Search

Evaluation of earthquake-induced liquefaction potential is crucial in the design phase of construction projects. Although several machine learning models achieve good prediction accuracy on their particular datasets, they may not perform well in other liquefaction datasets. To address this issue, we proposed a novel hybrid classifier ensemble to improve generalizability by combining the predictions of seven base classifiers using the weighted voting method. The applied base classifiers include back propagation neural network, support vector machine, decision tree, k-nearest neighbours, logistic regression, multiple linear regression and naïve Bayes. The hyperparameters and weights of the base classifiers were tuned using the genetic algorithm. To verify the robustness of the classifier ensemble, its performance was tested on three datasets collected from previous published researches. The results show that the proposed classifier ensemble outperforms the base classifiers in terms of a variety of performance metrics including accuracy, Kappa, precision, recall, F1 score, AUC and ROC on the three datasets. In addition, the importance of influencing variables was achieved by the classifier ensemble on the three datasets to facilitate the future data collecting work. This robust ensemble method can be extended to solve other classification problems in civil engineering. [ABSTRACT FROM AUTHOR]

Summary: Rockbust is a violent expulsion of rock due to the extreme release of strain energy stored in surrounding rock mass, leading to considerable damages to underground strucures and equipment, and threatening workers' safety. As the operational depth of engineering projects increases, a larger number of factors influence the mechanism of rockburst. Therefore, accurate classification of rockburst intensity cannot be achieved based on conventional criteria. It is urgent to develop new models with high accuracy and ease to implement in practice. This study proposed an ensemble machine learning method by aggregating seven individual classifiers including back propagation neural network, support vector machine, decision tree, k‐nearest neighbours, logistic regression, multiple linear regression and Naïve Bayes. In addition, we proposed nine data imputation methods to replace the missing values in the compiled database including 188 rockburst instances. Five‐fold cross validation and the beetle antennae search algorithm are used to tune hyperparameters and voting weights of the individual classifiers. The results show that the rockburst classification accuracy obtained by the classifier ensemble has increased by 15.4% compared with the best individual classifier on the test set. The predictor importance obtained by the classifier ensemble shows that the elastic energy index is the most sensitive input variable for rockburst intensity classification. This robust ensemble method can be extended to solve other classification problems in underground engineering projects. [ABSTRACT FROM AUTHOR]

Tunnel squeezing occurs when time-dependent rock creep produces large tunnel convergence. The occurrence of tunnel squeezing may result in buget increase and time waste during tunnel construction. The aim of this study was to propose a robust classifier ensemble to predict squeezing conditions in rock tunnels. Seven individual machine learning classifiers were aggregated using weighted voting methods to establish the classifier ensemble. The weight and hyperparameters of each individual classifier were tuned using the firefly algorithm. The classifier ensemble was trained and tested on a dataset collected from published literature. Missing values in the database were replaced by various imputation methods. The results indicate that the proposed classifier ensemble achieved an accuracy of 96%, higher than that of the traditionally used individual classifiers. This robust ensemble method can be applied to other classification problems in civil engineering. [ABSTRACT FROM AUTHOR]

Lightweight concrete (LWC) is widely used in the construction industry due to a variety of advantages. However, compared with traditional normal-weight concrete, more influencing variables (e.g. types of lightweight aggregates) must be considered to optimize multiple properties including uniaxial compressive strength (UCS), density and cost. This makes the mixture design of LWC more difficult or sometimes impossible using laboratory experiments. To address this issue, this study proposes a multi-objective optimization (MOO) method using machine learning and metaheuristic approaches for LWC mixture design through a two-step approach. In the first step, a least squares support vector regression (LSSVR) model is constructed to predict multiple properties of LWC. The hyper-parameters of the LSSVR model are tuned using the firefly algorithm (FA). A dataset containing a large number of different mixtures of LWC is compiled from published literature. High prediction accuracy (0.97 for UCS and 0.90 for density) is achieved on the test dataset (including 30% of all the instances). In the second step, a newly developed multi-objective FA (MOFA) model is used to optimize the LWC mixture, while satisfying the constraints. The Pareto fronts of the triple objectives (UCS, cost and density) are successfully obtained. The proposed MOO method is powerful and efficient in finding optimal LWC mixtures with conflicting objectives and therefore decision making can be facilitated in early phases of construction. [ABSTRACT FROM AUTHOR]

• BPNN has good prediction accuracy for UCS, while RF performs better in predicting slump. • PSO is efficient in tuning hyperparameters of machine learning models. • The Pareto front of the mixture optimization problem is obtained by MOPSO. For the optimization of concrete mixture proportions, multiple objectives (e.g., strength, cost, slump) with many variables (e.g., concrete components) under highly nonlinear constraints need to be optimized simultaneously. The current single-objective optimization models are not applicable to multi-objective optimization (MOO). This study proposes an MOO method based on machine learning (ML) and metaheuristic algorithms to optimize concrete mixture proportions. First, the performances of different ML models in the prediction of concrete objectives are compared on data sets collected from the published literature. The winner is selected as the objective function for the optimization procedure. In the optimization step, a multi-objective particle swarm optimization algorithm is used to optimize mixture proportions to achieve optimal objectives. The results show that the backpropagation neural network has better performance on continuous data (e.g., strength), whereas the random forest algorithm has higher prediction accuracy on more discrete data (e.g., slump). The Pareto fronts of a bi-objective mixture optimization problem for high-performance concrete and a tri-objective mixture optimization problem for plastic concrete are successfully obtained by the MOO model. The MOO model can serve as a design guide to facilitate decision-making before the construction phase. [ABSTRACT FROM AUTHOR]

Depletion of river sand due to large-scale concrete production has caused many environmental problems. To address this issue, river sand can be replaced with sand manufactured from waste deposits. To facilitate manufactured-sand concrete production, this study proposes three tree-based models: one individual model (regression tree (RT)), and two ensemble models (random forest (RF) and gradient boosted regression tree (GBRT)) to predict its mechanical properties, such as uniaxial compressive strength (UCS), and splitting tensile strength (STS). These tree-based models were trained and tested on a dataset collected from previous literature. In addition, to understand the importance of each input variable on the mechanical properties of manufactured-sand concrete, the variable importance is calculated using the RF algorithm. The results show that the highest correlation coefficients are achieved by GBRT in predicting UCS (0.9887) and STS (0.9666), which respectively increase by 3.0%–10.8% and 16.0%–21.6% in comparison with the models in previous literature. The mechanical properties UCS and STS are highly sensitive to the curing age with relative importance of 36.8% and 40.3%, respectively. To facilitate the application of the tree-based models in predicting mechanical properties of manufactured-sand concrete, a graphical user interface has been designed in this study. • The mechanical properties of manufactured-sand concrete were predicted by tree-based models. • Firefly algorithm was used to tune the hyperparameters of the tree-based models. • Importance of input variables was calculated using random forest algorithm. • A GUI was developed to facilitate the application of the proposed models. [ABSTRACT FROM AUTHOR]

Prediction of rockburst in underground engineering is of significance as it is close to the safety of support structures, personnel, and working environments. To improve the accuracy of rockburst classification, this paper proposed an ensemble classifier RF-FA model in which the random forest classifier (RF) and firefly algorithm (FA) were combined to achieve the optimum performance on rockburst prediction. Five key parameters of surrounding rock, i.e., the depth H, the maximum tangential stress σθ, the uniaxial compressive strength σc, the tensile strength σt, and the elastic energy index Wet, are selected as input variables while the rockburst intensity including none, light, moderate, and strong classes was chosen as output. A total of 279 cases worldwide were collected and used for train and test the proposed RF-FA model. The results showed that the FA can optimize the hyperparameters of RF efficiently and the optimum model exhibited a high performance on rockburst data from the independent test set and new engineering projects. The feature importance obtained by the ensemble RF-FA model indicated that the elastic energy index plays the most important role in rockburst. Besides, the proposed model showed much better accuracy on rockburst classification compared with previously existing RF models and empirical criteria, which means it is a useful and robust tool for rockburst prediction. [ABSTRACT FROM AUTHOR]

• A GA-RF model was developed for predicting UCS and FST of AAMs; • The curing time and water content significantly influenced the UCS, while Na/Al and water content were more important to FST; • The recommended Ca/Si varied from 1 to 2; Na/Al was slightly lower than 1 and Si/Al ratios changed between 2.5 and 3.5. The final setting time (FST) and uniaxial compressive strength (UCS) are critical parameters for designing the mixture proportions of alkali-activated materials (AAMs). To understand the influence of the mixture composition on FST and UCS of AAMs, two datasets containing 616 samples for UCS and 278 samples for FST were compiled from published literature. A random forest (RF) model was developed on these datasets to predict FST and UCS of AAMs. The hyperparameters of the RF model were optimized using the Genetic Algorithm (GA). Results show that the hybrid GA-RF model achieved the highest prediction accuracy on the test set of UCS (0.932) and FST (0.997), compared to other machine learning models. The developed model was then used to interpret the influence of mixture composition on FST and UCS. The curing time and water content significantly influenced the UCS, while Na/Al and water contents were more important to FST. The microstructure development of the AAMs was affected by Ca/Si, Na/Al and Si/Al ratios. To achieve better UCS, the recommended Ca/Si varied from 1 to 2; Na/Al was slightly lower than 1 and Si/Al ratios changed between 2.5 and 3.5. This study can facilitate the mixture optimization for FA-slag based AAMs. [ABSTRACT FROM AUTHOR]

The precast segmental self-centering concrete-filled steel tube (PSCFST) bridge is not only the ideal choice for fast and environmentally friendly construction but also has good seismic and resilience properties. Our research group has carried out shaking table test research on the PSCFST bridge, but due to the limitation of test equipment and site, no damage test has been carried out. To further study the seismic performance of PSCFST bridges when subjected to larger ground motions, machine learning (ML) models are developed to predict the seismic performance of PSCFST. A novel combined prediction model based on Conv1D-LSTM was proposed to predict the PSCFST bridge acceleration response. Two other commonly used ML methods including XGBoost and Random forest regression (RFR) are also used for comparison purposes. A database of ML prediction models is established based on 116 sets of input ground motion (GM) and superstructure acceleration response from shaking table tests. Then, the data of RSN292 60% GM were selected as the prediction test data. Furthermore, based on the Opensees platform, the PSCFST fiber finite element (FFE) model was established and validated by the shaking table test results, then the dynamic time history analysis of the ground motion with larger amplitude (greater than the input ground motion assignment of the shaking table test) was carried out. The superstructure acceleration response of 70%–120% of the RSN 292 GM is obtained by the FFE model and used as the data set for the ML prediction model. After that, the superstructure acceleration response is obtained through three prediction models. Comparing the simulation and prediction results shows that all the Conv1D-LSTM, XGBoost, and RFR models can reliably predict the acceleration response of the PSCFST bridge. In all cases, the Conv1D-LSTM model performed outperforms the XGBoost and RFR models. The determination coefficients (R 2) of Conv1D-LSTM, XGBoost, and RFR model for the prediction of superstructure response are 0.9643, 0.8780, and 0.9623, respectively. • Proposed a novel combined prediction model based on Conv1D-LSTM to predict the PSCFST bridge acceleration response. • Established the fiber finite element analysis model of PSCFST single span bridge and validated by the shaking table test. • Three AM models are used to predict the acceleration response of the PSCFST bridge under large amplitude ground motions. [ABSTRACT FROM AUTHOR]