Comparative analysis of Auto-associative NNs and SVMs applied to patterns classification in the damage detection system

The paper presents briefly a principle functioning of a structure health monitoring (SHM) system which utilizes the phenomenon of elastic waves propagation and soft computing methods. Efficiency and robustness of the design SHM system was verified based on signals measured in laboratory strip specimen. Piezoelectric transducers technology was used here in order to actuate and sense elastic waves signals. Following a procedure of signal de-nosing a principal component analysis (PCA) was computed and then a number of signal parameters were used for training the SHM system. In this study, a binary decision tree and support vectors machines (SVMs) were applied for multi-level pattern classification. The obtained results were discussed and compared with those related to previously used Auto-associative Neural Networks (ANNs). On this basis, conclusions were drawn on a performance obtained of the classifiers studied.