Submit

Login Login

Register Register

Home / Journals / CMC / Online First / doi:10.32604/cmc.2026.086951
Journal Menu
Special Issues
Table of Content

All issues

Online First

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Open Access

ARTICLE

Morphology-aware CNN with Statistical Descriptor Regression and Conditional Image Synthesis for Predicting Crack Patterns in Steel Fiber-Reinforced Concrete

Khaja Wahaajuddin Kawkabi, Andre Steve Nziko Talla Nziko, Prince Manyanya Ngangura, Xu Long*
School of Mechanics and Transportation Engineering, Northwestern Polytechnical University, Xi’an, China
* Corresponding Author: Xu Long. Email: email
(This article belongs to the Special Issue: Physics-Informed AI and Multiscale Modeling for Design and Reliability of Advanced Materials and Structures)

Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.086951

Received 08 June 2026; Accepted 03 September 2026; Published online 20 September 2026

Abstract

Accurate characterization of crack evolution in steel fiber-reinforced concrete (SFRC) remains a fundamental challenge due to heterogeneous mesostructure and the highly nonlinear nature of fracture processes governed by fiber–matrix interactions. This paper proposes a morphology-aware two-stage framework that reformulates the ill-posed curve-to-image mapping into two well-posed steps. Stage 1 (StatNet) learns to regress an eleven-component vector of crack morphology descriptors directly from uniaxial stress-strain response and mix-design metadata using dedicated encoders. Stage 2 (PatternDecoder) synthesizes high-resolution crack-probability maps conditioned on predicted descriptors and a latent embedding of the stress-strain curve. The framework is trained and validated on 220 meso-scale Abaqus cohesive-zone model simulations with explicit fiber beam elements. The model achieves a global pooled coefficient of R2 = 0.940 across all morphology descriptors on the held-out validation folds, per-sample inference time of 14.57 ms. Comparisons against DirectCNN (direct convolutional neural network) and CGAN (conditional generative adversarial network) baselines confirm the superiority of the proposed approach in terms of structural fidelity, parameter efficiency, and computational cost. Residual diagnostics confirm homoscedasticity and near-normal errors, while gradient sensitivity analysis highlights physically consistent transitions at peak stress. The morphology-aware decomposition transforms an ill-posed mapping into an interpretable, computationally efficient surrogate model that bridges mesoscale fracture mechanisms with rapid crack-pattern assessment.

Keywords

Steel fiber-reinforced concrete; crack pattern prediction; morphology-aware CNN; surrogate modeling; cohesive zone model

  • 82

    View

  • 16

    Download

  • 1

    Like

Related articles
Copyright© 2026 Tech Science Press
© 1997-2026 TSP (Henderson, USA) unless otherwise stated

Share Link