Ensemble learning-based structural health monitoring by Mahalanobis distance metrics

Environmental variability is still a major challenge in structural health monitoring. Due to the similarity of changes caused by environmental variations to damage, false positive and false negative errors are prevalent in detecting damage that cause serious economic and safety issues. To address this challenge and its disadvantages, this article proposes a novel ensemble learning-based method in a nongenerative sequential algorithm for structural health monitoring under varying environmental conditions by three kinds of Mahalanobis distance metrics in three main levels. At each level, one attempts to find a few and adequate nearest neighbors of each feature to remove environmental variability via an innovative approach. The major contribution of this article is to develop a novel data-based method by the concepts of ensemble learning and unsupervised learning. The great advantages of the proposed method include developing a nonparametric data-based framework without estimating any unknown parameter, dealing with the negative effects of environmental variability, improving the performance of Mahalanobis distance, and increasing damage detectability. The performance and effectiveness of this method are validated by modal features of two real bridge structures along with several comparisons. Results demonstrate that the proposed ensemble learning-based method highly succeeds in detecting damage under environmental variability, and it is superior to some state-of-the-art techniques.