Convolutional neural network for predicting crack pattern and stress-crack width curve of air-void structure in 3D printed concrete

Extrusion-based 3D concrete printing (3DCP) results in deposited materials with complex microstructures that have high porosity and distinct anisotropy. Due to the material heterogeneity rapid growth of cracks, fracture analysis these air-void structures is often complex, resulting a computational cost. This study proposes convolutional neural network (CNN)-based methodology for using as input. More specifically, lattice model used build dataset comprises input well output information, including crack patterns crack-width curves. To establish relationship between morphology associated microstructures, U-net first presented. With obtained pattern input, principal component (PCA) CNN are then integrated predict stress-crack width The predicted from demonstrate quantitative agreement numerical analyses, 0.85 Intersection over Union prediction 0.75 R2 curves prediction. indicates models can be an alternative traditional analysis. feature maps during or deconvolutional process given explain why proposed perform on system. Moreover, generalization discussed transfer learning fine-tuning show potential expressing varied pore information. In end, cropped XCT created explore further application printed materials.