Search

Civil infrastructure relies heavily on structural health monitoring systems. However, these systems often encounter challenges due to sensor failures and environmental damage. Consequently, numerous anomalous data points are generated, significantly distorting the accuracy of structural safety assessments. While deep neural networks have emerged as a promising tool for efficiently identifying abnormal data, the meticulous optimization of hyperparameters during training remains a challenge. To address this challenge, this paper introduces a novel approach termed multiple transfer learning, designed to continually enhance a model's classification performance without the need for meticulous hyperparameter configurations. This methodology achieves adaptive training by iteratively migrating across bridge anomaly datasets, bypassing the need for elaborate hyperparameter setting. In this study, five distinct hyperparameter working conditions are established and evaluated to validate the effectiveness of the multiple transfer learning method. The findings highlight the robustness of this approach, demonstrating that multiple transfer learning achieves satisfactory recognition accuracy levels irrespective of the initial hyperparameter setting during network model training. This method circumvents the need for continuous hyperparameters optimization, enabling the adaptive detection of abnormal bridge data. • A training method called multiple transfer learning based on transfer learning is proposed in this paper to solve setting hyperparameters. • The relationship between the number of multiple transfer learning times and the accuracy of anomaly data identification is displayed. • Comparison experiments with different working conditions were set up to verify the effectiveness of multiple transfer learning. [ABSTRACT FROM AUTHOR]

This paper proposes a data-driven model for predicting the code-based seismic vulnerability index calibrated on a dataset comprising almost 300 buildings. The vulnerability index, estimated following the Italian Seismic Code, involved rigorous investigations, including geometric surveys, experimental tests, and numerical modelling. Leveraging data from these investigations, encompassing approximately 15 categorical and numerical explanatory variables, the authors developed several regression and classification predictive models, such as Logistic Regression and Artificial Neural Networks (ANN). The optimal models perform binary classification to determine the categorization into two macro-classes, defined by an arbitrary vulnerability threshold. The ANN model stands out as the best performer. When adjusting the vulnerability threshold to obtain a balanced dataset, such a model achieves an accuracy higher than 85%. The paper also discusses the importance of each feature by calculating the SHapley Additive exPlanations (SHAP) values. The proposed model can aid decision-makers in allocating resources effectively to mitigate seismic risks of built environments. • Data-driven model for code-based seismic vulnerability assessment. • 15 explanatory variables are considered. • Regression models perform poorly. • Binary classification is the only reliable approach. • The ANN model has an 85% accuracy on binary classification with balanced dataset. [ABSTRACT FROM AUTHOR]

• Develops a deep Bayesian survival model that integrates deep neural network and Bayesian inference into the survival analysis for rail useful lifetime modelling. • Demonstrates survival analysis is an effective way to handle censored data problems in rail useful lifetime modelling. • Conducts extensive evaluation to verify the effectiveness of this model, using Concordance index, root mean square error, and a proposed CW-index. • The estimated lifetimes obtained by this model are the closest to real lifetimes compared to other commonly used models. • This model can provide the confidence interval of the estimated lifetime, which is more useful than point estimates in decision-making and maintenance planning of railroad systems. Reliable estimation of rail useful lifetime can provide valuable information for predictive maintenance in railway systems. However, in most cases, lifetime data is incomplete because not all pieces of rail experience failure by the end of the study horizon, a problem known as censoring. Ignoring or otherwise mistreating the censored cases might lead to false conclusions. Survival approach is particularly designed to handle censored data for analysing the expected duration of time until one event occurs, which is rail failure in this paper. This paper proposes a deep Bayesian survival approach named BNN-Surv to properly handle censored data for rail useful lifetime modelling. The proposed BNN-Surv model applies the deep neural network in the survival approach to capture the non-linear relationship between covariates and rail useful lifetime. To consider and quantify uncertainty in the model, Monte Carlo dropout, regarded as the approximate Bayesian inference, is incorporated into the deep neural network to provide the confidence interval of the estimated lifetime. The proposed approach is implemented on a four-year dataset including track geometry monitoring data, track characteristics data, various types of defect data, and maintenance and replacement (M&R) data collected from a section of railway tracks in Australia. Through extensive evaluation, including Concordance index (C-index) and root mean square error (RMSE) for evaluating model performance, as well as a proposed CW-index for evaluating uncertainty estimations, the effectiveness of the proposed approach is confirmed. The results show that, compared with other commonly used models, the proposed approach can achieve the best concordance index (C-index) of 0.80, and the estimated rail useful lifetimes are closer to real lifetimes. In addition, the proposed approach can provide the confidence interval of the estimated lifetime, with a correct coverage of 81% of the actual lifetime when the confidence interval is 1.38, which is more useful than point estimates in decision-making and maintenance planning of railroad systems. [ABSTRACT FROM AUTHOR]

• This paper proposes a novel monocular vision based 3-dimentional displacement measurement approach for structures. • A stationary monocular camera measurement system is designed to achieve 3-dimensional displacement measurement. • State-of-the-art deep learning approaches are adopted to realize the depth direction displacement measurement. • Two tests are conducted on steel cantilever beams in a laboratory to validate the proposed measurement system. Vibration displacement of civil structures are crucial information for structural health monitoring (SHM). However, the challenges and costs associated with traditional physical sensors make displacement measurement difficult. In recent years computer vision (CV) techniques have been employed for measuring vibration displacement in civil structures. There has been a growing interest in CV-based three-dimensional (3D) displacement measurement, as it provides comprehensive information for structural health assessment. Most existing methods use multi-view geometry, requiring multiple cameras for depth measurement. This paper proposes a new system for measuring the 3D vibration displacement utilising a single camera. Instead of using multi-view geometry, deep neural networks are utilised to learn the depth of scenes from monocular images. Compared with the multi-view methods, the proposed 3D measurement system with monocular vision is more cost-effective and much more convenient to set up and use in practice, avoiding the complicated calibration and object matching between multiple cameras. Experimental tests are conducted in the laboratory to investigate the feasibility of the proposed system. Physical displacement sensors are equipped with the testing structure to provide the ground truth data. The results demonstrate that the proposed monocular 3D displacement system is able to produce reasonable 3D full-field displacement measurement, which makes monocular image based CV system a promising approach to achieve 3D displacement measurement, with its obvious advantages in cost and convenience compared to the traditional sensor-based or multi view CV-based methods. [ABSTRACT FROM AUTHOR]

Cold-Formed Steel Lipped (CFSL) channels are susceptible to a localized failure mechanism known as web crippling, triggered by concentrated loads or reactions applied to the web of the section. These loads induce buckling and distortion in the web, ultimately leading to the member's collapse. It is a challenging task to accurately determine the web crippling capacity of a CFSL channel due to its complexity and various influencing factors. This paper presents hybrid soft computing techniques for accurately predicting the web crippling capacity of CFSL channels subjected to two flange load cases. The developed soft computing techniques combine Artificial Neural Networks (ANN) with either Genetic Algorithms (GA) or Particle Swarm Optimization (PSO) to improve computational efficiency and accuracy. The finite element models of CFSL channels are developed and validated by experimental results and then employed to generate a database, which is used to train machine learning models, including ANN, GA-ANN, and PSO-ANN. Analysis is undertaken on the reliability of existing design formulas for determining the web crippling capacity of CFSL channels. It is shown that the PSO-ANN model outperforms the other models in terms of prediction accuracy. The existing design codes and formulas are not reliable in estimating the web crippling capacity of CFSL channels. However, the proposed model yields good correlation with finite element analysis results. A user- friendly graphical interface tool is developed for the practical design of cold-formed steel lipped channels. • FE model is developed that simulates the web crippling behavior of CFSL channels. • Hybrid ML models are developed for predicting web crippling capacity of CFSL channels. • Reliability of existing formulas for estimating web crippling capacity is investigated. • A user-friendly graphical interface tool is developed for design purpose. [ABSTRACT FROM AUTHOR]

Constructing an interpretable model for the long-term deformation Structural Health Monitoring (SHM) of earth-rock dams is of great significance for improving the safety state evaluation and monitoring effect. In this paper, a physics-data-driven model for the deformation SHM of earth-rock dams is proposed based on deep mechanism knowledge distillation. Firstly, the deterministic model is established based on the Finite Element Model (FEM) and outputs the hydraulic load component curve and aging component curve. Then a regression prediction model (HTSGAN) between influencing factors and deformation measurements at multiple measurement points is established based on the Graph Convolutional Network (GCN) and attention mechanism. Finally, the Teacher-Hydraulic-Time-Seepage Graph Attention Networks (T-HTSGAN) model is established based on the feature-based multi-teacher knowledge distillation using the knowledge of hydraulic loading physics and soil-rock creep physics of the FEM for mechanism constraints. The model effectively solves the problems of poor model interpretability and lack of physics knowledge constraints in previous earth-rock dam SHM models. The research results are applied to a project of a 185.5-meter-high concrete-faced rockfill dam, and the predictive performance of the model is more effective and stable through the comparison of six baseline models. The comparative analysis of the component curves proves the effectiveness of the proposed knowledge distillation method for mechanism constraints and improves the interpretability of the neural network model. Therefore, the model is more suitable for engineering applications. • A new physics-data-driven deformation monitoring model for earth-rock dams is proposed. • Mechanical constraints of dam SHM model is further improved by a new neural network model mechanism constraint method. • The application results of a high CFRD show that the model has high accuracy and good engineering applicability. [ABSTRACT FROM AUTHOR]

To predict the dynamic response of high-speed trains (HSTs) passing through cable-stayed bridges (CSBs), this paper proposed a prediction framework based on the genetic algorithm (GA) and the neural network. Firstly, an HST-CSB coupled dynamic model was established using the rigid-flexible coupling method. Subsequently, a fused Temporal Convolutional Network and Long Short-Term Memory (TCN-LSTM) neural network model was developed, and its hyperparameters were optimized using the GA. The dataset was created based on the HST-CSB coupled dynamic model for training and testing the model. Additionally, the generalization capability of the model after adopting different track deformations was discussed. The main results were as follows: Comparative analysis with various neural network models revealed that the GA-TCN-LSTM model exhibited the most accurate prediction results across different train speeds. The significant temperature fluctuations of the bridge led to a drastic deterioration of the track smoothness, resulting in decreased predictive performance of the neural network models. The GA-TCN-LSTM model could optimize the hyperparameters, demonstrating good robustness and generalization capabilities. The predictive performance of the model improved when using the German low-interference spectrum, attributed to the greater irregularities in the German low-interference spectrum, resulting in larger train dynamic responses and stronger feature characteristics in the training data for the neural network. Different neural network models exhibited varying adaptability to track irregularities, with the GA-TCN-LSTM model demonstrating high accuracy in predicting dynamic responses under different train speeds. • A coupled dynamic model of high-speed train and cable-stayed bridges is established. • A model combining Temporal Convolutional Networks and Long Short-Term Memory network is established. • The genetic algorithm is employed to optimize the hyperparameters to mitigate the impact of them on model. • The generalization capability of the model under different conditions is discussed the performance. [ABSTRACT FROM AUTHOR]

This paper aims to develop a unique artificial neural network (ANN)-based equation as well as MATLAB- and Python-based graphical user interfaces (GUIs) using the most comprehensive and up-to-date database for predicting the behaviour of axially loaded concrete-filled double skin tubular (CFDST) short and slender columns with normal- and high-strength materials. Two machine learning (ML) methods, which are ANN and extreme gradient boosting (XGBoost), are trained and tested using 1721 sets of data, with 129 of them collected from experimental studies and 1592 generated by finite element (FE) simulations. The accuracy of the developed ML models is assessed through comparing their predictions with the experimental and FE results. To demonstrate the effect of each parameter on the predicted results, the SHapley Additive exPlanations (SHAP) method is used. The developed ML models are also used to conduct parametric studies to examine the effect of geometric and material parameters on the predicted results. The accuracy of the ML models and the proposed ANN-based equation in predicting the ultimate axial capacity of CFDST columns is compared with that of six design methods including two design code provisions and four design equations proposed by researchers. A numerical example is presented to illustrate the design procedure of the CFDST column using the proposed ANN-based equation. The results indicate that the ANN model performs better on unseen data than the XGBoost model with lower root mean square error for the test set. The results also show that the ML models and the proposed ANN-based equation are superior to the other design models in prediction accuracy. • 1721 data samples of CFDST columns are used to train and test the ML models. • Optimized ANN and XGBoost models are used to predict the behaviour of CFDST columns. • ANN-based equation, MATLAB- and Python-based GUIs are developed. • Parametric studies are conducted using the trained ML models. • The superiority of the proposed design models over six design equations is shown. [ABSTRACT FROM AUTHOR]

Several breakthroughs have appeared in different applications due to deep learning which can be applied in Bridge Weigh-In-Motion (BWIM) systems as well. Therefore, deep learning applications should be examined in BWIM systems. However, some deep learning-based solutions have already been proposed in the international literature, they cannot be compared to static methods because available datasets are not appropriate to train and test artificial neural networks. In addition, numerous other aspects have been already considered during tests. In the current paper, a numerically simulated and validated comprehensive dataset is provided. It is designed to meet requirements of incremental development. This dataset makes it possible to benchmark static algorithms and deep learning-based methods on the same dataset. In the second part of the paper, a deep learning-based solution is proposed. The developed solution shows promising results on the provided dataset and also demonstrates the applicability of the dataset. • Deep learning-based algorithm is developed for Bridge Weigh-In-Motion system. • Validated FEA-based influence surfaces and strain gauge data are used as input. • BME-Simulated 1 (BME-S1) dataset with more than 100000 vehicle models is provided. • A BWIM system is developed with class B+ in accordance with COST 323 standard. [ABSTRACT FROM AUTHOR]

Dynamic analysis of forced-vibration systems in civil engineering could be computationally inefficient or even hard to converge if the systems are stiff or highly complicated. Rapid advances in machine learning make it possible to formulate surrogate models for forced-vibration systems using neural networks. The widely used neural networks such as the convolutional neural network (CNN), recurrent neural network (RNN), etc., usually require a constant sampling rate and data length, thus they are difficult to be implemented for real-time calculation of the dynamic system with varying sampling rates. Recently, the continuous-time state-space neural network (CSNN) has shown the capability to lift these restrictions and has been drawing growing attention from the community. In this paper, we propose a generalized CSNN model for various forced-vibration systems (linear and nonlinear). The CSNN model comprises two sets of independent neural networks aiming to compute the state derivative and system response, respectively. Both neural networks adopt linear and nonlinear layers in parallel, instead of only fully connected nonlinear layers as adopted in the literature. This configuration is aimed to enhance the CSNN model with its capability to recognize the linear and nonlinear behaviors of systems. Additionally, the bias options in the CSNN model are all turned off to improve the stability of the model in the long-term time-series forecast, premised on the assumption that the forced-vibration systems are dissipative systems without drift, which is the most common case in civil engineering. Integration on the state derivative at the current time step is executed to obtain the state at the next time step using the explicit 4th-order Runge–Kutta method. Both numerical and experimental illustrative examples are provided, demonstrating that the CSNN model can achieve high performance and training efficiency with a few hyper-parameters, and thus is highly promising for engineering applications. • A novel continuous-time state-space neural network (CSNN) model is proposed for data-driven modeling of forced-vibration linear/nonlinear systems. • The CSNN model does not require any prior knowledge of the system, and is not restricted to the pre-designated sampling rate and input data length. • The CSNN model has much fewer hyper-parameters than traditional neural network models and is unlikely to induce overfitting. • Numerical and experimental studies demonstrate that the CSNN model is superior to LSTM and PhyCNN in modeling both linear and nonlinear systems. [ABSTRACT FROM AUTHOR]

[Display omitted] • An innovative two-stage machine learning framework (TMLF) was proposed. • The uncertainties of material and geometrics were considered. • The strengths of the RHS-CHS T-joint follow the gamma distribution. • ANN model shows the highest accuracy in predicting the RHS-CHS T-joints' strength. • The TMLF can accurately predict the distribution of structural component strength. Machine learning technics have been extensively used in the strength prediction of structural components in recent years, nevertheless, these established strength prediction models usually can not address the inherent uncertainties introduced by the material and geometrics of structural components. This paper intends to propose an innovative two-stage machine learning framework for developing probabilistic strength prediction models of structural components with the consideration of uncertainties of material and geometrics based on limited test results. Rectangular hollow section to circular hollow section (RHS-CHS) T-joints are used as an example for evaluating the proposed framework. To this end, 58,744 training and 19,870 validation datasets are first generated through numerical simulation. Three machine learning algorithms are used and evaluated in this paper to develop the best model for predicting the strength of the RHS-CHS T-joints. The analysis results show that the artificial neural networks (ANN) showed the best generalization performance. The combination of the uncertainties of material and geometrics is considered through the Latin hypercube sampling method. Then, a new strength database considering the randomness of the structural parameters (e.g., material and geometrics) are developed through the trained ANN model and the probabilistic distribution of the strength of the RHS-CHS T-joints is analyzed through the Anderson-Darling test method. The analysis results indicate that the random strengths of the RHS-CHS T-joints follow a gamma distribution while the material and geometrics follow the normal distribution. Finally, the machine learning models for probabilistic strength prediction of the RHS-CHS T-joints are developed and validated. The analysis results indicate that the developed machine learning model can accurately capture the distribution of the strength of RHS-CHS T-joints, confirming the efficiency of the proposed two-stage machine learning framework for developing probabilistic strength prediction models of structural components. A software named "Probabilistic strength prediction of RHS-CHS T-joint" is proposed based on the developed probabilistic strength prediction model for practical application. [ABSTRACT FROM AUTHOR]

• An optimization strategy was conducted to obtain a final set of ANN model. • The developed ANN model exhibited superior performance than current codes and empirical equations. • An explicit equation based on Artificial Neural Networks were derived for practical application. In this paper an Artificial Neural Network (ANN) model is developed for the prediction of the ultimate compressive load of rectangular Concrete Filled Steel Tube (CFST) columns, taking into account load eccentricity. To this end, an experimental database of CFST specimens from the literature has been compiled, totaling 1224 individual tests, both under concentric and under eccentric loading. Except for eccentricity, other parameters taken into consideration include the cross section width, height and thickness, the steel yield limit, the concrete strength and the column length. Both short and long specimens were evaluated. The architecture of the proposed ANN model was optimally selected, according to predefined performance metrics. The developed model was then compared against available design codes. It was found that its accuracy was significantly improved while maintaining a stable numerical behavior. The explicit equation that describes mathematically the ANN is offered in the paper, for easier implementation and evaluation purposes. [ABSTRACT FROM AUTHOR]

• A generative adversarial network (GAN) based method was proposed to establish a synthesized crack image dataset with pixel-wise annotations. • The generated crack annotations and crack images during the training epochs were demonstrated. • Comparative study was conducted to validate the performance of the synthesized crack image dataset for training the crack detection DNN. Crack detection is vital to maintain the structural safety of in-service bridges, which is an increasing demand in the industrial community. The deep learning-based crack detection is an emerging method that provides a novel way to deal with this problem. Although, the deep neural networks (DNN) can learn to detect cracks themselves, they require large numbers of crack images to learn the features of cracks in the real world. Besides, except for the crack forms, there always exist all kinds of noise motifs that will disturb the correct detection of crack regions. The lack of crack sample images has been an obstacle for the improvement of deep learning-based crack detection method. This paper proposes a generative adversarial network (GAN) based method to establish a synthesized crack image dataset with pixel-wise annotations, which provides a novel way aside from the traditional data augmentation method. The Deep Convolutional GAN (DCGAN) model was adopted for the generation of synthesized crack annotations while the Pixel2Pixel model was used to generate the corresponding synthesized crack images. The generated crack annotations and crack images during the training epochs were demonstrated to show how the GANs learn to generate the synthesized images. Moreover, comparative study was conducted to validate the performance of the synthesized crack image dataset for training the crack detection DNN. The results showed the DNN trained by the synthesized images can achieve 74.34% of the MeanIoU that was reached by the same DNN model trained with real images. As for the way of using the synthesized and real crack images for training a crack detection DNN, the way that use the synthesized crack images for pre-training and then use the real images for fine-tuning is better than the way that directly mix the synthesized and real crack images for training. This work provides reference for the GAN-based establishment of crack image dataset and the evaluation of the image quality for training crack detection DNNs. [ABSTRACT FROM AUTHOR]

• A numerical database of 450 CFST columns is developed using ABAQUS. • The results of 531 CFST columns are used to train and test the optimised ANN model. • Predictive equations and Graphical User Interface are developed for CFST columns. • Strength reduction factors are proposed using Monte Carlo simulation. • The proposed ANN model and equations predict well the strengths of CFST columns. This paper presents an optimised Artificial Neural Network (ANN) model for predicting the ultimate axial strengths of concentrically loaded Concrete-Filled Steel Tubular (CFST) short and slender columns strengthened with Carbon Fibre-Reinforced Polymer (CFRP). Since experimental data on CFRP strengthened CFST columns is limited, an accurate Finite Element (FE) model is developed and used to provide additional numerical data. A multi-layered feed-forward back-propagation network is proposed, optimised, and trained using the results of 76 experimental tests and 450 generated FE models. The accuracy of the ANN model is assessed through comparing its computed results with experimental data. A reliability analysis is performed using Monte Carlo Simulation (MCS) to evaluate the safety of the solutions computed by the ANN model. ANN-based equations and Graphical User Interface (GUI) are developed based on the trained ANN model for the determination of the ultimate axial strengths of CFST columns. The results show that the developed ANN model is capable of accurately predicting the ultimate axial strengths of CFRP strengthened CFST columns with a high degree of accuracy. [ABSTRACT FROM AUTHOR]

The estimation of the maximum displacements of RC beams subjected to impact loads is of pivotal importance to define demand models to be employed in structural design. The dynamic behavior is quite complex and the maximum displacements can be obtained through experimental tests or high-fidelity FEM models, while simplified analytical models accounting for few degrees of freedom have generally low accuracy. In this context, this paper presents a study on the training and interpretation of a machine-learning model that strategically combines several algorithms for the said purpose. To train the model, a comprehensive database of 342 samples of rectangular simply supported beams under rigid impact loads is adopted. Seven ML regression algorithms are adopted in this study to develop a predictive model: (1) Support Vector Regression (SV R), (2) Gaussian Process Regression (GPR), (3) Random Forest Regressor (RFR), (4) XGBoost , (5) CatBoost , (6) LightGBM , and (7) Artificial Neural Networks (ANN). Finally, a stacking model combining SV R , GPR , RFR , XGBoost , CatBoost and ANN is built to improve accuracy, wherein 75% and 25% of the data are used for training and validation, respectively, in combination with the 5 cross-validation technique. This effort resulted in approximately 90% validation accuracy exceeding current mechanics-based/semi-empirical models. Finally, the SHapley Additive exPlanations (SHAP) algorithm is used to estimate the relative importance of the input features; the most influential factors are impact velocity and impactor mass which are representing the initial kinetic energy. Moreover, a full parametric analysis of the proposed model is provided and the influence of each input is discussed. The model is finally applied to perform a structural check for three damage levels. The proposed model can be used in practical applications where a reliable estimation of the maximum displacements of RC beams under impact loading is required. • Dataset of 342 samples of rectangular RC beams subjected to impact loads. • Seven ML regression algorithms are compared to predict maximum displacements. • Stacking model combining SV R , GPR , RFR , XGBoost , CatBoost and ANN. • SHapley Additive exPlanations (SHAP) algorithm to estimate importance of the inputs. • A full parametric analysis of the proposed model is provided. [ABSTRACT FROM AUTHOR]

• Deep neural network is employed for wind pressure prediction for the first time. • Errors of all taps on roof are exhaustively analysed by multiple error metrics. • The accuracy in roof ridge and corner bay is evidently higher than the literature. • A nested DNNs model is proposed to improve the accuracy in the corner zone. Although the problems of wind pressure prediction on roofs have been studied extensively, the prediction accuracy is still unsatisfactory owing to the limited capacity of shallow learning of artificial neural networks (ANNs), especially in the areas characterized by flow separation. Moreover, the literatures have only conducted error analysis on part of the predicted taps instead of all taps, and have not made precision comparison with related studies. In this paper, the model of deep neural networks (DNNs) with the ability of deep learning are built to predict the mean and root-mean-square (RMS) wind pressure coefficients on low-rise buildings. For quantitatively comparing the presented results with those from the literature, the same buildings from the NIST-UWO database as studied by the literature are also predicted in this work. In order to improve the DNNs model for predicting the corner zone with higher pressure gradients, a nested DNNs model is further proposed by returning the mean coefficients predicted by one DNNs model as the input of another DNNs model for the RMS coefficients. The prediction results of the DNNs model are also compared with those from the methods of random forest (RF) and general regression neural network (GRNN). The study shows that the DNNs model obviously enhances the prediction accuracy around the roof ridge and the corner bay in comparison with the ANNs model. For the mean or RMS coefficients of all the taps on the whole roof, the correlation coefficient between predicted and experimental results exceeds 0.997, the mean-square-error (MSE) is less than 5 %, and the relative errors of>95 % predicted samples are lower than 10 %. The prediction accuracy in the corner zone is further improved by the nested network where the samples within errors less than 5 % are evidently increased and the errors > 20 % disappear. Also, the prediction results from the DNNs model are apparently better than those from RF and GRNN. This study could provide a benchmark for future studies on the prediction of mean or RMS wind pressure coefficients on the roofs. [ABSTRACT FROM AUTHOR]

This paper presents a novel data-driven approach for active structural control through the use of deep reinforcement learning, wherein, the control system learns to react in an optimal manner through a training process that utilizes deep neural networks within a reinforcement learning framework. The key advantage of this paradigm is the data-driven approach to active control which helps circumvent the need for high-fidelity modeling that typically requires extensive prior knowledge about the structure of interest. Furthermore, the proposed framework is applicable for designing a variety of active controllers, and different external load types, for example, wind and seismic loads for any desired building. The efficacy of the proposed framework is demonstrated in the context of seismic response control through three numerical case studies. The results confirm that the proposed approach yields significant structural response reductions in the linear and nonlinear regimes. Furthermore, implementation issues such as sensitivity to structural property variations, and time delay are thoroughly investigated. • Data-driven active control framework circumventing detailed modeling using deep RL • Customizable environment for various building geometries and control systems • Active controller trained for linear/nonlinear buildings and tested by unseen motions • Trainable using synthetic ground motion data and generalizable to real earthquakes • A robust data-driven control model to system property variations and time delay [ABSTRACT FROM AUTHOR]

In this paper, we present a comprehensive study of mechanical characteristics of a reconfigurable origami-inspired structure using Finite Element and Artificial Neural Network (ANN) approaches. We introduce a design of crease section in the proposed reconfigurable structure, which undergoes enormous and complicated deformation in the folding and unfolding process. Although this crease design can make the deploying process more straightforward, it can jeopardize the stability and life cycle of the structure. We explore how the geometric parameters of the design affect the stability and fatigue failure, including length ratio, total height, story's height, thickness, crease indexes, and circumscribed circle's radius. In order to reduce the computational time, we develop an ANN employing the obtained Finite Element method (FEM) results. The ANN results demonstrate that decreasing the circumscribed circle's radius, the length ratio, and the total height enhance the stability of the origami-inspired structure. It is found that crease indexes affect stability based on the radius of the circumscribed circle. In addition, we investigate how these parameters simultaneously contribute to this design's buckling load and life cycle. The results provide a detailed design parameter characteristic map that can be used to optimize origami structure performance. • Study the stability and fatigue failure of a reconfigurable origami-inspired structure using Kresling pattern. • Develop an Artificial Neural Network and use the Finite Element Method available in ANSYS software. • The stability is increased with the decrease of the length ratio (b/a) and the radius of the circumscribed circle. • Increasing the thickness of the structure leads to an increase in the buckling load. • Decreasing the length ratio and total height can improve the life cycle and stability. [ABSTRACT FROM AUTHOR]

A critical challenge in structural control is the design of controllers capable of rapid adaptation to uncertainties, including changes in local dynamics and unknown parameters. A solution is the use of data-based controllers, but these typically require pre-training and may be computationally expensive to utilize. In this paper, the authors proposed a fast neuro-controller capable of pure online learning. The controller is built using an ensemble of recurrent neural networks (ERNNs) based on long short-term memory (LSTM) cells running in parallel. A multi-rate sampler is used to sequentially construct the individual RNN input vectors using local measurements at sampling rates pre-determined from the structure's natural frequencies. Each RNN performs an individual estimation with the associated sampling rate, and these estimations are assembled into an attention and a dense layer to produce an estimated control force. Critical advantages of the proposed architecture are (1) rapid online adaptation and processing; (2) capability to adapt to local dynamics and non-stationarities; and (3) pure online learning architecture minimizing reliance on pre-training. First, the performance of the ERNNs control algorithm is validated on an actively controlled single-degree-of-freedom system, and stability of the controller is demonstrated. Second, the ERNNs is simulated on a full-scale 20-story building equipped with a semi-active tuned liquid wall damper (TLWD) system that consisted of distributed semi-active tuned liquid multiple column dampers. The ERNNs controller effectively improved the passive performance of the control strategy and performed similarly to that of a full state-feedback sliding mode controller under multi-hazard scenario. The evolution of attention weights under both the wind and seismic hazards demonstrated that the importance of each RNN follows the importance of the structural frequency responses, showing that the controller was capable of capturing and adapting fast-changing dynamics. • A novel pure data-driven neural controller for multi-hazard mitigation. • Fast online adapting capability to system dynamic uncertainties. • The proposed controller is robust to uncertainty in system dynamic properties. • The proposed controller provides great control with limited measurements. • Fast computational speed for preventing control latency. [ABSTRACT FROM AUTHOR]

• This work studied web crippling behaviour of aluminium alloy lipped channels. • Experimental tests, finite element models, and deep neural network were built. • The effects of related factors on web crippling strength were studied. • The prediction accuracy of current design guidelines was assessed. • New formulas for web crippling strength were proposed. This work presents a deep-learning model referred to as a deep belief network (DBN) to investigate the end-two-flange (ETF) web crippling behaviour of roll-formed aluminium alloy lipped channels (both unfastened and fastened to the supports) with centred and offset web circular holes. A total of 1,080 data points was generated for training the DBN, using an elasto-plastic finite element model that was validated using 30 new experimental results presented in this paper. When the DBN predictions were compared to the experimental results, they were found to be 5% conservative on an average. A parametric analysis was then carried out using the DBN predictions, to study the effects of hole size, hole position, section thickness, and the bearing plate on web crippling strength of perforated roll-formed aluminium alloy channels. The DBN predictions were utilised to evaluate the performance of the current design rules of American Iron and Steel Institute (AISI 2016), Australian and New Zealand Standards (i.e., AS/NZS 1664, AS/NZS 4600:2018), and Eurocode (CEN 2007). The current design standards were demonstrated to be over conservative in predicting the web crippling strength of roll-formed aluminium alloy perforated channels. From the parametric analysis, new web crippling strength and web crippling strength reduction factor formulae for roll-formed aluminium alloy perforated channels were proposed. Additionally, a reliability analysis revealed that the proposed equations accurately predicted the ETF web crippling strength of roll-formed aluminium alloy perforated lipped channels. [ABSTRACT FROM AUTHOR]

• A practical vibration-based CNN combined FE model and field-testing data is proposed for in-service steel railway bridges. • A data augmentation strategy is proposed to classify various acceleration responses related to damage scenarios. • t-Distributed Stochastic Neighbour Embedding (t-SNE) and Gradient weighted Class Activation Mapping (Grad-CAM) are used for visualization of feature extraction and feature mapping. • The proposed method can be used to assess the health state of structure to fulfil the regulatory, such as Australian Standard, and safety aspects of steel railway bridges. • The proposed techniques can be utilized for rail infrastructure management without the need of involving specialized expertise and complex signal processing. Railway bridges exposed to extreme environmental conditions can gradually lose their effective cross-section at critical locations and cause catastrophic failure. This paper has proposed a practical vibration-based deep learning approach for damage classification of various extents and degrees of cross section losses due to damages like corrosion in operational railway bridges using vibration-based Convolutional Neural Networks (CNN)s. Firstly, field testing of an in-service railway bridge is conducted and the modal parameters of the bridge are obtained to validate the developed Finite Element (FE) model of the bridge. In the next phase, corrosion scenarios of the main bridge members are generated as various quantities of cross section losses of these members by the validated FE following the Australian Standard AS7636. In the deep learning part, a 1D CNN aligned with novel specific data augmentation strategies is developed to classify various acceleration responses related to each damage scenario simulated by the validated FE. Furthermore, a visualization of feature extraction and feature mapping using t-Distributed Stochastic Neighbour Embedding (t-SNE) and Gradient-weighted Class Activation Mapping (Grad-CAM) is illustrated. The case studies on the simulated and field data-validated FE model results applying background noises and variations, and the real field testing data suggest that the proposed method can reach a perfect damage classification close to an accuracy of 100%. [ABSTRACT FROM AUTHOR]

Bond strength estimation plays an important role in structure engineering. This paper proposes to adopt machine learning approaches to conduct a data-driven analysis of bond strength between ultra-high performance concrete (UHPC) and reinforcing bars. To make up for the lack of experimental data, a new database is established by integrating 557 instances from several published works. A total of nine machine learning models which can be divided into three types are implemented to train the bond strength estimators based on the database, including linear models, tree models, and artificial neural networks. Four strong metrics, i.e. Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R 2), and Ratio of Accurate Estimation (RACC), are used to evaluate the performance of models. Among them, Artificial Neural Network and Random Forest achieve great estimation performances in the top two, which far exceed the empirical formulas. They have 74% and 73% of estimated data to keep the relative error within 10%, respectively. The statistical relative importance of different factors from tree models consistently shows that the ratio of embedded depth to the diameter of reinforcing bars has a significant impact on the bond strength of UHPC, which is conformable with the observations in experiments. • The largest database of bond strength between UHPC and reinforcing bars was built. • Nine machine learning models were used to train the bond strength estimator. • The ratio of the embedded depth had a significant effect on the bond strength. [ABSTRACT FROM AUTHOR]

• By conducting an extensive parametric study including 475 finite element model (FEM) of steel–concrete composite beams (SCCB), an attempt was made to provide a reliable database. • The Lateral distortional buckling (LDB) strength of SCCB as one of the failure modes has been investigated. • The MR, and ANN have been used to predict the formulae for better prediction of the LDB of SCCB by practical engineers. • The accuracy of the developed formula is verified versus the actual data using several suitable criteria. • The existing predictions given by the researchers were compared with the formula obtained from the MR, and ANN. Lateral distortional buckling (LDB) mode has been investigated as one of the failure modes in steel–concrete composite beams (SCCB). The LDB mode in such beams is known by lateral displacement accompanied by rotation of the compressed bottom flange that occurs due to the web distortion. However, many studies have been conducted to investigate the behavior of this failure mode, no equation has yet been found that can accurately calculate the ultimate LDB resistance of SCCB. Consequently, in the current paper, by conducting an extensive parametric study including 475 finite element models (FEM), an attempt was made to provide a reliable database. Then, based on the provided database, two methods of artificial neural network (ANN) and multiple regression (MR) were employed, and based on them, high-precision equations were proposed to predict the ultimate LDB strength of SCCB. Finally, the proposed formulas were compared with the existing formulas for estimating the LDB strength of SCCB. The results showed that the proposed formulas not only present a reasonable accuracy compared to the existing formulations but also can be used by engineers as practical equations in the design process of SCCB. [ABSTRACT FROM AUTHOR]

• Machine learning (ML)-based approach is developed for seismic failure mode and maximum bearing capacity prediction for corroded RC columns. • The best models for seismic failure model prediction is the RF and CatBoost model which has an accuracy of 89%. • The best model for bearing capacity prediction is the CatBoost model which has a R2 of 0.92. • The ML-based models can conveniently predict the life-cycle seismic performance of RC columns. Corrosion of steel reinforcements is a major factor that will adversely affect the seismic performance of the reinforced concrete (RC) columns. This paper investigates the application of machine learning (ML)-based approach for seismic failure mode and maximum bearing capacity prediction for corroded RC columns. A comprehensive database consisting of 180 cyclic tests of corroded RC columns are collected. Six ML algorithms including three single learning methods (k -Nearest neighbors, Decision tree, Artificial neural network) and three ensemble learning methods (Random forest, AdaBoost, CatBoost) are selected to develop the predictive model. The performance of the six models are evaluated and the application of ML-based approaches for life-cycle seismic performance assessment of RC column is demonstrated with a case-study column. The results show that the Random forest and CatBoost models have the best performance for seismic failure mode prediction with an accuracy of 89%. The best model for bearing capacity prediction is the CatBoost model which has a R2 of 0.92, and the CatBoost model is superior to the traditional mechanism-based code models for bearing capacity prediction. The ML-based models can conveniently predict the seismic failure mode and bearing capacity of RC columns in its life-cycle context without complicated numerical simulations or theoretical calculations. [ABSTRACT FROM AUTHOR]

• A fiber model is described for circular DCFST slender columns with ultra-high-strength concrete. • Constitutive laws for ultra-high-strength concrete with confinement effects are included. • New numerical solution algorithms based on King-Werner's approach are developed. • The behavior and design of slender DCFST columns loaded eccentrically are presented. • The fiber model and design equations developed by the ANN yield accurate solutions. The use of ultra-high-strength concrete in double-skin concrete-filled steel tubular (DCFST) composite columns has received considerable attention in recent years. However, no numerical study on the behavior of eccentrically loaded circular DCFST slender beam-columns made of ultra-high-strength concrete has been reported. This paper proposes computational modeling procedures for accurately predicting the axial load–deflection responses and axial load-moment interaction diagrams of circular DCFST slender beam-columns filled with ultra-high-strength concrete when subjected to the combined action of bending and axial compression. The computational procedures based on the fiber element methodology incorporate the effects of concrete confinement, second-order and initial geometric imperfection on the behavior of DCFST slender beam-columns. The enhanced King-Werner's algorithm is implemented in the fiber-based procedures to solve the nonlinear moment equilibrium equation. The accuracy of the present model is verified with experimental results published in the literature and finite element analysis results. The numerical results reveal that the confined concrete model can capture the response of ultra-high-strength concrete with a sudden drop after reaching the peak load and the ductility in the softening branch. The significance of several influencing factors on the structural behavior is investigated in the comprehensive parametric study including 154 specimens with a wide range of column geometric and material properties. The design of circular DCFST slender beam-columns with ultra-high-strength concrete in accordance with Eurocode 4 is discussed. The artificial neural network (ANN) model is developed to propose new empirical formulas for evaluating the member resistance of DCFST slender beam-columns. The proposed equations give accurate and reliable predictions of the column's resistance and can be employed in practical design. [ABSTRACT FROM AUTHOR]

• A recursive long short-term memory network approach for predicting nonlinear structural seismic response is proposed. • A recursive mechanism is adopted for predictions with arbitrary lengths and sampling rates. • The LSTM network is applicable to ground motions with different spectral characteristics and amplitudes. • The LSTM network achieves extremely low computational cost compared with the traditional methods. Artificial neural networks have been used to predict nonlinear structural time histories under seismic excitation because they have a significantly lower computational cost than the traditional time-step integration method. However, most existing techniques require simplification procedures such as downsampling to maintain identical length and sampling rates, and they lack sufficient accuracy, generality, or interpretability. In this paper, a recursive long short-term memory (LSTM) network was proposed for predicting nonlinear structural seismic responses for arbitrary lengths and sampling rates. Referring to the traditional integral solution method, the proposed LSTM model uses the recursive prediction principle and is therefore applicable to structures and earthquakes with different spectral characteristics and amplitudes. The measured ground motions and multilayer frame structures were used for model training and validation. The rules of hyperparameter selection for practical applications are herein discussed. The results showed that the proposed recursive LSTM model can adequately reproduce the global and local characteristics of the time history responses on four different structural response datasets, exhibiting good accuracy and generalization capability. [ABSTRACT FROM AUTHOR]

• Two soft computing models are developed for the estimation of masonry compressive strength. • The unit and mortar strengths and the height to thickness ratio have been found as main influencing parameters. • Proposed ANN model proves quite more accurate compared to existing codes and literature models. • A database of 410 individual specimens has been compiled for the development and evaluation of the models. Masonry is a building material that has been used in the last 10.000 years and remains competitive today for the building industry. The compressive strength of masonry is used in modern design not only for gravitational and lateral loading, but also for quality control of materials and execution. Given the large variations of geometry of units and joint thickness, materials and building practices, it is not feasible to test all possible combinations. Many researchers tried to provide relations to estimate the compressive strength of masonry from the constituents, which remains a challenge. Similarly, modern design codes provide lower bound solutions, which have been demonstrated to be weakly correlated to observed test results in many cases. The present paper adopts soft-computing techniques to address this problem and a dataset with 401 specimens is considered. The obtained results allow to identify the most relevant parameters affecting masonry compressive strength, areas in which more experimental research is needed and expressions providing better estimates when compared to formulas existing in codes or literature. [ABSTRACT FROM AUTHOR]

• A benchmark cable net roof with square and rectangular plans, hyperbolic paraboloid shape was designed. • Parametrization of pressure coefficients was obtained as a function of the geometry. • Wind-induced displacements were estimated by FEM analyses. • Vertical displacements of the cable net were predicted using artificial neural networks. This paper discusses the implementation of an Artificial Neural Network (ANN) model for predicting the wind-induced, vertical displacements of cable net roofs with a hyperbolic paraboloid shape. The ANN model was calibrated by training and comparing three preliminary ANN topologies. The first one was trained using both experimental pressure coefficients on similar building geometries and wind-induced vertical displacements estimated by Finite Element Method (FEM). This first set was subsequently used to calibrate a second ANN model, which was based on surrogate modeling of pressure coefficients, calculated by polynomial fitting of the experimental data. Further generalization of the pressure coefficients was obtained through parametrization of the polynomial fitting as a function of geometrical properties. Pressure loads were used to calibrate the third ANN, which was employed to synthetically generate vertical displacements of a varied geometrical sample. The ANN-based models exhibit remarkable results; the coefficient of determination between physically-estimated flexible roof displacements and approximated values was usually above 0.90 for all the tested geometries, suggesting that this type of surrogate model is capable of replicating complex, geometrically nonlinear structural behavior. [ABSTRACT FROM AUTHOR]

• Finite difference model is proposed for temperature prediction in composite columns. • A database containing 15,200 composite column specimens is established. • Equivalent temperature method is proposed for temperature calculations. • Artificial neural networks method for fire resistance design of composite columns. • Closed-form equations are derived for predicting fire resistance of composite columns. Concrete encased steel (CES) columns, also known as steel reinforced concrete (SRC) composite columns, exhibit superior fire resistance due to concrete acting as a protection layer for the embedded steel section. While modern design codes have provided design guides for the fire resistance of CES columns, they are only applicable to those made of normal strength concrete. For high strength CES columns, advanced analysis is needed to capture the brittleness of high strength concrete at elevated temperature. In this paper, two methods, namely the artificial neural network (ANN) and the analytical equations, are proposed to predict the fire resistance of axially-loaded CES columns made of high strength concrete. To train the ANN, a finite difference model is developed to compute the temperature field in CES columns and it is used to establish a database containing 15,200 specimens. The cross-sectional dimensions and materials grades of the specimens are carefully selected to cover a wide range of values including those commonly adopted in real-life applications. The inputs of the ANN are identified through an extensive parametric analysis. The selected ANN consists of 7 inputs, 3 outputs and 2 hidden layers and achieves a high determination coefficient R2 value of 0.999. For practical implementation, analytical equations are also derived and achieve high R2 values above 0.953. The predictive power of the ANN and the analytical equations are examined against the observations obtained from actual fire tests, showing reasonable accuracy of prediction. Both methods are simple, of high accuracy and have implicitly accounted for temperature-dependent material degradation, and hence do not require input of temperature-dependent material properties and advanced analysis software. [ABSTRACT FROM AUTHOR]