Structural damage localization through an innovative hybrid ensemble approach

Abstract

Damage localization in structures can be achieved by using an appropriate data interpretation algorithm based on the expected structural response. According to the several algorithms reported in literature, a different degree of accuracy is obtained according to complexity requirements. This paper presents a hybrid algorithm approach as alternative to combine some of the reported methods by employing an ensemble architecture. Thus, this damage assessment algorithm integrates advantage of individual techniques in order to increase the performance of the whole expert system. The proposed architecture employs a network of piezoelectric devices to produce guided waves along the structure. The traveling of guided waves is affected by damage producing scattering, reflection and mode conversion, which can be characterized with statistical processing and pattern recognition methods. In this paper, supervised learning by means on ensemble learning, cross-correlation features, and PCA statistical indices are combined for locating damages. An experimental validation is conducted on an aircraft turbine blade structure instrumented with an array of piezoelectric devices (PZT), where it is demonstrated the potential of the methodology to significantly enhance localization tasks.