A machine learning methodology for structural damage classification in structural health monitoring

One of the goals of structural health monitoring (SHM) applications is to determine the presence and the severity of a damage. In some cases, this is an element to forecast the behaviour and take decisions to allocate maintenance or replace the structure or the piece. An appropriate decision can reduce the risk of an accident, making more efficient the management of maintenance tasks and reducing the costs while improving the performance of a system. In this way, the development of a good SHM system is a need. Through the use of: (i) advanced methodologies of digital signal processing; (ii) the acquisition of information from a set of piezoelectric sensors appropriately placed in the surface of a structure; and (iii) the use of techniques such as principal component analysis (PCA) [1, 2, 3] and machine learning, it is possible to generate a solution that meets the necessity about the knowledge of the structural state. This work presents a methodology which allow to determine the presence of a structural damage and its classification in spite of temperature changes. The methodology is tested with a composite plate instrumented with a PZT sensor network and some added masses as damages. The whole system is validated to different temperatures