An Innovative Data-Driven Approach for Detection of Linear and Nonlinear Damage in Building Structures Using Signal Processing and BiLSTM-Based Deep Learning

This paper introduces a new method to detect damage in building structures using a data-driven approach. The method undergoes testing on two benchmarks, one featuring linear damage and the other nonlinear damage. The approach consists of three main phases. In the first phase, the short-time Fourier transform (STFT) and successive vibrational mode decomposition (SVMD) process are used to decompose signals to extract features. The features include instantaneous frequencies, spectral entropy, and intrinsic mode functions (IMFs). In the second phase, feature vectors form from these features, representing the structure’s transient behavior. The deep learning models used, long short-term memory (LSTM) and bidirectional-LSTM (BiLSTM), capture the long-term dependencies in the features from linear and nonlinear damage scenarios and then classify them. The results indicate that this method effectively classifies damages with less than 2% error for both models. The method’s robustness undergoes testing under various signal-to-noise ratios. The findings show that the BiLSTM model surpasses the LSTM and performs well even in challenging environmental conditions. The presented method contributes significantly to structural health monitoring by providing a reliable and accurate way to detect building structure damage. Including deep learning-based classifiers enhances the method’s accuracy and robustness, making it suitable for real-world applications.