Clustering acoustic emission signals by mixing two stages dimension reduction and nonparametric approaches

• DOI: 10.1007/s00180-018-00864-w
• Publication Date: 2019-01-16T05:43:51Z
• Authors: O. I. Traore; P. Cristini; N. Favretto-Cristini; L. Pantera; P. Vieu; S. Viguier-Pla
• Keywords: curve smoothing; [SPI.ACOU]Engineering Sciences [physics]/Acoustics [physics.class-ph]; [SPI.ACOU] Engineering Sciences [physics]/Acoustics [physics.class-ph]; 330; Functional Principal Component Analysis; 01 natural sciences; 7. Clean energy; semi-metric; Functional clustering; [MATH.MATH-ST]Mathematics [math]/Statistics [math.ST]; 0103 physical sciences; [MATH]Mathematics [math]; [MATH.MATH-ST] Mathematics [math]/Statistics [math.ST]; hierarchical clustering

In the context of nuclear safety experiments, we consider curves issued from acoustic emission. The aim of their analysis is the forecast of the physical phenomena associated with the behavior of the nuclear fuel. In order to cope with the complexity of the signals and the diversity of the potential source mechanisms, we experiment innovative clustering strategies which creates new curves, the envelope and the spectrum, from each raw hits, and combine spline smoothing methods with nonparametric functional and dimension reduction methods. The application of these strategies prove that in nuclear context, adapted functional methods are effective for data clustering.