Robust multitask compressive sampling via deep generative models for crack detection in structural health monitoring

Abstract

In structural health monitoring (SHM), there is an increasing demand for real-time image-based damage detection. Such a technology is essential for minimizing hazard loss caused by delayed emergency response after earthquakes or other natural disasters, or service interruption during structural inspection. Compressive sampling (CS) is a promising solution to achieve such a goal by greatly reducing the power consumption on high-resolution image transmission when using wireless devices. However, conventional CS failed to achieve high enough compression ratios, while existing generative-model-based CS requires laboriously training a high-quality generator with many large-scale images. To overcome such a bottleneck that hinders the practical use of CS in SHM, we propose a multitask CS algorithm that only relies on existing generators trained by low-pixel crack images. By exploiting the new discovery that similar crack images share a similar sparsity pattern in their latent vectors mapped by the generator, our algorithm achieves higher crack detection accuracy and robustness within a much shorter time when using a high data compression ratio. We verify the effectiveness of the proposed CS algorithm using synthetic and real image data. The results demonstrate that this work has moved a step closer toward successful implementation of operational CS-based crack detection systems in real-time SHM.

Vector Quantized Variational Autoencoder-Based Compressive Sampling Method for Time Series in Structural Health Monitoring

“…For example, while many time series signals are sparse after wavelet transform, this result does not apply to acceleration signals measured from some real structures [15], limiting the application of the CS method in many SHM applications. In recent years, many researchers have attempted to take advantage of the strong feature-learning ability of deep neural networks (DNNs) [16][17][18][19] to relieve the signal sparsity constraint in conventional CS, e.g., Bora et al [20] proposed to use well-trained deep generative networks that capture the high-dimension image signals in a low-dimension space as an implicit regularizer for the ill-posed CS problem; Huang et al [21] and Zhang et al [22] successfully applied this idea to achieve high segmentation accuracy on building crack images with a high compression ratio; Dave et al [23] proposed to use the architecture of an untrained deep convolutional generative adversarial networks as a prior to solve any differentiable linear inverse problem for image data, assuming such an architecture has already provided enough constraints to capture the underlying distribution of natural images [24].…”