Computational study demonstrates high accuracy for a hybrid AI model detecting seismic irregularities in architectural plans, highlighting improved early-stage earthquake-resistant design.
Key Points
Develop and evaluate a hybrid AI framework that integrates convolutional neural networks with structural mechanics calculations to detect seismic irregularities in early architectural floor plans.
Integrated a convolutional neural network (CNN) with analytical formulas for center of rigidity and center of mass to identify torsional irregularity (A1) and slab discontinuity (A2) per the Turkish Earthquake Code (TBDY 2018).
Trained and evaluated the system on 1,576 architectural plan samples derived from eight statically verified base configurations using an 80/20 train-validation split.
Achieved a validation accuracy of 99.2% and an F1-score of 0.98 for classifying seismic structural irregularities.
Outperformed purely image-based visual models, which demonstrated poor detection near boundary conditions where structural eccentricity closely approached code thresholds.