Uncertainty quantification in digital image correlation for experimental evaluation of deep learning based damage diagnostic

As the temporal and spatial resolution of monitoring data drastically increases by advances in sensing technology, structural health monitoring applications reach the thresholds of big data. Deep neural networks are ideally suited to use large representative training datasets to learn complex damage features. One such real-time deep learning platform that was developed to solve damage detection and localisation challenge in the authors previous paper. This network was trained by using simulated structural connection with a variety of loading cases, damage scenarios, and measurement noise levels for robust diagnosis of damage. In this article, this platform is validated by using the data collected by Digital Image Correlation (DIC) which offers a non-contact method to measure full-field strain by increasing the flexibility of their implementation. Nevertheless, the capabilities of DIC while measuring small strain responses is limited. This article first investigates the accuracy of the strain measurements of a structural component subjected to operational loads which are often smaller than 50 mu epsilon. The accuracy of three DIC systems with different camera resolutions is compared with the measurements collected by strain gauges and finite element model. Then, the performance and efficiency of damage diagnosis approach is evaluated on two induced damage conditions.

Automated summary

This article discusses a method for structural health monitoring using deep neural networks and validates its effectiveness. It first explores the application of Digital Image Correlation in small strain situations, and then evaluates the performance of the damage diagnosis method under two induced damage conditions.