Feature Selection in Structural Health Monitoring Big Data Using a Meta-Heuristic Optimization Algorithm

This paper focuses on the processing of structural health monitoring (SHM) big data. Extracted features of a  structure are reduced using an optimization algorithm to find a minimal subset of salient features by removing noisy, irrelevant and redundant data. The PSO-Harmony algorithm is introduced for feature selection to enhance the capability of the proposed method for processing the  measured big data, which have been collected from sensors of the structure and uncertainties associated with this process. Structural response signals under ambient vibration are preprocessed according to wavelet packet decomposition (WPD) and statistical characteristics for feature extraction. It optimizes feature vectors to be used as inputs to surrogate models based on the wavelet weighted support vector machine (WWLS-SVM) and radial basis function neural network (RBFNN). Two illustrative test examples are considered, the benchmark dataset from IASC-ASCE SHM group and a 120-bar dome truss. The results indicate that the features acquired by WPT from vibrational signal have higher sensitivity to the damage of the structure. Furthermore, the proposed PSO-Harmony is compared with four well-known metaheuristic optimization algorithms. The obtaind results show that the proposed method has a better performance and convergence rate. Finally, the proposed feature subset selection method has the capability of 90% data reduction