Uncertainty‐assisted deep vision structural health monitoring

Abstract

Computer vision leveraging deep learning has achieved significant success in the last decade. Despite the promising performance of the existing deep vision inspection models, the extent of models’ reliability remains unknown. Structural health monitoring (SHM) is a crucial task for the safety and sustainability of structures, and thus, prediction mistakes can have fatal outcomes. In this paper, we use Bayesian inference for deep vision SHM models where uncertainty can be quantified using the Monte Carlo dropout sampling. Three independent case studies for cracks, local damage identification, and bridge component detection are investigated using Bayesian inference. Aside from better prediction results, the two uncertainty metrics, variations in softmax probability and entropy, are shown to have good correlations with misclassifications. However, modifying the decision or triggering human intervention can be challenging based on raw uncertainty outputs. Therefore, the concept of surrogate models is proposed to develop the models for uncertainty‐assisted segmentation and prediction quality tagging. The former refines the segmentation mask and the latter is used to trigger human interventions. The proposed framework can be applied to future deep vision SHM frameworks to incorporate model uncertainty in the inspection processes.

Deep learning for post‐hurricane aerial damage assessment of buildings

“…More rigorous quantification of confidence in model predictions requires a nondeterministic approach that deploys uncertainty quantification and propagation. While outside the scope of this work, such outcome can be achieved, for instance, using a Bayesian deep learning framework, which enables modeling epistemic and aleatoric uncertainty (Gal & Ghahramani, 2016; Kendall & Gal, 2017; Sajedi & Liang, 2020).…”

Pixel‐level multicategory detection of visible seismic damage of reinforced concrete components

“…Even if sample-level data balancing is achieved, data imbalance at the pixel level alone can hinder and stagnate the training process. A method was developed by Sajedi and Liang (2020), which functions in the inference phase to minimize the negative impact of data imbalance. In this study, a different method is proposed that mitigates the pixel-level data imbalance in the training process of a CNN.…”

Semi‐supervised learning based on convolutional neural network and uncertainty filter for façade defects classification

“…However, the majority of unlabeled data would lead to the wrong direction during the initial epochs, since the teacher model was too weak to generate high‐quality predictions while there were no ground‐truth labels to correct the predictions (Y. Li, Liu, & Tan, 2019). To select high‐quality samples, some research used the term of uncertainty in deep learning research for civil engineering problem such as structural health monitoring (Sajedi & Liang, 2020). Moreover, the research of (Yu, Wang, Li, Fu, & Heng, 2019) proposed to use an uncertainty‐aware self‐ensembling model to conduct semi‐supervised learning for segmentation of 3D medical images, which is very inspirational for involving uncertainty in semi‐supervised learning.…”