ABSTRACT
Recently, there has been growing interest in deep learning (DL) methods for damage detection which avoid the process of manual feature extraction and have strong feature extraction capability. However, the effectiveness of DL-based damage detection is often limited by class imbalance in the dataset, where instances of damage are much fewer compared to undamaged cases. This study proposes a framework for damage detection via data augmentation to address the issue of class imbalance in datasets for structural health assessment. Firstly, the raw data are normalized and sliced to create original datasets of both healthy and damaged conditions. Subsequently, a deep generative model, named Denoising Diffusion Probabilistic Model (DDPM), is employed to generate additional samples based on the limited number of original datasets of damaged condition. DDPM operates by gradually adding Gaussian noise to the original data with timesteps according to a pre-defined schedule in a forward process, followed by a reverse process to generate samples by denoising and reconstructing the added noise by a U-Net network. Finally, the augmented dataset, combined with the original ones of healthy condition, are utilized to train a deep Convolutional Neural Network (CNN) for damage identification. The proposed method is validated through a three-span continuous beam bridge, from which a large volume of vibration data of the healthy condition and a smaller volume of the damaged condition are simulated under white noise excitation. The impact of the nodes for damage diagnosis, imbalance ratio and measurement noise on the classification performance is evaluated in detail.
