Machine learning for structural engineering: A state-of-the-art review
Introduction
Machine learning (ML) is a class of artificial intelligence (AI) that focuses on teaching computers how to make predictions from available datasets and algorithms. Most importantly, it provides computer systems the ability to learn and improve themselves rather than being explicitly programmed. Although ML was born in 1943 and first coined in 1959, it actually started to flourish in the 1990 s, and has become the most successful subfield of AI. ML has also become one of the technology buzzwords of our age since it plays a pivotal role in many real-world applications such as image and speech recognition, traffic alerts, self-driving cars, medical diagnosis, etc.
In general, ML can be classified into three main categories based on the learning process: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning is the most basic type of ML whose algorithm is trained from a labelled dataset. This method is suitable for regression and classification problems, and it has been widely used in structural engineering for damage detection (classification problems) and strength predictions (regression problems). On the contrary, the algorithm used in unsupervised learning is trained from an unlabelled dataset. Meanwhile, in the reinforcement learning method, the algorithm is trained through a trial-and-error process. A significant number of ML algorithms have been adopted in structural engineering applications, e.g., neural networks (NN), decision tree (DT), regression analysis (RA), support vector machine (SVM), random forest (RF), boosting algorithm (BA), etc. Surrogate model, also known as metamodel, is a special case of supervised ML which has been widely used in the field of engineering design to reduce computational time of complex black-box ML models with relaxed accuracy. It is an interpretable model which is trained to approximate the predictions of a black-box ML model. In other words, surrogate models are simple analytical models which mimic the behaviour of complex ML models.
Structural engineering involves the structural analysis and design of load-bearing structures. For complex structural systems under extreme actions that exhibit highly nonlinear behaviour, the use of structural analysis and design methods requires a time-consuming calibration process, and they are somehow too complicated for practical implementation. In this case, ML can provide a promising alternative to save time and effort. One of the first structural engineering applications of ML was carried out by Adeli and Yeh [1] in 1989 using artificial neural network (ANN) to design steel beams [2]. Since then, the ANN algorithm has been successfully used in many pioneering works in structural engineering including structural analysis and design [3], structural damage detection [4], structural health monitoring (SHM) [5], structural optimisation [6], [7], [8], strength and resistance predictions [9], [10], and structural reliability [11]. However, the use of ML in structural engineering is still in its infancy at that time [12] due to the limitations of ML algorithms and computing power. This is evidenced by the fact that only a few relevant articles were published each year in the early stage of structural engineering applications (see Fig. 12a in Section 3).
Another reason that hinders the application of ML in structural engineering at its early stage is a lack of experimental databases to ensure the validation of ML models. However, in recent years, the research community has taken the necessary steps towards overcoming this barrier by establishing database platforms (e.g., DataCenterHub, DesignSafe, Mendeley Data, etc.) to collect data from structural engineering tests. DataCenterHub is a massive repository platform with over 250 datasets from nearly 50,000 experiments [13]. DesignSafe is extended from the network for earthquake engineering simulation (NEEShub), a cyberinfrastructure platform to share data and tools for earthquake engineering [14], [15] and disaster risk management [16], [17]. Some notable databases for structural engineering include NEEShub datasets for earthquake engineering that can be accessed from DataCenterHub [18] and image databases for crack damage detection (e.g., Structural ImageNet [19] with over 10,000 images, PEER Hub ImageNet [20] developed by the Pacific earthquake engineering research (PEER) centre with over 36,000 images, bridge crack library [21] with over 11,000 images, etc.). Detailed databases used in structural engineering are given in Section 2.5.
In addition to the establishment of the database platforms for structural engineering, there have also been recent advances in ML techniques. BA methods (see Section 2.2.6), especially extreme gradient boosting (XGBoost) [22] and categorical gradient boosting (CatBoost) [23], offer extremely powerful tools to solve the problems with large datasets in a fast and accurate manner. Convolutional neural network (CNN) [24] is considered as one of the state-of-the-art ML algorithms for image-based crack detection [20] due to its ability in rapidly detecting crack damage in structures. Recently, Google team has made a new breakthrough when creating a new ML method called AutoML-Zero [25] that can evolve itself without human intervention. In addition, the availability of open-source ML libraries with hands-on ML algorithms and ready-to-run packages (e.g., TensorFlow and Keras developed by Google and PyTorch by developed by Facebook) has facilitated the development of ML-based models for structural engineering applications.
With the rapid development of ML algorithms and computational power combined with the availability of databases collected recently, the research community has witnessed a boom in the use of ML in the structural engineering domain, especially over the last five years with a clear exponential growth in the number of publications each year (see Fig. 12a in Section 3). However, practical applications of ML in structural engineering are still very limited. One of the real-world applications of ML is to improve design of buildings through generative design, where the industry (e.g., Arup) has developed ML-powered tools to generate design alternatives that meets requirements of the end-users. Although a number of review articles published recently have touched on this topic, they just focused on a certain area of structural engineering (e.g., structural design and performance assessment [26], reliability and safety [27], [28], earthquake engineering [29], [30], structural design for fire [31], SHM and crack detection [32], [33], [34], [35], [36], [37], [38], [39], concrete property [40], [41], concrete mix design [42], capacity prediction of concrete structures [43], and design and inspection of bridges [44]). A comprehensive review on all areas of structural engineering is lacking.
This paper is therefore aims to present a comprehensive review on all applications of ML techniques to structural engineering. The present review is considered as the most ambitious and comprehensive work when covering a wide range of structural engineering applications (i.e., structural analysis and design, SHM and damage detection, behaviour and capacity of structural members and systems, fire resistance of structures, and property and mix design of concrete) and ML algorithms (i.e., NN, SVM, DT, RF, BA, and RA). The review looks at both isolated structural members (e.g., beam, column, slab/panel, wall, and joint) and whole structural systems (e.g., truss, frame, building, and bridge) made from different materials (e.g., concrete, steel, cold-form steel (CFS), fibre reinforced polymer (FRP) composite, and steel–concrete composite). The review also considers different behaviours of structures under shear, flexural, torsional, axial, bond, and buckling actions. It should be noted that evolutional algorithms (EA) including genetic algorithm and gene expression programming are types of ML, and thus the author initially intended to cover them in this paper. However, there is a very large number of publications (about 125 references) on the application of EA to structural engineering found in the literature, and thus the inclusion of this topic will make the manuscript too lengthy. In addition, there is already one review article published recently on this topic [45]. For this reason, this topic is not covered in the manuscript.
Section snippets
ML in a nutshell
This section provides the concepts and hands-on tools to implement ML methods. It covers a wide range of ML algorithms which are widely used in the structural engineering domain. In addition, available Python libraries, open-source codes and datasets for ML are also provided for the readers to practise and execute their ML models.
Bibliometric survey
In this section, a bibliometric study of the current literature on the use of ML methods for structural engineering applications is presented. The literature search is limited to Scopus indexed papers collected from well recognised academic databases including Web of Science, Scopus, Science Direct, Wiley Online Library, Taylor & Francis Online, Springer Link, ASCE Library, and SAGE. The keywords used in this search include ML and soft computing related terms (e.g., artificial neural networks,
ML applications
Based on the result of the bibliometric survey (Section 3), seven classes of ML methods and five different structural engineering topics have been identified as shown in Table 5. The seven groups of ML methods are (1) NN methods, (2) SVM methods, (3) BA methods, (4) RA methods (i.e., linear regression – RA1, multivariate regression – RA2, polynomial regression – RA3, LASSO regression – RA4, Ridge regression – RA5, and logistic regression – RA6), (5) RF method, (6) DT method, and (7) others. The
Summary of the findings from the reviewed literature
A comprehensive review on the application of various ML algorithms for different areas of structural engineering presented in 3 Bibliometric survey, 4 ML applications indicates the potential of ML in this field. Based on the results of this review, the following findings of ML algorithms and structural engineering applications are summarised as below:
Conclusions
ML has emerged as a promising predictive tool for a broad range of structural engineering applications, and thus it can be potential replacements for commonly used empirical models. The application of ML in structural engineering is booming evidenced by an exponential growth of the number of relevant publications in the literature in recent years. In this paper, an ambitious and comprehensive review on the applications of ML for structural engineering has been presented. The review covers a
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This is research was supported by the Australian Research Council (ARC) under its Future Fellowship Scheme (FT200100024). The financial support is gratefully acknowledged.
References (578)
- et al.
Neurobiological computational models in structural analysis and design
Comput Struct
(1991) - et al.
Use of neural networks in detection of structural damage
Comput Struct
(1992) - et al.
Optimum design of aerospace structural components using neural networks
Comput Struct
(1993) - et al.
Structural reliability analyis of elastic-plastic structures using neural networks and Monte Carlo simulation
Comput Methods Appl Mech Eng
(1996) - et al.
Machine learning applications for building structural design and performance assessment: state-of-the-art review
J Build Eng
(2021) - et al.
Review and application of artificial neural networks models in reliability analysis of steel structures
Struct Saf
(2015) - et al.
Machine learning for reliability engineering and safety applications: Review of current status and future opportunities
Reliab Eng Syst Saf
(2021) - et al.
Soft computing techniques in structural and earthquake engineering: a literature review
Eng Struct
(2020) - et al.
Advances in computer vision-based civil infrastructure inspection and monitoring
Eng
(2019) - et al.
A systematic review of convolutional neural network-based structural condition assessment techniques
Eng Struct
(2021)
A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications
Mech Syst Signal Process
Machine learning techniques for structural health monitoring of heritage buildings: a state-of-the-art review and case studies
J Cult Heritage
Structural crack detection using deep convolutional neural networks
Autom Constr
Emerging artificial intelligence methods in structural engineering
Eng Struct
Computational design optimization of concrete mixtures: a review
Cem Concr Res
Machine learning applied to the design and inspection of reinforced concrete bridges: Resilient methods and emerging applications
Struct
Finding structure in time
Cogn Sci
Deep learning in neural networks: an overview
Neural Netw
A decision-theoretic generalization of on-line learning and an application to boosting
J Comput Syst Sci
Boosting machines for predicting shear strength of CFS channels with staggered web perforations
Struct
Concrete-to-concrete interface shear strength prediction based on explainable extreme gradient boosting approach
Constr Build Mater
Machine-learning interpretability techniques for seismic performance assessment of infrastructure systems
Eng Struct
Predicting bearing capacity of double shear bolted connections using machine learning
Eng Struct
Explainable extreme gradient boosting tree-based prediction of load-carrying capacity of FRP-RC columns
Eng Struct
Efficient machine learning models for prediction of concrete strengths
Constr Build Mater
Data-driven machine-learning-based seismic failure mode identification of reinforced concrete shear walls
Eng Struct
Failure mode and effects analysis of RC members based on machine-learning-based SHapley Additive exPlanations (SHAP) approach
Eng Struct
Computer vision-based concrete crack detection using U-net fully convolutional networks
Autom Constr
Image-based concrete crack detection in tunnels using deep fully convolutional networks
Constr Build Mater
Automatic crack classification and segmentation on masonry surfaces using convolutional neural networks and transfer learning
Autom Constr
Predicting concrete compressive strength using hybrid ensembling of surrogate machine learning models
Cem Concr Res
A machine learning-based formulation for predicting shear capacity of squat flanged RC walls
Struct
General solution for shear strength estimate of RC elements based on panel response
Eng Struct
Classification of failure mode and prediction of shear strength for reinforced concrete beam-column joints using machine learning techniques
Eng Struct
Perceptron learning in engineering design
Comput-Aided Civ Infrastruct Eng
Neural networks in civil engineering: 1989–2000
Comput Aided Civ Infrastruct Eng
Integrated assessment of seismic damage in structures
Comput Aided Civ Infrastruct Eng
Structural optimization by gradient-based neural networks
Int J Numer Meth Eng
Optimal design of transmission towers using genetic algorithm and neural networks
Int J Space Struct
Prediction of ultimate shear strength of deep beams using neural networks
ACI Struct J
HPC strength prediction using artificial neural network
J Comput Civ Eng
Machine learning techniques for civil engineering problems
Comput-Aided Civ Infrastruct Eng
A cyberplatform for sharing scientific research data at DataCenterHub
Comput Sci Eng
Advancing earthquake engineering research through Cyberinfrastructure
J Struct Eng
The NEEShub cyberinfrastructure for earthquake engineering
Comput Sci Eng
DesignSafe: new cyberinfrastructure for natural hazards engineering
Nat Hazard Rev
Disaster risk management through the DesignSafe cyberinfrastructure
Int J Disaster Risk Sci
Deep transfer learning for image-based structural damage recognition
Comput-Aided Civ Infrastruct Eng
PEER Hub ImageNet: a large-scale multiattribute benchmark data set of structural images
J Struct Eng
Cited by (202)
Deep reinforcement learning for intelligent risk optimization of buildings under hazard
2024, Reliability Engineering and System SafetyCausal diagrams for civil and structural engineers
2024, StructuresMachine learning tools to improve nonlinear modeling parameters of RC columns
2024, Journal of Building EngineeringInvestigation of optimized machine learning models with PSO for forecasting the shear capacity of steel fiber-reinforced SCC beams with/out stirrups
2024, Journal of Building EngineeringPrediction of concrete and FRC properties at high temperature using machine and deep learning: A review of recent advances and future perspectives
2024, Journal of Building Engineering