Steel concrete composite beams are widely used in modern constructions due to their extraordinary strength, stiffness and structural efficiency. These systems rely on shear connectors to ensure effective force transfer between steel and concrete components, allowing them to operate as a single structural unit. Traditionally, welded head studs have served this purpose. However, their persistent nature complicates decomposition, recycling and reuse, especially at the end of the lifespan of the structure.1. In response, the researchers introduced classifiable shear connectors employing mechanical fixing methods such as bolts, twisted connections, or friction grip systems to facilitate disassembly, material recovery and waste reduction.2,3,4. These connectors are consistent with circular economy principles that emphasize design for disassembly, ease of use of materials, and recyclability. Furthermore, traditional welding systems rely on energy-intensive installations, contributing to the high carbon emissions of the construction sector.5.
Beyond sustainability, removable connectors offer practical advantages, especially in modular constructions where prefabricated parts are quickly assembled and disassembled, reducing labor, construction times and waste2. Research has confirmed that bolted connectors can match or outweigh welded studs in terms of ductility, load-slip performance, and strength, making them suitable for demanding applications such as skyscrapers and bridges.6. The growing number of research supports these findings. Ataei et al.7 We tested the bolted shear connector with a thin-walled beam, showing improved ductility and a clear failure mode. Fahmy et al.8 We evaluated the effects of bolt shape, grout strength, preload, and material properties, and conducted experimental and numerical studies on high strength removable connectors. Their results highlighted the reusability, strength and stiffness of bolted systems and introduced improved predictive equations.
Further research has proposed new design models. ataei9 Introducing design equations that take into account bolt clearance, preload, and material interactions to improve prediction accuracy over existing standards. Similarly, Kwon et al.10 Pavlovićetal.11 Equivalent shear resistance was demonstrated between the bolted connector and the weld connector, highlighting the resolution potential of the bolted fastening system.
Research by Ram and Die12 Lee and Bradford13 It focuses on environmentally friendly materials such as geopolymer concrete combined with bolted systems, achieving high ductility and CO₂ reduction. Rehman et al.14I have confirmed that such a system meets Eurocode 415 Ducility and load capacity requirements. Recent Works by Suwaed and Karavasilis16Ram et al.17and Patel et al.18 We further examined the structural and environmental benefits of blind bolts and removable connectors in complex and precast systems. Furthermore, Kozma et al.19Loqman et al.20and Hosseini et al.twenty one We have confirmed that the removable connector not only allows disassembly and reuse, but also provides higher fatigue resistance and slip capacity than welded studs. suwaedtwenty two We have introduced a friction-based high strength bolted connector that achieves high shear resistance and rapid disassembly.
Recent studies incorporate advanced numerical modeling to validate experimental results. Wang et al.twenty three FE analysis was used to study high strength bolted connectors under reverse push-off loads between Csillag and Pavlovićtwenty four We investigated the blind bolt connectors and SRR connectors on the FRP deck. Hosseinpour et al.twenty five We verified the compatibility of bolted connectors on thin-walled cold-formed beams and met the Eurocode 4 ductility standard. Various innovative connector types have appeared: Lock bolts (LB-DSC) with grout-filled tubes and cone-shaped lugs26Y-shaped bolt connector with tapered nut (Jung etal. 2022) and double-nut friction grip connector27 Song et al.28 Tested Tapered Iron Bolt (TIB) System and proposed new design formulation. Liu et al.29 and Jakovljevićetal.30 The adaptability of single nut embedded connectors and the reliability of removable connections across the precast and casting system have been confirmed. Furthermore, the fangs31 We evaluated high strength bolts and bolted connectors on UHPC and confirmed their excellent operation. Fahmy et al.8 and Ataei et al.4 Using a combination of experimental and FEM data, we proposed a predictive model for the load capacitance and stiffness of cold-formed composite beams. Yu and Kim32 The operational method of stud shear connectors was analyzed using a steel-UHPC composite structure, especially when using short studs. They improved the existing model by considering stud head rotation and plastic deformation, and helped them to better match the experimental results. Their research showed that both UHPC strength and stud size had a strong impact on shear performance, providing clearer guidance for designing more efficient combined systems.
Although machine learning demonstrates strong possibilities in predicting the shear behavior of steel concrete composite connectors, previous studies have identified several limitations. A common problem is the reliance on small or unbalanced data sets, limiting the performance of models across a variety of structural conditions. The practical utility of these models is often reduced by neglecting or simplifying key factors such as periodic loads, complex failure mechanisms, and environmental impacts. Furthermore, many approaches are not transparent, providing predictions without clearly explaining the contribution of input variables to the final outcome, making them less accessible to engineers in real design applications. Furthermore, there is little research that integrates both numerical and experimental data to improve model accuracy.
Several studies have demonstrated the effectiveness of machine learning in predicting shear behavior of stud connectors in steel concrete composite structures. Zhu and Farouk33 Using artificial neural networks enhanced by particle swarm optimization, we predict shear resistance of grouped stud connectors, surpassing traditional design codes based on 232 push-out tests. Similarly, Lee et al.34 We evaluated a variety of machine learning (ML) models and found that artificial neural networks provide the most accurate predictions. The use of global sensitivity analysis identified stud diameter, concrete strength, and stud yield strength as important influencing factors. Zhou et al.35 Using 194 test samples, various machine learning models were further applied, including extreme gradient boosts (xgboost) and random forests, to predict the shear strength of head studs in steel–UHPC systems. User-friendly tools have been developed to interpret the importance of functionality using Shapley Additive Description (SHAP) analysis and to support engineering applications.
In related studies, Li et al.36 Using 245 test results, we proposed an ensemble stacking model that combines Xgboost, LightGBM, and Induratere to estimate the shearing capacity of Perfobond Rib (PBL) connectors. Their model achieved high accuracy (R² = 0.92) and interpretability with concrete strength, number of reinforcement, number of reinforcement, and yield strength of iron identified as the most important features via SHAP analysis. Taffese et al.37 We also developed an explanatory AI model using LightGBM and SHAP to predict slip behavior in steel–UHPC interfaces, highlighting stud diameter and spacing as a dominant factor. These studies33,34,35,36,37 It highlights the expansion of the integration of interpretable machine learning technologies in structural prediction tasks, providing both accurate results and valuable design insights.
Due to fatigue behavior, Kang et al.38 To predict the lifespan of steel concrete bridge stud connectors, we introduced a Gaussian process regression model trained on a large fatigue test data set. Their approach outperformed traditional design codes and provided insight into the most influential fatigue-related variables through explanatory AI technology.
He, etc.39 It researches machine learning applications in the design, optimization and evaluation of steel concrete composite structures, providing a more comprehensive perspective. In addition to highlighting the effectiveness of ML models in the prediction and damage assessment task, this review also identified many important drawbacks, such as lack of data, lack of model transparency, and lack of real-world validation.
Despite these advances, most existing models are empirical and limited in scope. There is a lack of generalized, data-driven frameworks that explain the complex relationships between design parameters, material properties, and performance outcomes specific to removable connectors. At the same time, machine learning (ML) has proven effective in structural engineering applications, such as predicting material strength, corrosion and damage progression. ML models can capture nonlinear and multivariate interactions without simplifying assumptions, providing new opportunities to predict connector behavior. However, applications to removable systems remain known, primarily due to the lack of comprehensive, high quality datasets.
To fill this gap, this study develops and compares multiple ML models trained on a hybrid dataset that integrates experimental and numerical data for removable bolted shear connectors. The aim is to support the design of efficient, sustainable, reusable complexes and create reliable prediction tools to estimate the ultimate shear capacity.
