Blast response and damage assessment of reinforced concrete slabs using convolutional neural networks

Abstract

Concrete structures are essential for shelters, storage, transportation, and defense systems. However, they are vulnerable to terrorist attacks and explosions. The most exposed component of these structures is the reinforced concrete slab, which is also the primary force-transferring member. Therefore, the present study utilizes machine learning techniques to predict the maximum vertical displacement of reinforced concrete slabs subjected to air-blast loading. This can be achieved using 11 input parameters of the slab and TNT blast to predict the maximum displacement. The dataset comprises 146 samples from various experimental and numerical blast studies on reinforced concrete slabs in the open literature. Rather than presenting the data in a tabular format, each individual data sample is transformed into an image using distinct techniques: one uses a self-similarity matrix, and the other utilizes an image generator for the tabular data. Image generation transforms tabular data into images by assigning features to pixel positions. This results in spatial dependency of the input features. Using these images, various convolutional neural networks were adopted (ResNet-18, ResNet-50, ResNet-101, EfficentNet-b0, ShuffleNet, Xception, DarkNet-53, and DenseNet-20) and trained to predict the slab maximum displacement. Most models demonstrated promising results. The performance of the models was predicted based on the root mean squared error, mean absolute error, and coefficient of determination, and the impact of input features on the maximum displacement was examined. Along with this, the initial study of the blast damage assessment on reinforced concrete slabs is explained for future work to be performed based on the proposed method.