Our faculty member, Assoc. Prof. Zeynep Değer, took part in a research collaboration that resulted in a recent publication introducing an integrated modeling framework to improve the seismic performance assessment of reinforced concrete (RC) structural walls. RC walls are essential components in lateral load–resisting systems, yet predicting their cyclic behavior—especially local responses such as strains and rotations—remains a major challenge in finite element (FE) modeling. The study addresses these challenges by combining validated OpenSees material models with machine learning–based parameter estimation.
Fig. 1. Schematic view of the model subjected to (a) axial and lateral load combination; (b) axial, lateral load and moment combination.
The proposed framework uses OpenSees to simulate key behavioral mechanisms of RC walls, including strain penetration, shear deformation, bar buckling and rupture, and low-cycle fatigue, while also addressing numerical issues such as strain localization in walls exhibiting softening behavior. To enhance model calibration, machine learning algorithms were trained to predict essential parameters—particularly the peak strain of confined concrete—using geometric and material properties from a database of 140 tested wall specimens.
Fig. 2. Machine learning model development and verification layout.
Findings highlight the strong potential of integrating machine learning into FE simulation workflows to improve both accuracy and physical transparency. Although further work is needed to expand the approach to more complex wall geometries and behaviors, the study provides an important step toward more reliable, efficient, and generalizable modeling strategies for RC structural walls in seismic design and assessment.
To read the full article: www.sciencedirect.com/S2352710225030578