Autonomous detection of concrete damage under fire conditions

The rapid advancement in computer vision has facilitated new means for the automatic assessment of structural damages. This study aims to develop a deep learning-based autonomous damage detection framework for concrete structures under fire conditions. A hybrid deep learning network comprising of Convolution Neural Network (CNN) and Long Short Term Memory (LSTM) network is proposed herein. Initially, the CNN is applied in the feature extraction phase, and the LSTM is used for damage detection and classification. The proposed hybrid network is then deployed to evaluate the structural damage of three types of self-compacting concrete (SCC) specimens exposed to standard fire conditions. A series of systematic studies are performed to optimize the network architecture and hyper-parameters. The effectiveness of the proposed hybrid method is contrasted with existing CNN methods against real datasets. Our analysis shows that the proposed framework delivers a robust and improved performance against traditional deep learning methods. Overall, the proposed framework opens the door for adopting autonomous damage detection systems for post-fire conditions. •An autonomous crack detection framework for post-fire conditions is proposed.•A hybrid CNN-LSTM network for damage detection is developed.•The hybrid CNN-LSTM model is tested real-time datasets.•The proposed approach can achieve human-level generalization adaptability.