The main aim of the current paper is to optimize the seismic design of steel moment-resisting frames, with a focus on minimizing life-cycle costs. An efficient and powerful Physics-based metaheuristic algorithm, known as the center of mass optimization, is utilized to search the design space of steel moment-resisting frames' performance-based design optimization problem. Four seismic design optimization scenarios are examined, considering initial cost and seismic total cost as the objective functions to be minimized, with plastic hinge rotation constraints and inter-story drift constraints as the local and global indicators of structural nonlinear behavior, respectively. Within the context of this paper, seismic total cost is formulated as the simple sum of the initial cost and seismic life-cycle cost of the structures. Two design examples, comprising 6-story and 12-story steel frames, are illustrated. The obtained numerical results indicate that optimizing seismic total cost with constraints on both plastic hinge rotation and inter-story drift yields the most cost-effective designs. Additionally, the findings indicate that in optimization processes including both kinds of design constraints, the inter-story drift constraints dominate the optimal designs.
Triple friction pendulum bearings (TFPBs) enhance the seismic resilience of base-isolated buildings. However, optimizing TFPB designs is computationally expensive, necessitating surrogate modeling techniques. This study utilizes structural modeling through OpenSees, while the optimization process, employing the Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES), is conducted in MATLAB for the design of TFPB systems. The primary objective is to minimize peak floor acceleration and inter-story drift. A Support Vector Regression (SVR) surrogate simulates the seismic response of models subjected to synthetic earthquake records tailored to site-specific hazards. To closely investigate the SVR method, three kernel functions (Gaussian, Matérn 3/2, and Matérn 5/2) were compared. SVR surrogate modeling reduced computational costs by 77.78%, with prediction differences of less than 2.71% for both objectives, demonstrating high fidelity. Additionally, no single kernel function consistently outperforms others across all optimization problems within the SVR method. In this study, the Gaussian kernel yielded superior results for the peak floor acceleration objective, while the Matérn 5/2 kernel provided better performance for the inter-story drift objective. This indicates that the choice of kernel function may need to be tailored to specific optimization goals.
This research was conducted to explain the drivers and consequences of applying artificial intelligence (AI) approaches through MaxQDA and the best-worst method in the sustainable supply chain of the civil engineering projects. This study aims to be applied research. The participants of both qualitative and theoretical parts of the study were 17 individuals from senior consultants from the construction industry. They have been selected purposefully. Data gathering has been conducted by semi-structured interviews and comparative questionnaires. The reliability of the questionnaire has been assessed by means of non-compliance rates. Qualitative analysis has been performed in order to explain the drivers and consequences of applying AI approaches in the sustainable supply chain of the civil engineering projects. The factors were prioritized through the best-worst method. Findings revealed that the drivers and consequences of applying AI approaches in sustainable supply chains in the civil engineering projects are hardware infrastructures, software infrastructures, non-technical factors (managerial factors), competitiveness, and supply chain sustainability. The first index in ranking was non-technical factors with a weight of 0.344. On the other hand, the second index was sustainability of the supply chain with a weight of 0.267. The third index was software infrastructures, followed by hardware infrastructures in fourth place with weights of 0.229 and 0.122, respectively, and the supply chain competitiveness index was ranked last place with a weight of 0.038.
Earthquakes are among the most devastating natural disasters, and accurately forecasting their magnitude is critical for reducing their impact on life and property. In this study, the effectiveness of advanced machine learning techniques for improving earthquake magnitude prediction is examined using comprehensive seismic data. A hybrid methodology was developed that integrates machine learning models with metaheuristic feature selection methods to enhance accuracy and robustness. Feature selection was performed using Genetic Algorithm, Particle Swarm Optimization, and Simulated Annealing, while ten machine learning models were implemented — ranging from Linear Regression and Decision Trees to Gradient Boosting, XGBoost, LightGBM, and Long Short-Term Memory (LSTM) networks. An ensemble approach was proposed, incorporating Voting and Stacking Regressors to leverage model diversity for improved performance. The Stacking Regressor outperformed all other models with a minimum RMSE of 1.75, MAE of 1.35, and an R-squared value of 0.94. These results demonstrate the potential of ensemble techniques for accurate and reliable earthquake magnitude forecasting, which may contribute significantly to disaster preparedness and risk mitigation.
Over the past 20 years, numerous experimental and numerical studies have been conducted to understand the bond characteristics of externally bonded (EB) FRP-intact (heat-damaged) concrete joints. Consequently, a large database became available to develop models for bond-strength prediction at these joints. An artificial neural network (ANN)-bond model was built, trained and tested using MATLAB®, utilizing more than 500 data points before being statistically analyzed to substantiate its validity for field predictability. The ANN model converged fully at 405 epochs with natural distribution of training and testing data noticed. The use of fourteen hidden layers provided the least prediction error. The performance of the first bond model to consider the impact of elevated temperature was compared to that of well-known literature models The present model demonstrated higher prediction superiority over the different literature models as indicated by the present detailed statistical and sensitivity analyses. For example, using a group of prediction data, the coefficient of determination (R2) and the root mean square error for the present model attained their highest and lowest values at 0.88 and 0.15, respectively, compared to those of the other models tested. The present model captures the trend behavior of bond strength versus the key parameters; namely, compressive strength, maximum aggregate size of concrete, FRP thickness, elastic modulus, as well as FRP bond length and width ratio. The model reflected higher sensitivity to exposing concrete to elevated temperatures than that by the different literature models. The impact of the key parameters indicated becomes marginal when concrete is subjected to temperatures in excess of 400oC. To enable an ease use of the developed model, an empirical equation for bond strength is provided in this work in terms of the different parameters.
Liquefaction-induced lateral displacements (LILD) poses a significant challenge in geotechnical engineering due to its potential to cause extensive structural damage and ground instability. Consequently, predicting these lateral movements is vital for evaluating the effects of earthquakes on buildings and infrastructures in regions susceptible to liquefaction. Reliable predictions enable engineers and geotechnical specialists to enhance safety measures in the design of structures and reduce risks linked to soil failure. This research employs the Gradient Boosting Regression Tree (GBRT) approach, tuned through the Equilibrium Optimization Algorithm (EOA), to estimate LILD. A dataset containing 247 data points was used to build the predictive model. In contrast to earlier studies that incorporated all seven available variables for prediction, this work utilized Neighborhood Component Analysis (NCA) to identify the four most impactful parameters for modeling. The resulting R2 values for the training and testing datasets were 0.9974 and 0.9514, respectively, reflecting the model’s strong predictive capability for LILD. Comparisons with prior models highlight the enhanced performance of the GBRT approach in predicting LILD. Furthermore, the sensitivity analysis indicated that the free surface ratio exerts the most substantial influence on the prediction outcomes.
The utilization of building information modeling (BIM) to raise efficiency and overcome the construction industry's difficulties has grown in popularity. This technique will bring numerous benefits for the interested parties, including contractors, and employers. Despite its countless benefits, this technology faces some difficulties due to its underlying challenges. This paper focuses on the challenges faced by the BIM technique. To study these difficulties, two surveys are prepared, one about the Delphi technique to detect the criteria and the other one regarding the DANP (DEMATEL-based analytic network process) technique for modeling. By interviewing and investigating similar studies and using the Delphi technique, 39 criteria have been found and approved, the noticed criteria were identified, and modeled by using the DANP technique and finally, the relation among criteria, sub-criteria, and their internal relations are studied. Among the criteria, the financial criteria are ranked first, and among the sub-criteria, the high cost of hardware, software, and BIM tools is ranked first. Therefore, to reduce the negative effects of financial barriers, the employer should take responsibility for the implementation of BIM, and the use of BIM should be facilitated by tax reduction by the government and the use of insurance policies.
Self-compacting mortar and concrete are high-performance building materials used in the construction industry because of their excellent rheological and mechanical properties. However, the absence of specific standards for mix design presents hindrance for researchers, motivating this study. A prediction model was developed in this study to assess the suitability of mix designs to produce robust and stable SCC with desired viscosity and yield stress characteristics. Utilizing artificial neural network technique, a powerful machine learning tool for solving complex nonlinear problems, bibliographic and experimental data on composition proportions and material properties were collected. The model architecture was optimized through multiparametric analysis, testing around 22,000 models to achieve approximately 85% prediction accuracy. The particle size distribution of fine aggregates, along with the content and specific surface area of fine filler materials, emerged as the most significant predictive variables. This model could serve as a reliable tool for researchers and industries to design self-compacting mixtures, conserving laboratory time, as well as financial and natural resources.
The present study provides a novel strategy to find the reliability of soil slopes by optimizing ANFIS with GA, FFA and PSO. These three hybrid models are initialized with 206 datasets through MATLAB. The data sample is splitted in 30:70 for testing and training during model processing. The obtained model results are verified through regression plot, 36 statistical indices, Rank value, Taylor diagram and uncertainty analysis. R2 values found in regression for above mentioned models in training are 0.7624, 0.7011, and 0.7378 whereas in testing are 0.8142, 0.6720 and 0.7013 respectively. Some of the statistical errors such as Mean Square Error (MSE) values were 0.0148, 0.0182, 0.0159 in training and 0.1263, 0.0277 and 0.1289 in testing. Again the Root Mean Square Error (RMSE) values were found to be 0.1216, 0.1349, 0.1263 in training and 0.1256, 0.1664 and 0.1590 in testing. Furthermore, Mean Absolute Error (MAE) values were 0.0912, 0.0978, 0.0902 in training and 0.0968, 0.1283, and 0.1169 in testing. Such errors appear to have been close to zero. The total scores calculated for the hybrid models are 160,117, 141 and ranked the models as 1st, 3rd and 2nd, respectively. The results showed that ANFIS-GA is more efficient and accurate than the ANFIS-PSO model followed by ANFIS-FFA for computation of reliability in soil slopes. Furthermore, it may be suggested that ANFIS hybrid models might be a useful tool for solving slope stability problems.
Structural Health Monitoring (SHM) technics have attracted vast amounts of attention among infrastructural managers for evaluating the structure''s health state and making an appropriate decision in case of emergency. The present study is aimed at proposing three novel objective functions using modal structural characteristics, including modal strain energy (MSE), generalized flexibility matrix (GFM), and mode shapes for structural damage detection. Modal assurance criterion (MAC) was also employed to form the objective functions. Moreover, three metaheuristic optimization algorithms (EO, Jaya, and TLBO) were implemented to execute the model updating process. The goal is to determine the most precise combination of the objective functions and optimization algorithms by examining them using three structural numerical models, comprising a 29-element steel planar truss, a 28-element steel planar frame, and a 25-element steel spatial truss under multiple damage scenarios. Moreover, the performance of the mentioned optimization algorithms was compared in terms of accuracy, stability and convergence speed by considering the impact of noisy and limited modal data. The obtained results indicated that the objective function formulated using MSE and mode shapes produced the most reliable outcomes for damage identification in all examples. Moreover, TLBO algorithm was introduced as the most precise optimization tool in most cases.
Reliability and safety evaluation is a significant topic in structural engineering. The main issues in structural reliability assessment are the excessive computational cost as well as the accuracy. Artificial neural network (ANN) can be used for structural reliability assessment. The ANN used in this article is a multilayer perceptron network (MLP) type. This study aims to evaluate the reliability of truss structures using MLP. In order to train and test the neural network, a database is created for the problem. Truss samples are generated based on a uniform distribution of optimal truss sections. The probability of failure in each truss sample is calculated using the Monte Carlo simulation, taking into account the normal distribution of random variables such as the cross-sectional area of the bars and the applied load. The limitation of node displacement is considered as a limit state function. The data was split as 60% for training and 40% was used for testing and validation. The optimal number of neurons in each layer is determined through a trial-and-error process, based on the lowest error of the predicted data and the highest regression coefficient of responses. Finally, the probability of failure of three benchmark truss structures is calculated as numerical examples using the MLP and compared with the values obtained from simulation. It has been shown that after training and preparing the MLP neural network, the accuracy of the MLP prediction process is proportional to 106 and 103 interactions for MCS and LHS, respectively.
For the contractor to achieve maximum benefit, it is always necessary to minimize total project cost. Thus, a cost estimation process should be established to estimate the cost of completing the project with the required precision. Furthermore, construction projects are often carried out within a complex and dynamic environment where a number of interconnected factors can lead to uncertainty. To address these complexities, this study introduces a comprehensive approach for estimating construction project costs. Fifteen key risk factors and cost-influencing factors are extracted and identified based on a literature survey and experts' judgment. Different approaches, encompassing artificial neural network (ANN), adaptive neuro fuzzy inference system (ANFIS), and regression, are used to estimate the project cost. Data envelopment analysis (DEA) is used to ascertain the most influential factors and risks on the total project cost. Data from 48 construction projects are collected, and DEA is effectively employed to evaluate the factors. The results of DEA and correlation analysis show that eight significant factors affect project cost, including number of project team members, risk of fluctuation in material price, project size, labor productivity, consultant experience, labor availability, and material monopoly. ANN is selected as the preferred method for project cost estimation with the minimum MAPE (Mean Absolute Percentage Error). Sensitivity analysis is performed to demonstrate the applicability of DEA in identifying the influential factors. The results indicate that the proposed ANN model, incorporating eight factors, predicts project costs with an acceptable margin of error.
This paper compares the damage identification outcomes of the Machine Learning (ML) and Deep Learning (DL) algorithms. The algorithms in both approaches have employed vibration data from the benchmark railway bridge KW51. The One-Dimensional Convolutional Neural Network (1D CNN) model is exploited in the DL algorithm to classify the response measurements. The 1D CNN classification algorithm is compared with a statistical-based ML model and another CNN model. The alternative classification model uses human-derived damage-sensitive features extracted from Principal Components Analysis (PCA) in the supervised Linear Discriminant Analysis (LDA) method. The metrics of the confusion matrix are applied as a reference for arbitrating the healthier classification between the two models. The classification accuracy of the 1D CNN algorithm varies from 90% (when acceleration measurements in the lateral direction, y-axis are used) to 100% (when measurements in the vertical direction z-axis are used). In contrast, the accuracy of the comparative ML-based classification approach varies from 66% (when acceleration measurements in the lateral direction y-axis are used) to 94.4% (when data from two sensors are jointly merged). In 1D CNN model, the maximum value of Type I error (False Positive FP) reaches 4.8%, while it reaches 14.4% in ML model. The maximum value of Type II error (False Negative FN) was 8.9% for data from accelerometer aBD23Ay, whereas it was 4.8% in similar case studies analyzed by the 1D CNN method. Once again, in both ML and DL models, the error is more relatively palpable when acceleration measurements in the lateral direction, y-axis are used.
In this work, an inverse method for damage detection studies is developed based on Planet Optimization Algorithm (POA). To demonstrate effectiveness the POA, numerical investigations are implemented on the bivariate Michalewicz function, and 15 functions from the CEC2014 benchmark in a 30-dimensional space (including a comparison between POA and several well-known candidates). The achieved results illustrate that in terms of stability, robustness, and quality of the obtained solution, POA is one of most outstanding optimization algorithms. POA ranks No. 1 in functions F1, F4, F7, F8, F13, while POA is highly ranked and sufficiently competitive with the other contenders in the rest of tests. Based on these considerations, for the first time, an application for multi-damage detection of steel roof truss systems using POA is presented herein. Through comparison with other famous algorithms, POA has outperformed computational cost with fast convergence speed. POA has only, respectively, 99.5%, 98.6%, and 97.2% time-consuming for computational cost when compared with GWO, AOA, and PSO. Also, the results proved that this technique provides an efficient solution to the complex problem with many constraints in unknown search space.
The maintenance and rehabilitation of flexible pavements are crucial for achieving optimal performance and ensuring higher quality, enabling transportation planners to promptly formulate economically viable and sustainable pavement maintenance and rehabilitation strategies. Employing the fuzzy logic technique constitutes a productive methodology for assessing the degradation of flexible pavement. The fuzzy technique offers a convenient instrument for integrating subjective analysis uncertainty within the International Roughness Index (IRI) and evaluating maintenance requirements. This paper strives to construct a system rooted in fuzzy logic to appraise the requirements for maintenance and (IRI) evaluation within a network of pavement roads. This system utilizes data on pavement distress collected from the United States and Canada to achieve its objectives. Various types of pavement distress, such as fatigue cracking, rutting, longitudinal cracking, block cracking, transverse cracking, patching, ravelling, and potholes, are input variables; these parameters are fuzzified into fuzzy subsets with triangular membership functions. The performance evaluation of the analytical models was conducted using several performance indicator metrics, including the coefficient of determination (R2), root mean squared error (RMSE), and mean absolute error (MAE).
Using a collar is a suitable way to protect structures in the water and reduce the scour depth around the structures. In this research, experiments were first carried out to determine the performance of the collar in reducing the scour depth of the spur dikes with different angles, and then, the percentage of scour depth reduction was predicted using intelligent models. The parameters of collar shape coefficient (Sc), the ratio of collar area to the flow depth multiplied by spur dike length (Ac/LY), the ratio of collar height to the flow depth (Zc/Y), and the spur dike angle (ɵ=60, 90, 120) were considered as input data while the reduction of scour depth around the spur dike nose (R) was addressed as the output parameter. The present study seeks to evaluate the application of the Cuckoo search (CS) algorithm and Bat algorithm (BA) to improve the performance of the Support Vector Regression (SVR) model in predicting the amount of R. The results indicated that the proposed hybrid models provided better accuracy than the simple SVR model. In all groups and studied models, the best performance was related to the Support Vector Regression–Bat Algorithm (SVR-BA) and the weakest performance was related to the SVR model. For example, in the combination of all four input data, the amount of root mean square error (RMSE) for SVR-BA was approximately 24% lower and the amount of squared correlation coefficient (R2) was approximately 2% higher, compared to SVR model.
Traditional solar panel installation monitoring methods are often resource-intensive and error-prone, requiring a more effective approach. This study introduces an innovative solution to the challenge of efficiently monitoring large-scale solar panel installations. This is addressed by implementing a deep learning-based model using Mask Region-based Convolutional Neural Networks (Mask RCNN) to automate the detection of solar panels from time series high-resolution drone imagery. The proposed methodology involves data preprocessing, where drone images are georeferenced. The model was trained and validated on a limited collected dataset in diverse solar panel configurations from the first acquired image. The model achieved, on average, an accuracy of 96.25% with an accuracy of detected solar panels in four consecutive images acquired on four different dates as follows: 0.97, 0.95, 0.97, and 0.96.". It was particularly effective in identifying solar panel expansions over time– a clear indicator of its capability to monitor incremental changes effectively. The developed model improves efficiency and accuracy in solar panel monitoring, reduces operational costs, and adapts to various geographic and environmental conditions. Additionally, the automated process significantly reduces the time and labor involved in manual monitoring. Despite its advantages, the model's limitations include high computational demand during training and sensitivity to environmental factors, such as dust accumulation and image quality variances. These challenges necessitate robust computational resources and an initial investment in advanced drone technology. In conclusion, the developed deep learning model presents a practical tool for solar panel detection, offering substantial improvements in monitoring and managing solar energy resources.
Recently, artificial intelligence (AI) has been widely adopted in the design and analysis of passively controlled structures. This study provides a comprehensive overview of AI-based methodologies aimed at enhancing the resilience and efficiency of buildings and infrastructures against environmental stresses and natural disasters. Despite extensive research on passive control systems like base isolators and energy dissipators, a detailed review focusing on AI's application in optimizing these systems is still lacking. This research addresses this gap by exploring the potential of AI to improve the design and performance of passive control systems. Traditional design methods often fail to adequately address the complex interactions between environmental forces and passive control systems, resulting in suboptimal safety and performance. By leveraging AI's predictive and analytical capabilities, this study highlights insights and strategies for enhancing the effectiveness and reliability of passively controlled structures. The findings show the AI's role in advancing structural engineering, promoting the development of smarter, adaptive building infrastructure, and ultimately contributing to safer, more resilient communities in the face of environmental challenges.
This study proposes a broad investigation into the application of advanced ML techniques for anticipating the compressive load-bearing capacity of FRP-confined concrete columns. The methods include the Extremely Randomized Tree (ERT), Random Forest (RF), Gaussian Process Regression (GPR), and Back Propagation Neural Network (BPNN). The dataset consisted of 567 specimens, encompassing rectangular specimens confined by assorted types of Fiber-Reinforced Polymer (FRP) sheets. The results demonstrate the remarkable potential of these ML methods in accurately forecasting the confined compressive strength. The BPNN model emerged as the top performer, achieving the lowest RMSE of 1.3216 and the highest R2 of 0.96 on the test dataset. The GPR model also exhibited predictive solid capabilities, with the second-lowest RMSE and second-highest R2. The performance of these ML models was enhanced by optimizing them using the Marine Predators Algorithm (MPA). The BPNN-MPA, RF-MPA, and GPR-MPA models outperformed the standalone ML methods, showcasing RMSE values as low as 1.1173 and R2 values as high as 0.98 on the full dataset. The reliability analysis shows the superior performance of the BPNN-MPA and RF-MPA models, with reliability scores of 0.95 and 0.93, respectively. The findings underscore the transformative potential of AI-driven approaches in accurately predicting the FRP-confined compressive strength, a critical parameter for the design and analysis of structural retrofitting systems. The MPA optimizer played a crucial role in fine-tuning the hyperparameters of the ML models, leading to significant improvements in their predictive performance. While the study is primarily focused on the specific engineering problem of FRP-confined concrete columns, the proposed framework can be extended to a broader range of structural engineering challenges.
Predicting shear capacity of Fiber Reinforced Polymer (FRP) concrete beams is challenging due to multiple influencing parameters. The research uses an experimental dataset of 48 and 73 results from carbon and glass Fiber Reinforced Polymer (FRP) concrete beams to create deep neural network (DNN) models for predicting shear capacity. A novel feature of this research is that hyperparameter optimization is used to determine the optimal sets of hyperparameters. Three distinct DNN models have been developed to predict shear capacity. The first two models, DNN1 and DNN2, are designed to predict the shear capacity of glass FRP and carbon FRP concrete beams, respectively. The third model, DNN3, is a generic model that can predict the shear capacities of both types of beams. Additionally, three power form nonlinear regression models (NLR1, NLR2, and NLR3) were created for comparison with the DNN models. The DNN1 model outperformed the NLR1 model in terms of MSE, while DNN2 achieved the lowest MSE of 57.45 for training and 77.26 for testing, indicating better performance compared to the NLR2 model, with MSE values of 62.54 and 93.2 respectively. The results indicate that DNNs performed better than regression models in predicting shear capacity.
