Physics‐informed prediction of shear capacity in RC short columns using a constrained genetic algorithm

Accurate prediction of shear strength in short reinforced concrete (RC) columns is critical for seismic safety assessment, yet existing design provisions often exhibit significant scatter or excessive conservatism due to complex interactions between slenderness, axial compression, and arch action. This study presents a dimensionally homogeneous closed‐form shear strength model for short RC columns, optimized using a genetic algorithm (GA). Trained on a dataset of n = 99 shear‐critical rectangular columns and validated against an independent, out‐of‐sample dataset of n = 24 specimens, the model integrates non‐linear slenderness decay ( a / d ) −0.3912 and axial load enhancement terms. An applicability domain (AD) assessment using the Williams plot ( h * = 0.1515) suggests that the training domain resides safely within statistical leverage bounds. Out‐of‐sample verification demonstrates the proposed model's precision, achieving a coefficient of variation of COV = 16.8%, outperforming standard design codes including ACI 318–19 (COV = 31.0%), ASCE/SEI 41–17 (COV = 30.7%), and Eurocode 2 (COV = 204.3%). Error tracing across the unseen dataset reveals that prediction overestimates occur when elements transition from direct diagonal strut action into intermediate flexure‐shear regimes or under severe axial splitting. To prevent unsafe extrapolation, explicit operational limits are established: 1.0 a / d 2.50, N /( A g f ′ c ) 0.50, f ′ c 70 MPa, and ρ w f wy 8.0 MPa. The resulting explicit closed‐form equation provides a robust, mechanistically consistent tool for structural assessment and code implementation.