RESEARCH

Structural behaviour and performance assessment of proposed concrete-filled quadruple steel tubular (CFQST) cross-sections under compressive loading: Experimental and numerical study (Journal of Building Engineering, 2025). DOI: https://doi.org/10.1016/j.jobe.2025.113828

We introduce CFQST columns (four inner steel tubes + outer tube) and show large gains over RC/CFST/CFDST: up to +43% axial capacity, +137% lateral capacity, +103% axial ductility, +81% lateral ductility. Optimal performance occurs near χ ≈ 0.33–0.38 and RDF = 0.50.

HIGHLIGHTS

Program: 20 columns (1 RC, 3 CFST, 1 CFDST, 15 CFQST) fabricated and tested; FE models developed/validated in Abaqus (C3D8R, CDP).
Phase results: Outer tube thickness ↑ (2→3→4 mm) raises capacity/ductility with diminishing returns beyond 3 mm.
Hollow ratio χ: Nonlinear trend; optimal near 0.33 (square) / 0.38 (circular); higher χ reduces capacity but increases ductility.
Tube shape: Square inner tubes outperform circular (≈1-2% axial capacity, up to 28% axial ductility, ≈5-8% lateral capacity gain).
RDF: Best at 0.50; RDF=0.33 or 0.67 lowers capacity/ductility relative to 0.50.
Lateral performance: CFQST vs RC: +137% lateral capacity; vs CFST: up to +54% capacity and +63% ductility.
Analytical model: Introduces ζ (RDF) to amplify confinement; predicts axial capacity with max error 6.36%, R² = 0.84.

KEYWORDS: CFQST, CFST, CFDST, composite columns, axial capacity, lateral capacity, ductility, hollow ratio (χ), radial distance factor (RDF), Abaqus, CDP, analytical model

Advanced predictive machine and deep learning models for round-ended CFST column (Scientific Reports, 2025). DOI: https://doi.org/10.1038/s41598-025-90648-2

Using a curated 200-specimen dataset of round-ended CFST stub columns, we benchmark LightGBM, XGBoost, CatBoost and DNN/CNN/LSTM to predict axial capacity (Pcc). CatBoost attains the lowest error (testing RMSE ≈ 396.5 kN, R² ≈ 0.932); DNN/CNN show higher R² but larger errors. Width (b) is the strongest positive driver; length (h) reduces capacity.

HIGHLIGHTS

Dataset & target: 200 CFST round-ended stub columns; inputs = {f′c, h, b, d, ts, fys}; output = Pcc.
Models: XGBoost, LightGBM, CatBoost, plus DNN, CNN, LSTM; tuned via Bayesian optimization + 5-fold CV.
Test metrics (representative): CatBoost RMSE ≈ 396.5 kN, R² ≈ 0.932; XGBoost R² ≈ 0.906; LightGBM R² ≈ 0.916; DNN R² ≈ 0.958/RMSE ≈ 496 kN; CNN R² ≈ 0.951/RMSE ≈ 536 kN; LSTM weakest (R² ≈ 0.891/RMSE ≈ 801 kN).
Uncertainty: U₉₅ analysis shows CatBoost lowest uncertainty; LSTM highest.
Explainability: SHAP → b most positive; h most negative; d, ts, fys positive at higher values. PDPs confirm monotonic trends.
Baseline comparison: Against 10 analytical models, ML/DL demonstrates superior accuracy and consistency across metrics.
Tooling: A lightweight GUI (Tkinter) enables practitioners to input six features and instantly get Pcc predictions.

KEYWORDS: CFST, round-ended columns, axial capacity (Pcc), machine learning, deep learning, SHAP, PDP, Bayesian optimization, GUI

Integrated behavioural analysis of FRP-confined circular columns using FEM and machine learning (Composites Part C: Open Access, 2024). DOI: https://doi.org/10.1016/j.jcomc.2024.100444

We introduce and analyse double-skin double-filled tubular (DSDFT) columns vs DSHT using validated Abaqus FE models and ML. DSDFT boosts ultimate axial load by ~20–101%, with trends mapped for FRP type, FRP layers, and steel-tube diameter; BI-LSTM/LSTM give the best predictions.

HIGHLIGHTS

Scope & configs: Circular DSHT (hollow) vs DSDFT (double steel tubes + inner concrete); 48 FEM columns, parameters: steel-tube diameters, FRP type (GFRP/AFRP), FRP layers, concrete strength.
FE modelling: Abaqus; CDP concrete; S4R shells for FRP/steel, C3D8R for concrete; tie + contact (μ≈0.6); axial displacement loading; BCs shown in Fig. 5.
Validation: FEM vs tests shows small errors—ultimate load max dev ≈12.14%, min ≈0.07%; strain dev up to ≈24.02% (min ≈0.52%).
• Key gains (DSDFT vs DSHT): Ultimate axial load ↑ by ~19.5–101.2%; ultimate strain generally ↑ (e.g., +3.18% at 60.3 mm outer tube).
FRP thickness: More layers → +10–15% load (GFRP) and up to +40% (AFRP) in DSDFT, with reduced ultimate strain (≈4–5%↓).
• FRP type: GFRP outperforms AFRP +5-13% load and ~+25% ultimate strain vs AFRP cases.
Steel-tube diameter effect: DSHT: larger outer diameter → load ↓ (≈37% from 60.3→101.6 mm); DSDFT: larger diameter → load ↑ (~40%) and strain ↑ (~9%) (60.3→114.6 mm).
Dataset & ML: 82 entries; inputs = {As·fy, effective concrete area (excluding hollows/steel), FRP plies, f′c}; output = ultimate axial load. BI-LSTM/LSTM best (testing R² ≈ 0.955/0.945); MARS and LS-SVM adequate (R²_test ≈ 0.90).
Explainability: SHAP shows Input 2 (effective concrete area) and Input 1 (As·fy) dominate predictions; higher inputs → higher capacity.

KEYWORDS: FRP confinement, DSHT, DSDFT, GFRP, AFRP, Abaqus, CDP, axial capacity, machine learning, LSTM, BI-LSTM, LS-SVM, MARS.

Proposed numerical and machine learning models for fiber-reinforced polymer concrete-steel hollow and solid elliptical columns (Frontiers of Structural and Civil Engineering, 2024). DOI: https://doi.org/10.1007/s11709-024-1083-1

Validated Abaqus FE models + 7 ML algorithms predict ultimate axial load and confined strain for GFRP-confined elliptical DSTC/DSFTC columns. Extra Trees tops performance (testing R² ≈ 0.9956, RMSE ≈ 0.033); Random Forest is second-best.

HIGHLIGHTS

Scope: Elliptical DSTC (hollow) and DSFTC (filled) columns with external GFRP confinement; parametric sweeps on aspect ratio, concrete grade, FRP plies, steel tube sizes.
FE modelling: Abaqus; interaction and boundary conditions verified; responses validated against independent compression tests.
Targets & models: Predict ultimate axial load and confined ultimate strain using AdaBoost, LightGBM, CatBoost, RF, ETR, XGB, DNN with 5-fold CV.
Performance: ETR best, RF second; testing up to R² ≈ 0.9956 with RMSE ≈ 0.033 for load; strong generalization shown in external checks.
Insights: Capacity governed mainly by steel-area×fy and FRP confinement stiffness; DSFTC gains more in strain capacity than DSTC; AR increase reduces load..

KEYWORDS: Elliptical columns, DSTC, DSFTC, GFRP confinement, Abaqus, FEM, Machine learning, Extra Trees, Random Forest, Capacity prediction

Synergetic concrete shape and cable layout optimization of pre-stressed concrete beams (Structural and Multidisciplinary Optimization, 2023). DOI: https://doi.org/10.1007/s00158-023-03545-5

We present SSCO, a FORTRAN tool that simultaneously optimizes concrete shape (via fuzzy, zero-order boundary morphing) and cable layout (B-spline) to eliminate tensile stress while cutting weight. Across 1–5 span PSC beams, SSCO achieves ~12–24% weight reduction in 3–6 iterations with <1 s runtime, and explicitly handles friction losses.

HIGHLIGHTS

Method: Concrete as 9-noded Lagrangian FE; cable as curvilinear 3-noded bar on a 3rd-order B-spline; fuzzy zero-order shape updates; AMCR remeshing each iteration.
Objective & constraints: Minimize tensile stress and concrete weight with limits on σ, MCC (≥40 mm), deflection (≤ min(Span/250, 20 mm)).
Performance: 3-6 iterations, total runtime <1 s; ~12–24% weight saving depending on spans and loads (PB1–PB3).
Validation & trends: PB1 mid-span deflection nearly halves after optimization; PB2/PB3 show alternating span-wise deflection changes; final shapes remain fabricable (no new holes).
Friction sensitivity: SSCO alters both shape and cable when friction losses are considered, yielding asymmetric profiles and different boundary offsets (see PB3 comparisons).

KEYWORDS: PSC beams, simultaneous shape cable optimization, B-spline tendon, fuzzy zero-order, AMCR, friction loss, FEM, pre-stress, weight reduction, SSCO

A Comparative Study of Gradient Descent Method and a Novel Non-Gradient Method for Structural Shape Optimization (IJMEMS, 2022). DOI: https://doi.org/10.33889/IJMEMS.2022.7.2.017

We compare a fuzzy-controlled, zero-order GSO method against OptiStruct’s gradient descent for structural shape optimization. GSO converges on a coarser mesh (avg. element size 50 vs 10) with ~0.23 s/iteration (vs ~3.64 s), gives smoother, fabricable shapes, and shows stronger material-sensitivity in composite beams.

HIGHLIGHTS

Method (GSO): Fuzzy triangular membership targets a prescribed σᵗ; design-node moves use move factor and direction node logic; design elements + isoparametric mapping rebuild a clean mesh every iteration (9-noded Lagrangian).
Convergence & mesh: Validation shows GSO converges at avg. element size 50 vs OptiStruct at 10 (fixed beam test).
Speed: Mean time/iteration ≈ 0.23 s (GSO) vs ≈ 3.64 s (OptiStruct) across cases.
Performance (weight): Comparable reductions overall e.g., B1: 25.58% (GSO) vs 15.74% (OptiStruct); B2: 20.18% vs 41.27%; CC: 15.04% vs 18.69%.
Shape quality: GSO yields smoother, more fabricable shapes with fewer sharp corners; OptiStruct results show more stress-concentration corners.
Material sensitivity: In composite beam B3a–B3e, as E₂ rises (0.2 → 2.0×10⁵ N/mm²), GSO weight saving increases 75.32% → 79.32%; OptiStruct shows no clear trend.

KEYWORDS: Shape optimization, Non-gradient, GSO, Fuzzy membership, Design elements, Isoparametric mapping, OptiStruct, Mesh convergence, Composite beams, Weight reduction

Fuzzy-based integrated zero-order shape optimization of steel-concrete-steel sandwich beams (Current Science, 2021). DOI: https://doi.org/10.18520/cs/v121/i7/941-949

We introduce a gradient-free (zero-order) GSO method that reshapes only the faceplate-core interface of steel-concrete-steel (SCS) beams to meet a target shear stress without changing the overall section. Implemented in FORTRAN with auto-remeshing, it cuts steel where stress is low, keeps serviceability (deflection change ~10⁻⁴), agrees with experiments (e.g., B5: 5.57 vs 5.49 mm deflection), and runs ~0.14 s/iteration.

HIGHLIGHTS

Interface-only morphing: perpendicular growth/shrinkage at the faceplate–core boundary to hit a prescribed σt, without altering the overall section.
Fuzzy control law: triangular membership function and move factor drive node updates toward σ ≈ σt with automatic remeshing.
Mesh/solver: 48 nine-noded Lagrangian elements; design-element mapping maintains mesh quality during shape changes.
Validation vs experiments: interfacial shear and deflection closely match Solomon et al.; e.g., B5 deflection 5.57 mm (present) vs 5.49 mm (exp).
Notable result: curved beams optimize asymmetrically (see Fig. 7h) due to nonlinear stress distribution; average iteration runtime ≈ 0.14 s.

KEYWORDS: Sandwich beams, Gradient-free optimization, Zero-order, Fuzzy membership, Target shear stress, Finite element, GSO

Fuzzy Based Bridge Rating System (FBRS) for Condition Assessment of Existing Railway Bridges (International Journal of Steel Structures, 2025). DOI: https://doi.org/10.1007/s13296-025-00989-x

We propose a fuzzy logic bridge rating framework with a 10-point scale and component-importance weighting (triangular MFs + Fuzzy Weighted Geometric Mean) implemented in Python as FBRS. Applied to 3 railway bridges, FBRS produces nuanced, less-conservative ratings than the IRICEN 5-point “worst-component” method.

HIGHLIGHTS

Problem with IRICEN 5-point: overall rating equals the worst component (ORN) → overly conservative, coarse resolution.
FBRS method: 10-point triangular MFs for condition; triangular MFs for importance; aggregation via FWGM; centroid defuzzification to CRI.
Case 1 (ROB, Moradabad): IRICEN ORN=2 vs CRI=3.43 (borderline Moderate–Severe; retrofit vs replacement).
Case 2 (Plate girder, 3×18.3 m): ORN=3 (major repair) vs CRI=8.32 (Good–Very Good → routine/minor maintenance).
Case 3 (Riveted steel truss, 6×30.48 m): ORN=3 vs CRI=7.76 (Good).
Outcome: FBRS delivers balanced, actionable ratings, prioritising maintenance more efficiently; suited for integration into Bridge Management Systems.

KEYWORDS: Railway bridges, visual inspection, fuzzy logic, FBRS, 10-point rating, FWGM, centroid defuzzification, IRICEN, CRN/ORN/CRI, bridge asset management.

Dual-axis buckling of laminated composite skew hyperbolic paraboloids with openings (Journal of the Brazilian Society of Mechanical Sciences and Engineering, 2018). DOI: https://doi.org/10.1007/s40430-018-1406-z

We develop a refined C⁰ nine-node shell FE model (second-order transverse-shear, cross-curvature included) to study biaxial (Nx, Ny) and shear (Nxy) buckling of laminated skew hypar shells with square cutouts. Buckling resistance falls with larger cutouts & thickness, rises with skew angle and clamping, and is highest for symmetric angle-ply laminates.

HIGHLIGHTS

Model & elements: Refined displacement field with second-order transverse-shear (zero shear at faces) + cross-curvature; implemented as 9-noded curved isoparametric C⁰ element in FORTRAN.
Dual-axis loading: Buckling under biaxial compression (Nx, Ny) and in-plane shear (Nxy); skew-edge DOFs handled by transformation.
Validation: Converges at 16×16 mesh; results agree with HSDT/3D elasticity references for plates/shells and with ABAQUS for cutouts.
Cutout effect: Larger central cutout → lower non-dimensional buckling parameter λ (both uniaxial and biaxial).
Skew angle & BCs: λ increases with skew angle α; CCCC gives the highest λ, SSSS lower; more clamped edges → higher λ.
Thickness: λ decreases as thickness (h/a) increases (per the study’s non-dimensionalization).
Lamination scheme: Symmetric angle-ply yields the largest λ; antisymmetric cross-ply the smallest among cases studied.
Load-ratio trends: For fixed Nxy/Ny, λ decreases as Nx/Ny increases; for fixed Nx/Ny, λ increases with Nxy/Ny.

KEYWORDS: Buckling, laminated composite, skew hyperbolic paraboloid, hypar shells, cutout, biaxial compression, in-plane shear, higher-order shear theory, finite elements, FORTRAN.

Impact of incidence angle of seismic excitation on vertically irregular structures (Earthquakes and Structures, 2024). DOI: https://doi.org/10.12989/eas.2024.27.3.000

Using nonlinear time-history with a bidirectional hysteresis model on regular and setback frames (R1, R2, IR1, IR2), we rotate 20 near-fault and 20 far-field records from 0°–360°. Result: the incidence angle has marginal effect; corner columns in vertically irregular frames show higher demand independent of angle.

HIGHLIGHTS

Scope & models: Regular R1/R2 and vertically irregular IR1 (one-way setback), IR2 (two-way setback); plan 10×10 m, two bays, storey height 3 m.
Records & rotation: 20 far-field + 20 near-fault ground motions; rotated every 15° from 0°–360° to probe incidence effects.
Analysis: SAP2000 nonlinear time-history; plastic hinges in columns; bidirectional yield criterion (Fx/Fxo)^2+(Fy/Fyo)^2 ≥ 1. Cases at Rμ=1 (first yield) and Rμ=4 (inelastic).
Key finding (regular): Increasing storeys does not introduce sensitivity to angle-responses ≈ angle-invariant.
Key finding (irregular): Corner columns of IR1/IR2 show exaggerated demand; this amplification does not depend on angle and diminishes toward central columns (often reduced base shear centrally).
Design implication: For both far-field and near-fault, the incidence angle needn’t be a primary design consideration for these configurations; vertical irregularity drives the distribution of demand.

KEYWORDS: Incidence angle, bidirectional ground motion, vertical irregularity, setbacks, near-fault, far-field, plastic hinges, nonlinear time history, SAP2000, polar plots.

Seismic acceleration amplification factor for pin-supported moment-resisting RC frame structures for Chi-Chi earthquake (Indian Journal of Engineering & Materials Sciences, 2022).

Five pin-supported RC MRFs (2–10 storeys) were analysed with linear time-history under Chi-Chi records (PGA 0.01–0.32 g) to derive a new acceleration amplification factor (AAF) model. The proposed AAF depends on height (z/h), period (T) and PGA range, and matches mean+sd floor accelerations better than UBC, ASCE, IITK, Akhlaghi, Fathali models.

HIGHLIGHTS

Set-up: Five RC MRFs (2/4/6/8/10 storeys), pin supports, hard rock; ETABS linear time-history; T ≤ 1.5 s; damping 5%.
Records: 81 Chi-Chi ground motions split into three PGA bands: 0.01–0.067 g, 0.067–0.20 g, 0.20–0.32 g.
Proposed AAF model:

Findings vs codes: Existing models often over/under-estimate especially for T < 0.7 s; proposed model stays within ~15–22% of mean+sd across bands and heights.
Trends: Floor spectra decrease as T increases; peak floor acceleration can be ≈4×, 1.5×, 2× PGA for the low/mid/high bands, respectively.
Design note: AAF for NSCs must reflect (z/h, T, PGA) - height only (e.g., IITK) is inadequate for pin-based frames.

KEYWORDS: Acceleration amplification factor, non-structural components, Chi-Chi, pin-supported RC frames, ETABS, PGA banding, floor spectra, UBC, ASCE