Prediction of Cracks and Recognizing Its Patterns in Geopolymer Concrete Beams Using Machine Learning

Abstract

Geopolymer concrete has gained increasing attention as a sustainable and environmentally friendly alternative to conventional Portland cement concrete, offering improved durability and a lower carbon footprint. However, like traditional concrete, it remains vulnerable to cracking under mechanical or environmental stresses, which can significantly impact structural performance. This survey reviews recent advancements in the application of machine learning (ML) and artificial intelligence (AI) for predicting cracks and recognizing their patterns in geopolymer concrete beams. The study examines diverse approaches such as Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Random Forests, and emerging models including Vision Transformers and Graph Neural Networks. Key focus areas include image-based crack detection, pattern classification, data preprocessing, and integration of multimodal sensing technologies. The survey highlights trends such as the shift toward explainable AI, federated learning for privacy-preserving model training, and digital twin-based predictive systems. Despite significant progress, challenges remain in the availability of large annotated datasets, generalization across different geopolymer mixes, and achieving real-time deployment on edge devices. The findings suggest that hybrid and adaptive ML frameworks can play a crucial role in enhancing predictive accuracy and robustness, paving the way for intelligent, sustainable structural health monitoring systems in modern infrastructure.