suppression still however remain a concern in AE data analysis. Clustering technique have been proposed in the present work as a tool to overcome this problem. Clustering can, in fact, support the identification of existing underlying relationships among sets of variables related for example to crack growth mechanism or noisy perturbations. It may represent a basic tool not only for classification of known categories, but also for discovery of new relevant classes.In this work different algorithms for automatic clustering and separation of AE events based on multiple features have been adopted. Noise was separated from events of interest and subsequently removed using a combination of different methods like PCA and k-means method. Several validation techniques have also been introduced for AE expression data analysis. Normalization and validity aggregation strategies have been proposed to improve the prediction about the number of relevant clusters. The remaining data have been processed using a self-organizing map (SOM) algorithm.

► We proposed a multi-step procedure by cluster analysis for noise removal from AE events.
► Validity indexes have been adopted for cluster validation.
► A Principal Component analysis (PCA) and self-organizing map (SOM) algorithms were developed to process the data.
► The dataset was reduced of about 65% after the filtering processes.