AI in civil engineering最新文献

This paper presents an intelligent lithology identification method that utilizes the feature fusion of single polarized and orthogonal polarized rock images. The traditional thin section identification method heavily relies on manual expertise, leading to subjective results and requiring significant time and labor. To overcome these limitations, we establish a microscopic feature fusion model using a convolutional neural network (CNN). This model leverages the complementarity information from single polarized and orthogonal polarized features. By extracting features from microscopic rock images using convolutional kernels and integrating multi-feature information at both the input and feature levels, the proposed method enhances the classification accuracy of the model, providing a more efficient and objective solution for lithology identification. To evaluate the identification performance, several metrics including accuracy (Acc), precision (P), recall (R), F1-score, and a confusion matrix are employed. The results demonstrate that the fusion model achieved a maximum accuracy of 98.66% on the testing set, representing a 4.91% improvement over using single polarized images alone and a 1.55% improvement over orthogonal polarized images alone. The integration of advanced deep learning models with microscopic image analysis techniques enables researchers and non-geologists to automate the identification and classification of extensive rock sample datasets efficiently. Moreover, the proposed method proves particularly useful in cases with complex mineral compositions and similar structures, as it provides more reliable and accurate analytical results.

The demand for construction materials in Pakistan has experienced a significant increase, particularly due to the China-Pakistan Economic Corridor (CPEC) project, which necessitates substantial amounts of resilient resources for infrastructure development. Parameters of rock strength, including uniaxial compressive strength (UCS), Young’s modulus (E), and Poisson’s ratio (ν), are critical attributes of rock materials vital for applications such as rock slope stability assessment, tunnel construction, and foundation design. Conventionally, the measurement of UCS, E, and ν in laboratory settings resource-intensive, requiring considerable time and financial investment. This study proposes to provide a comprehensive assessment framework using an adaptive boosting machine (AdaBoost), extreme gradient boosting machine (XGBoost), and category gradient boosting machine (CatBoost), to indirectly estimate UCS, E, and ν through streamlined mineralogical analyses. The performance of the boosting trees was analyzed using Taylor diagrams and a suite of five regression metrics: coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), variance accounted for (VAF), and the A-20 index. The results indicate that the proposed boosting trees robust predictive capabilities for the constructed database. Notably, AdaBoost demonstrated the highest efficacy in predicting the strength of carbonate rock, achieving R² values of 0.98, 0.99, and 0.97, with the lowest RMSE values of 0.3164, 0.63, and 0.18, for UCS, E, and ν, respectively. Moreover, variable importance analysis highlighted that the presence of micrite and calcite has a significant impact on predicting UCS, E, and ν of carbonate rock. Furthermore, the AdaBoost model was validated using an independent dataset, which corroborated its predictive reliability. In conclusion, the proposed models present a highly effective methodology for the indirect prediction of essential mechanical properties of carbonate rocks, offering substantial time and cost efficiencies compared to traditional laboratory techniques.

Geopolymer concrete is acknowledged as a sustainable alternative to conventional Portland cement concrete owing to its ability to reduce carbon emissions and reutilize industrial by-products. This paper reviews the application of Artificial Intelligence-based Life Cycle Analysis (LCA) techniques in the sustainability assessment of geopolymer concrete. The assessment covers the entire life cycle of geopolymer concrete, spanning from the extraction of raw materials to its ultimate disposal, with a particular focus on its environmental, economic, and social impacts. The incorporation of AI techniques into the LCA process offers notable advantages, such as the efficient management of large datasets, enhancement of data quality, prediction of environmental impacts, and facilitation of informed decision-making. Key sustainability metrics to be considered include environmental impacts such as carbon footprint and energy consumption, economic factors like cost-effectiveness, as well as social implications. The amalgamation of AI within the LCA framework provides a comprehensive and efficient approach to evaluating the sustainability of geopolymer concrete, thereby facilitating its application in sustainable construction practices.

In this study, a comprehensive assessment of slope failure risk in man-made slopes was conducted, focusing specifically on the embankments in the excavated regions along the Tibati-Sengbe road in the Adamawa region of Cameroon. The primary objective of this study was to analyze the stability of these slopes and determine the safety factors that should be considered in their stabilization. To achieve this goal, a field survey was conducted to identify and characterize the areas at risk. The stability assessment was performed employing sophisticated numerical methods, including the Limit Equilibrium Method (LEM) utilizing the Bishop Method, the Finite Element Method (FEM) through the Plaxis Method, and the Analytical Method (AM) based on Taylor's Abacus. Ten slopes with homogeneous soil composition but varying geotechnical and geometric properties were selected as the objects for simulations, which were performed using the software packages ROCSCIENCE (Phase 2) for LEM and PLAXIS for FEM. The results indicated a high degree of consistency between the FEM and LEM methodologies, with an R² correlation approaching 1 in their comparison. Nonetheless, the AM yielded conflicting results in 60% of cases, emphasizing the fundamental significance of numerical methods in evaluating slope stability. The findings of this study discredit the effectiveness of analytical methods in determining safety factor calculations and highlight the accuracy and reliability of the FEM and LEM techniques given their consistent results.

Accurately predicting 3D concrete printing (3DCP) properties through the utilization of machine learning holds promise for advancing cost-effective, eco-friendly construction practices that prioritize safety, reliability, and environmental sustainability. In this study, a comprehensive exploration of seven regression models was undertaken, complemented by the application of Bayesian optimization techniques to forecast critical metrics such as compressive strength, pump speed, and carbon footprint within the realm of 3DCP technology. Drawing upon a compilation of various 3DCP mixtures sourced from existing literature, an intricate carbon footprint calculation methodology was devised, resulting in the establishment of a bespoke database tailored to the study’s objectives. The performance evaluation of the developed models was conducted through the analysis of key statistical indicators, including R², RMSE, MAE, and Pearson correlation. To enhance the robustness and generalizability of the models, a rigorous tenfold cross-validation strategy coupled with a strategic introduction of noise was employed during the validation process. The incorporation of Shapley Additive Explanations (SHAP) analysis provided insightful interpretability into the predictive capabilities of the models, enabling a nuanced understanding of the underlying relationships between input variables and target outputs. Furthermore, the application of multi-objective optimization techniques facilitated judicious decision-making processes, enabling the identification of optimal 3DCP mixture compositions that concurrently enhance performance metrics, reduce operational costs, and mitigate CO₂ emissions.

Effective water management and flood prevention are critical challenges encountered by both urban and rural areas, necessitating precise and prompt monitoring of waterbodies. As a fundamental step in the monitoring process, waterbody segmentation involves precisely delineating waterbody boundaries from imagery. Previous research using satellite images often lacks the resolution and contextual detail needed for local-scale analysis. In response to these challenges, this study seeks to address them by leveraging common natural images that are more easily accessible and provide higher resolution and more contextual information compared to satellite images. However, the segmentation of waterbodies from ordinary images faces several obstacles, including variations in lighting, occlusions from objects like trees and buildings, and reflections on the water surface, all of which can mislead algorithms. Additionally, the diverse shapes and textures of waterbodies, alongside complex backgrounds, further complicate this task. While large-scale vision models have typically been leveraged for their generalizability across various downstream tasks that are pre-trained on large datasets, their application to waterbody segmentation from ground-level images remains underexplored. Hence, this research proposed the Visual Aquatic Generalist (VAGen) as a countermeasure. Being a lightweight model for waterbody segmentation inspired by visual In-Context Learning (ICL) and Visual Prompting (VP), VAGen refines large visual models by innovatively adding learnable perturbations to enhance the quality of prompts in ICL. As demonstrated by the experimental results, VAGen demonstrated a significant increase in the mean Intersection over Union (mIoU) metric, showing a 22.38% enhancement when compared to the baseline model that lacked the integration of learnable prompts. Moreover, VAGen surpassed the current state-of-the-art (SOTA) task-specific models designed for waterbody segmentation by 6.20%. The performance evaluation and analysis of VAGen indicated its capacity to substantially reduce the number of trainable parameters and computational overhead, and proved its feasibility to be deployed on cost-limited devices including unmanned aerial vehicles (UAVs) and mobile computing platforms. This study thereby makes a valuable contribution to the field of computer vision, offering practical solutions for engineering applications related to urban flood monitoring, agricultural water resource management, and environmental conservation efforts.

A number of bridges have collapsed around the world over the past years, with detrimental consequences on safety and traffic. To a large extend, such failures can be prevented by regular bridge inspections and maintenance, tasks that fall in the general category of structural health monitoring (SHM). Those procedures are time and labor consuming, which partly accounts for their neglect. Computer vision and artificial intelligence (AI) methods have the potential to ease this burden, by fully or partially automating bridge monitoring. A critical step in this automation is the identification of a bridge’s structural components. In this work, we propose an extensible synthetic dataset for structural component semantic segmentation of portal frame bridges (PFBridge). We first create a 3 dimensional (3D) generic mesh representing the bridge geometry, while respecting a set of rules. The definition of new, or the extension of the existing rules can adjust the dataset to specific needs. We then add textures and other realistic elements to the model, and create an automatically annotated synthetic dataset. The synthetic dataset is used in order to train a deep semantic segmentation model to identify bridge components on bridge images. The amount of available real images is not sufficient to entirely train such a model, but is used to refined the model trained on the synthetic data. We evaluate the contribution of the dataset to semantic segmentation by training several segmentation models on almost 2,000 synthetic images and then finetuning with 88 real images. The results show an increase of 28% on the F1-score when the synthetic dataset is used. To demonstrate a potential use case, the model is integrated in a 3D point cloud capturing system, producing an annotated point cloud where each point is associated with a semantic category (structural component). Such a point cloud can then be used in order to facilitate the generation of a bridge’s digital twin.

The use of agricultural by-products, such as Rice Husk Ash (RHA), in concrete production has gained significant attention as a sustainable alternative to traditional construction materials. This study aims to evaluate and compare the effects of Nano-Rice Husk Ash (NRHA) and Micro-Rice Husk Ash (MRHA) on the compressive strength of concrete. Concrete samples were prepared with varying replacement levels of NRHA (0% to 3%) and MRHA (0% to 14%) and underwent thorough examination through both slump and compressive strength tests conducted at 7, 21, 28, and 56 days. The results showed that NRHA achieved maximum compressive strength at a 1% replacement level, while MRHA reached its peak at a 0.5% replacement level. However, a comparison of the compressive strength of NRHA at 1% (22 N/mm²) against MRHA at 0.5% (21.5 N/mm²) revealed that the marginal difference in strength made MRHA a more cost-effective option due to the lower expenses involved in its preparation. Thus, MRHA presents a more economical solution for achieving comparable compressive strength. Furthermore, the study applied linear, non-linear, and mixed regression analyses to model the properties of NRHA and MRHA concrete based on a comprehensive set of variables. The analysis found that the blended ordinary and logarithmic models provided the best fit, offering superior accuracy compared to linear and non-linear models.

Traditional methods for proportioning of high-performance concrete (HPC) have certain shortcomings, such as high costs, usage constraints, and nonlinear relationships. Implementing a strategy to optimize the mixtures of HPC can minimize design expenses, time spent, and material wastage in the construction sector. Due to HPC's exceptional qualities, such as high strength (HS), fluidity and resilience, it has been broadly used in construction projects. In this study, we employed Generalized Regression Neural Network (GRNN), Nonlinear AutoRegressive with exogenous inputs (NARX neural network), and Random Forest (RF) models to estimate the Compressive Strength (CS) of HPC in the first scenario. In contrast, the second scenario involved the development of an ensemble model using the Radial Basis Function Neural Network (RBFNN) to detect inferior performance of standalone model combinations. The output variable was the 28 Days CS in MPa, while the input variables included slump (S), water-binder ratio (W/B) %, water content (W) kg/m³, fine aggregate ratio (S/a) %, silica fume (SF)%, and superplasticizer (SP) kg/m³. An RF model was developed by using R Studio; GRNN and NARX-NN models were developed by using the MATLAB 2019a toolkit; and the pre- and post-processing of data was carried out by using E-Views 12.0. The results indicate that in the first scenario, the Combination M1 of the RF model outperformed other models, with greater prediction accuracy, yielding a PCC of 0.854 and MAPE of 4.349 during the calibration phase. In the second scenario, the ensemble of RF models surpassed all other models, achieving a PCC of 0.961 and MAPE of 0.952 during the calibration phase. Overall, the proposed models demonstrate significant value in predicting the CS of HPC.

This study is primarily aimed at creating three machine learning models: artificial neural network (ANN), random forest (RF), and k-nearest neighbour (KNN), so as to predict the crippling load (CL) of I-shaped steel columns. Five input parameters, namely length of column (L), width of flange (b_f), flange thickness (t_f), web thickness (t_w) and height of column (H), are used to compute the crippling load (CL). A range of performance indicators, including the coefficient of determination (R²), variance account factor (VAF), a-10 index, root mean square error (RMSE), mean absolute error (MAE) and mean absolute deviation (MAD), are used to assess the effectiveness of the established machine learning models. The results show that all of the three ML (machine learning) models can accurately predict the crippling load, but the performance of ANN is superior: it delivers the highest value of R² = 0.998 and the lowest value of RMSE = 0.008 in the training phase, as well as the highest value of R² = 0.996 and the smaller value of RMSE = 0.012 in the testing phase. Additional methods, including rank analysis, reliability analysis, regression plot, Taylor diagram and error matrix plot, are employed to assess the models’ performance. The reliability index (β) of the models is calculated by using the first-order second moment (FOSM) technique, and the result is compared with the actual value. Additionally, sensitivity analysis is performed to check the impact of the input variables on the output (CL), finding that b_f has the greatest impact on the crippling load, followed by t_f, t_w, H and L, in that order. This study demonstrates that ML techniques are useful for developing a reliable numerical tool for measuring the crippling load of I-shaped steel columns. It is found that the proposed techniques can also be used to predict other kinds of failures as well as different kinds of perforated columns.