Concrete structures deteriorate over time due to environmental exposure and mechanical stress, leading to various types of damage such as cracking, spalling, corrosion, and exposed rebar. Automated detection using deep learning-based computer vision techniques is limited by the lack of high-quality, annotated data sets. To address this challenge, this paper presents multidamage monitoring of concrete structures (MDMCS), a data set of 1,200 images with precise pixelwise annotations involving four types of damage (cracking, spalling, corrosion, and exposed rebar) and diverse lighting conditions and material textures. The data set was evaluated using six state-of-the-art segmentation models, validating the efficacy of the data set and providing benchmarks for damage detection models. MDMCS will facilitate advances in artificial intelligence-powered structural monitoring and robot-assisted automatic inspection for improving the operation and maintenance of concrete structures.

Conventional asphalt concrete has a limited lifespan due to cracking, deformation, and environmental degradation, driving the development of fiber-reinforced asphalt concrete (FRAC). However, key gaps remain in current data-driven FRAC studies due to small and homogeneous datasets, “black-box” machine learning models, and trade-offs between mechanical-sustainable performance, failing to provide a transparent understanding of features governing FRAC behaviors. This paper proposes a framework integrating explainable artificial intelligence and life cycle assessment (LCA) to advance mechanical and sustainable design of FRAC. A dataset of 2490 laboratory samples covers 15 input features and 3 mechanical outputs. Eight machine learning models, along with a voting ensemble strategy, were optimized using Genetic algorithm for hyperparameter tuning. The optimized voting ensemble achieved an average prediction performance of R² = 0.87, RMSE = 1.09, MAPE = 11.96%, and MAE = 0.60 across the three mechanical targets, indicating robust and reliable predictive capability. SHapley Additive exPlanations (SHAP) analysis and linear non-gaussian acyclic causal inference quantified global/local feature impacts and pairwise interactions. LCA evaluated economic and environmental impacts and derived strength-normalized sustainability metrics. Finally, an interactive graphic user interface platform was developed for predictions, SHAP interpretations, and LCA outcomes. This data-driven approach establishes a paradigm for intelligent FRAC design, harmonizing mechanical performance with sustainability.

Crack detection accuracy in computer vision is often constrained by limited annotated datasets. Although Generative Adversarial Networks (GANs) have been applied for data augmentation, they frequently introduce blurs and artifacts. To address this challenge, this study leverages Denoising Diffusion Probabilistic Models (DDPMs) to generate high-quality synthetic crack images, enriching the training set with diverse and structurally consistent samples that enhance the crack segmentation. The proposed framework involves a two-stage pipeline: first, DDPMs are used to synthesize high-fidelity crack images that capture fine structural details. Second, these generated samples are combined with real data to train segmentation networks, thereby improving accuracy and robustness in crack detection. Compared with GAN-based approaches, DDPM achieved the best fidelity, with the highest Structural Similarity Index (SSIM) (0.302) and lowest Learned Perceptual Image Patch Similarity (LPIPS) (0.461), producing artifact-free images that preserve fine crack details. To validate its effectiveness, six segmentation models were tested, among which LinkNet consistently achieved the best performance, excelling in both region-level accuracy and structural continuity. Incorporating DDPM-augmented data further enhanced segmentation outcomes, increasing F1 scores by up to 1.1% and IoU by 1.7%, while also improving boundary alignment and skeleton continuity compared with models trained on real images alone. Experiments with varying augmentation ratios showed consistent improvements, with F1 rising from 0.946 (no augmentation) to 0.957 and IoU from 0.897 to 0.913 at the highest ratio. These findings demonstrate the effectiveness of diffusion-based augmentation for complex crack detection in structural health monitoring.

The study explores dataset preparation, machine learning (ML) model training, interpretation, and life cycle assessment (LCA) to predict and enhance the performance and sustainability of rubberized concrete. A large dataset comprising 1209 collected samples with nine input features was used to train and evaluate six machine learning models. Among the six models, the Light gradient boosting machine (LightGBM) model achieved the highest prediction accuracy on the testing dataset, with an R² value exceeding 0.96, a MAPE of 8.31 %, a MAE of 2.36 MPa, and a RMSE of 3.25 MPa. The SHAP algorithm was used to interpret predictions and identify key factors influencing compressive strength. Rubber content and water-to-cement ratio reduced strength, while longer curing time, more superplasticizer, higher fine aggregate content, and a greater silica fume-to-cement ratio improved it. Coarse aggregate, crumb rubber size, and cement content had minimal impact. Optimal performance was achieved with: Rubber content <55.92 kg/m³, w/c ratio <0.4, curing time >26 days, superplasticizer >3.7 kg/m³, fine aggregate >642 kg/m³, and silica fume/cement ratio >2.5 %. LCA results show that, although rubberized concrete offers no clear advantage over conventional concrete in cost, carbon, or energy, the lower strength and higher superplasticizer use of rubberized concrete lead to greater strength-normalized impacts, and therefore its value lies more in waste recycling and toughness than in strength-based sustainability. The unique advantage of this research lies in the development of a ML-LCA integration framework that synthetically balances performance prediction and sustainability assessment of rubberized concrete, with delivering actionable mix design recommendations, identifying key and low-impact variables, and revealing trade-offs among strength, cost, and environmental performance.

Geopolymer concrete (GPC) is a sustainable alternative to Portland cement concrete by eliminating Portland cement. However, using alkaline activators compromises the sustainability of GPC. This paper presents a comprehensive assessment of the cradle-to-gate life cycle cost, carbon footprint, and energy consumption of 2304 GPC mixtures which represent the state-of-the-art dataset, aiming to establish a holistic understanding of the impacts of GPC design parameters on mechanical and sustainability performance. The cost-benefit characteristics of solid wastes are considered, and strength-normalized sustainability parameters are discussed. Results reveal that the formulation of GPC plays critical roles in mechanical and sustainability metrics. Inappropriate use of alkaline activators and solid wastes can largely compromise both mechanical strength and sustainability metrics, lower than their Portland cement counterparts. Based on the large dataset, this paper identifies the appropriate upper and lower bounds for various ingredients to guide the design of GPC for balanced mechanical strength and sustainability metrics.

As urbanization accelerates, aging infrastructure demands more advanced inspection methods for structural health monitoring. The growing integration of artificial intelligence (AI) and computer vision technologies has significantly enhanced damage detection accuracy while simultaneously reducing inspection time and operational costs. Despite these advantages, the adoption of AI-based technologies in infrastructure maintenance remains limited due to challenges related to data. One major issue is the lack of comprehensive, task-specific annotated datasets. Another is the poor quality of images captured by drones or mobile devices, which are often affected by noise, blurring, and inconsistent lighting. Although recent advances in generative AI offer promising support for structural health monitoring, it remains unclear which models are best suited for specific tasks. This study examines the use of generative AI in structural health monitoring, focusing on key challenges such as limited datasets and low-quality image restoration. The review covers a range of generative AI technologies, outlining their principles, strengths, limitations, and representative applications to support the selection of appropriate tools for specific tasks. Generative AI models enable accurate image segmentation and structural anomaly detection using limited training data. The paper also explores new opportunities for integrating multi-modal generative AI to enhance human–computer interaction in support of structural health monitoring. A framework is proposed to streamline the use of generative AI technologies for data augmentation, image restoration, damage inspection, and human–computer interaction in structural health monitoring.