Physics-guided residual regression and empirical-error-constrained design exploration for frp-reinforced concrete slab–column connections

Abstract Punching shear is a brittle limit state in concrete flat slabs, and its assessment is still uncertain when fiber-reinforced polymer (FRP) bars replace steel flexural reinforcement. This study develops a physics-guided residual-ridge model for interior, concentrically loaded FRP-reinforced slab–column connections without shear reinforcement. The development database includes 143 tests from 28 experimental programs: 93 GFRP, 39 CFRP, and 11 BFRP specimens. A modified KDS 14 formulation provides the mechanical baseline, and ridge regression estimates the logarithmic test-to-model residual. Candidate models were selected by nested five-fold cross-validation grouped by experimental program. The selected model achieved a pooled grouped out-of-fold $$\:{R}^{2}$$ of 0.911, RMSE of 91.0 kN, MAE of 61.3 kN, and MAPE of 15.5%; the corresponding mechanics-model values were 0.668, 175.7 kN, 125.4 kN, and 27.7%. The model was also evaluated using three full-scale GFRP slabs from one independent program. Their measured capacities were 124–135 kN, while the predicted capacities were 131.7–139.0 kN, and all observations were within the pre-specified empirical 90% multiplicative error bands. For the 93-specimen GFRP subset, ACI 440.11-22 produced a mean test-to-prediction ratio of 2.09, while the proposed model gave 1.04. A bounded two-objective enumeration examined 2,310 alternatives; 1,908 met the empirical lower-band resistance and applicability requirements, and 16 were nondominated for effective depth and FRP area per unit width. A browser-based graphical user interface applies the final equations, error band, and design enumeration, and it exports the nondominated solutions. The model is intended for comparative assessment and preliminary design exploration within the reported domain; the empirical bounds are not code resistance factors.