ShM of bridges by improved complete ensemble empirical mode decomposition with adaptive noise (iceemdan) and clustering

Structural health monitoring (SHM) is a very broad field and it is fast growing. For instance, it is being used to identify damages in civil structures by using increasingly modern systems and tools, both to obtain data and for its subsequent processing and analysis. The advancement of technology in data processing and data mining have demonstrated the efficiency of supervised and unsupervised machine learning algorithms in different fields such as medicine, biology, finance, aeronautics and engineering. However, still a problem is the limited application to real civil structures, especially in bridges, which deserves a more exhaustive study in this field. When dealing with vibration data, in many cases, the Fast Fourier Transform (FFT) is used to obtain the damage sensitive features. However, for non-stationary and non-linear signals the Hilbert Huang Transform (HHT) is more efficient considering Empirical Mode Decomposition (EMD) method to decompose the signal into its main components. The present paper shows the feature extractions using an Improved Completed Ensemble Empirical Mode Decomposition with Adaptive Noise technique (ICEEMDAN) and the damage identification and localization by a clustering-based approach. The effectiveness of the methodology is shown using a real case of study in which four structural damage scenarios were imposed in a Warren truss bridge and the vibration caused by a crossing vehicle was recorded by accelerometers. The clustering results showed good correspondence with the damage scenarios located in different bridge zones and, therefore, the proposed approach demonstrates the feasibility for damage feature extraction as well as damage identification and localization.