Concrete surface crack detection is a critical research area for ensuring the safety of infrastructure, such as bridges, tunnels and nuclear power plants, and facilitating timely structural damage repair. Addressing issues in existing methods, such as high cost, lengthy processing times, low efficiency, poor effectiveness and difficulty in application on mobile terminals, this paper proposes an improved lightweight concrete surface crack detection algorithm, YOLOv8-Crack Detection (YOLOv8-CD), based on an improved YOLOv8. The algorithm integrates the strengths of visual attention networks (VANs) and Large Convolutional Attention (LCA) modules, introducing a Large Separable Kernel Attention (LSKA) module for extracting concrete surface crack and local feature information, adapted for features such as fracture susceptibility, large spans and slender shapes, thereby effectively emphasizing crack shapes. The Ghost module in the YOLOv8 backbone efficiently extracts essential information from original features at a minimal cost, enhancing feature extraction capability. Moreover, replacing the original convolution structure with GSConv in the neck network and employing the VoV-GSCSP module adapted for the YOLOv8 framework reduces floating-point operations during feature channel fusion, thereby lowering computational complexity whilst maintaining model accuracy. Experimental results on the RDD2022 and Wall Crack datasets demonstrate the improved algorithm increases in mAP50 by 15.2% and 12.3%, respectively, and in mAP50-95 by 22.7% and 17.2%, respectively, whilst achieving a reduced model computational load of only 7.9 × 10⁹, a decrease of 3.6%. The algorithm achieves a detection speed of 88 FPS, enabling real-time and accurate detection of concrete surface crack targets. Comparison with other mainstream object detection algorithms validates the effectiveness and superiority of the proposed approach.

Compressive strength of concrete is an important parameter in the design of concrete structures and the prediction of their durability. Therefore, it is of great significance to predict the compressive strength of concrete. In this study, a fully connected neural network model is developed using the PyTorch framework to predict the compressive strength of concrete and compared with six other machine learning models. These models are multiple linear regression, K‐nearest neighbor regression, support vector machine, decision tree, random forest, light gradient boosting machine, and artificial neural network. The model is trained using 4,253 data with seven input parameters, including cement (C), fly ash (F), mineral powder (K), fine aggregate (FA), coarse aggregate (CA), water reducer admixture (WRA), and water (W). Three thousand six hundred twenty‐one data in the datasets are used to train the prediction model after data cleaning, and 632 data are used to validate the model. The results show that the fully connected neural network model based on PyTorch frame can predict the compressive strength of concrete with higher accuracy. Therefore, it is a reliable and useful method to optimize the artificial network model. So, it has important application value in practice. The strength of concrete can be predicted in advance, making the project more efficient and reducing costs. Besides, by adjusting the mix ratio, combining the strength prediction results in different environments and industries to ensure the quality of construction.

Concrete surface crack detection is a critical problem in the health monitoring and maintenance of engineering structures. The existence and development of cracks may lead to the deterioration of structural performance, potentially causing serious safety accidents. However, detecting cracks accurately remains challenging due to various factors such as uneven lighting, noise interference, and complex backgrounds, which often lead to incomplete or false detections. Traditional manual inspection methods are subjective, inefficient, and costly, while existing deep learning-based approaches still have the problem of insufficient precision and completeness. Therefore, this paper proposes a new crack detection model based on an improved TransUNet: AG-TransUNet, an adaptive multi-head self-attention mechanism, and a gated mechanism-based decoding module (GRU-T) is introduced to improve the accuracy and completeness of crack detection. Experimental results show that the AG-TransUNet outperforms the original TransUNet with a 4.05% increase in precision, a 2.59% improvement in F1-score, and a 0.36% enhancement in IoU on the CFD dataset. The AG-TransUNet achieves a 2.21% increase in precision, a 5.63% improvement in F1-score, and a 9.07% enhancement in IoU on the concrete crack dataset. In addition, in order to further quantitatively analyze the crack width, the orthogonal skeleton method is used to calculate the maximum width of a single crack to provide a reference for engineering maintenance. Experiments show that the maximum error between the real values and detection results is about 5%. Therefore, the proposed method better meets the needs of crack detection in practical engineering applications and provides a solution for improving the efficiency of crack detection.