Julian Carrillo

Accurate prediction of residual flexural strengths is essential for the structural design and performance assessment of steel fibre reinforced concrete (SFRC). While machine learning models have recently demonstrated strong predictive capability, most of the existing approaches rely on heuristic feature selection, unconstrained architectures, and random validation splits that may overestimate generalisation capacity. This study proposes a Hard-Constrained Physics-Informed Neural Network (PCNN) for predicting the residual flexural strengths f_R1–f_R4 of steel fibre reinforced concrete (SFRC). Non-negativity, monotonic dependence on the reinforcement index, and bounded sequential evolution across crack mouth opening displacement (CMOD) stages are enforced architecturally through custom layers, guaranteeing mechanical admissibility by construction. The model was trained and validated on 235 original beam specimens from 18 experimental groups (expanded to 860 via controlled integer replication) using GroupKFold by Study cross-validation to eliminate data leakage. PCNN achieves f_R1 MAE = 1.256 MPa and R² = 0.606 across held-out experimental campaigns, competitive with Ridge Regression (MAE = 1.144, R² = 0.688). Monotonicity audits show a 0% violation rate for PCNN versus 7.3% and 10.2% for Random Forest and GBDT, respectively. PCNN is the only model in the comparison that guarantees physical admissibility for all possible input configurations. The results demonstrate that embedding mechanical principles directly into neural network architecture enhances robustness and interpretability without compromising accuracy. The proposed Physics-Informed Neural Networks framework provides a transferable methodology for physics-guided machine learning in structural materials modelling.

Accurate prediction of residual flexural strengths is essential for the structural design and performance assessment of steel fibre reinforced concrete (SFRC). While machine learning models have recently demonstrated strong predictive capability, most of the existing approaches rely on heuristic feature selection, unconstrained architectures, and random validation splits that may overestimate generalization capacity. This study proposes a Hard-constrained Mechanics-Informed Neural Network framework for predicting SFRC residual strengths across independent experimental studies. A multi-study database comprising 860 observations from 18 experimental assemblies was analysed to identify statistically dominant and physically interpretable predictors. Based on exploratory statistical analysis and micromechanical assessment, a hierarchical shared-backbone neural network architecture was developed. Physical admissibility was enforced structurally through non-negativity constraints, monotonic dependence on effective fibre bridging capacity, and bounded signed transitions between CMOD stages. Model performance was evaluated using five-fold group-based cross-validation, where entire studies were excluded from training to rigorously assess generalization. The proposed architecture achieved predictive performance comparable to or exceeding conventional neural networks and ensemble models, while eliminating physically inadmissible predictions and reducing fold-to-fold variability. The results demonstrate that embedding mechanical principles directly into neural network architecture enhances robustness and interpretability without compromising accuracy. The proposed hard-constrained Physics-Informed Neural Networks (PINN) framework provides a transferable methodology for physics-guided machine learning in structural materials modelling.

Confined masonry (CM) houses are widely used in low- and mid-rise houses worldwide, especially in countries with emerging economies. CM emerged in Latin America as a response to the severe damage observed in unreinforced masonry buildings following significant earthquakes in the early twentieth century. In Colombia, CM construction dates back to the 1930s, with houses up to five stories tall. A key challenge in seismic risk estimation is predicting the behavior of prevalent construction typologies in a region to inform risk mitigation and management strategies. As a part of a larger project called the National Seismic Risk Model for Colombia, this study presents a seismic fragility assessment of CM houses, evaluating the effects of masonry strength, masonry unit type, wall density, construction quality, and number of stories. A total of 72 archetype houses, representative of Colombian CM structures, were analyzed across low, intermediate, and high seismicity zones. Nonlinear models were proposed and calibrated using quasi-static cyclic tests on confined masonry walls and shaking table tests on scaled buildings. For each archetype, nonlinear time history analyses were conducted in OpenSees, using approximately 370 hazard-consistent ground-motion records to represent the hazard at each site of interest. The results from these structural analyses are then combined with damage information to obtain fragility curves for this building typology. The fragility results of this study, combined with vulnerability functions, will serve as inputs to calculate risk metrics later on as part of the National Seismic Risk Model for Colombia. The present work provides a region-specific fragility assessment, calibrated with local experimental data, which enhances its applicability within Colombia and potentially to other Latin American countries with similar construction practices and standards. These findings can contribute directly to national seismic risk mitigation initiatives.

Plastic straws (PS) are often discarded after a brief period of use, contributing significantly to global pollution. This article explores the mechanical properties of fiber-reinforced concrete (FRC) incorporating PS as an innovative reinforcing material. Our study evaluated the tensile strength and overall mechanical performance of concrete in both its fresh and hardened states after the inclusion of PS. To this effect, three tensile tests were performed, as well as 18 workability tests according to ASTM C-143 and 18 flexural tests following ASTM C78. Statistical methods were employed to analyze how the quantity and size of PS impact the mechanical properties of concrete. Although PS exhibited low stiffness and tensile strength, they demonstrated a high deformation capacity compared to other polymers. Additionally, the statistical analysis indicated that the dosage of PS added significantly influences the performance of FRC under flexural loading. Overall, the mechanical performance observed was comparable to that of other FRCs utilizing industrial polymers. This suggests that PS can effectively serve as a reinforcing element in concrete, given their good residual strength, which is a crucial characteristic in applications involving FRC.

Correlations between fiber characteristics and the compressive, flexural-tensile, and shear strengths of steel fiber reinforced concrete (SFRC) are proposed in the literature to facilitate the code-compliant design. In this study, the correlation between flexural-tensile and direct shear stresses on plain concrete (PC) and SFRC beams and columns is evaluated. The database included 1767 data from direct shear tests of beam and column, and from four-point bending tests (4PBT) of PC and SFRC beams. Data comprised measured results from this study and from the results of other research programs published in the literature. The main characteristics of the database were fiber aspect ratios (l _f /d _f between 65 and 80), fiber volume fraction (V _f between 0.25% and 2%), and compressive strength of concrete (f′ _c between 28 and 78 MPa). From the trends observed in the experimental results, an empirical equation is proposed to estimate the direct shear and flexural-tensile performance of SFRC. The results of the statistical analysis showed that the equation is a suitable tool to estimate the maximum direct shear and flexural-tensile strengths of SFRC. The correlation between shear and flexural-tensile strengths would facilitate SFRC code-compliant design.

Non‐engineered construction constitutes a significant portion of the existing housing stock in Colombia and other South American countries. These masonry constructions frequently lack technical and engineering characteristics, affecting their performance under seismic and gravitational loads. The assessment of their mechanical properties, such as compressive strength, modulus of elasticity, shear strength, and shear modulus triggers challenges when using existing models, as experimental results of unreinforced masonry houses often deviate from these theoretical models. To address these challenges, machine learning models using neural networks were proposed to predict masonry mechanical properties. To develop the neural networks, a comprehensive database of experimental results from various tests conducted in Colombia was complied, categorizing data into four main masonry types based on material and masonry units: (1) horizontally hollow clay bricks, (2) vertically hollow clay bricks, (3) solid clay bricks, and (4) vertically hollow concrete bricks. These categories encompass the prevalent types of masonry units used in unreinforced masonry constructions. The neural networks developed herein demonstrate superior performance when compared to existing predictive models in the literature, facilitating their application in structural analyses of masonry. In addition, recalibrated equations were proposed to better reflect the properties of masonry of unreinforced constructions. A user‐friendly application was developed to enable practical use of both the database and the predictive models, supporting researches and engineers in characterizing masonry for seismic resilience and retrofitting.