A one-dimensional convolutional neural network-based approach with data augmentation for missing data recovery in bridge SHM systems

Abstract

This article proposes a deep learning approach that uses one-dimensional convolutional neural networks (1D CNN) to reconstruct missing vibration data in structural health monitoring (SHM) systems. The data were collected from a laboratory-scale cable-stayed bridge model, represented as univariate time series with randomly missing segments. In order to enhance the model’s learning capability and generalisation, Gaussian noise-based data augmentation is applied during training. The models are evaluated using RMSE, MAE, and R² metrics. The research results show that the 1D CNN model demonstrates superior capability in extracting local features from input signals, while also offering fast training speed, high stability, and a lightweight architecture, making it highly suitable for real-world applications. In addition, incorporating Gaussian noise with an appropriate standard deviation can significantly improve the reconstruction accuracy compared to the non-augmented data model. The proposed method has demonstrated strong potential for recovering lost or corrupted data in practical SHM systems, thereby improving the reliability of structural analysis and diagnostics.