Hybrid AI Model for Early Detection of Structural Irregularities in Earthquake-Resistant Architectural Design

Abstract

The integration of artificial intelligence (AI) into early-stage architectural design remains limited, particularly in detecting seismic structural irregularities before structural engineering analysis. This study proposes a hybrid AI-driven framework for the early detection of structural irregularities defined in the Turkish Earthquake Code (TBDY 2018), focusing on torsional irregularity (A1) and slab discontinuity (A2). The developed system, Irregularity Detector (ID.1), employs a Convolutional Neural Network (CNN) architecture to extract spatial and geometric features from architectural plan images and integrates these features with analytical structural parameters, including calculations of the center of rigidity and center of mass. The dataset consists of 1,576 architectural plan samples derived from eight statically verified base configurations, with a validation split of 0.2 used during model training. Results demonstrate that purely visual models exhibit limited performance, particularly in boundary conditions where structural eccentricity approaches code-defined thresholds. In contrast, the proposed hybrid approach significantly improves detection reliability by combining image-based learning with engineering-based mathematical analysis. The model achieved a validation accuracy of 99.2% and an F1-score of 0.98, confirming its effectiveness in classifying structural irregularities. The findings highlight the limitations of conventional visual augmentation methods in structural analysis and emphasize the importance of physics-informed feature integration in deep learning applications for architectural design. The proposed system enables automated feedback during the architectural design phase, reducing reliance on time-consuming manual structural assessments and minimizing costly design revisions. By bridging the gap between architectural design and structural engineering, this research contributes to the digital transformation of seismic-aware design processes. Future work will focus on expanding the model using large-scale real-world datasets, incorporating additional irregularities such as vertical structural irregularities, integrating with Building Information Modeling (BIM) workflows, and adapting the framework to international seismic codes such as Eurocode 8.