Damage detection and noise cancelation algorithm in building structures using autocorrelation data and improved pattern construction method

Although many studies analyze data in structural health monitoring under noisy conditions, few focus on results in high noise levels, where detecting changes, whether clear damage or hidden deterioration, presents a significant challenge due to environmental noise. In this study, a methodology is presented that combines robust algorithms proven effective in noisy environments. This approach not only identifies changes but also localizes and assesses the severity of deterioration and damage while mitigating noise impact on the final results. The methodology employs autocorrelation as an alternative to time history responses, which serves inputs to a discrete wavelet transform. Several statistical indices are then embedded to create the features of the algorithm. Gray wolf optimization is applied as a novel feature selection method in damage detection problems, yielding outstanding results. Integrating these selected features and applying a moving maximum filter produces a superior pattern that significantly enhances the outcomes. Finally, a supervised Ensemble learner is used to classify the data. The effectiveness of the proposed method is demonstrated through three case studies: a benchmark study consisting of 17 diverse linear and nonlinear scenarios under forced vibrations, a numerical deterioration model with 5 different scenarios in response to ambient vibrations, and an experimental model with 10 predefined damage scenarios subjected to forced chirp signals. The results show high accuracy, exceeding 90% across all scenarios, including original data and data contaminated with Gaussian noise at signal to noise ratio levels of 10, 5, 1, and 0.5 decibels.