A digital twin approach based method in civil engineering for classification of salt damage in building evaluation

The integration of digital twins and machine learning models in civil engineering has revolutionized the inspection and maintenance of buildings and structures. Digital twins, as precise virtual replicas of physical assets, enable continuous monitoring and predictive maintenance, enhancing the reliability and efficiency of structural assessments.

This research aims to develop a convolutional neural network (CNN)-based approach for classifying salt damage in concrete structures, integrating digital twin technologies to enhance structural health monitoring and damage detection. This study leverages transfer learning techniques, utilizing six state-of-the-art pre-trained architectures, including VGG-16, InceptionV3, ResNet50, VGG-19, DenseNet121, and MobileNet. After extensive evaluation, VGG-16, was chosen as the final model for fine-tuning, achieving high accuracy in the classification of salt damage. The digital twin approach provides a virtual representation of structures to enable predictive maintenance and reduce subjectivity in inspections.

The fine-tuned CNN models demonstrated state-of-the-art accuracy in detecting salt damage, significantly outperforming traditional visual inspection methods. The use of digital twins enabled continuous monitoring and effective prediction of structural damage. The developed models offer a robust and efficient alternative to manual inspections, supporting the transformation of structural health monitoring in civil engineering. The results underline the potential of combining digital twins and deep learning to achieve precise and reliable structural assessments.