عليرضا فرزين پور

Occurrence of Northridge earthquake in 1994 and its effects on structures and infrastructures was the beginning for two important subjects in structural and earthquake engineering. The first has been development and explanation of performance-based design (PBDS) frameworks as a new approach that enables structural engineers to target specific stakeholder-driven building performance objectives, and the second was about innovative structural systems, such as self-centering mechanism. Self-center moment resisting frames (SC-MRFs) which allows isolation of damages in certain components of the connection are an example of this type of structures that can be easily repaired and remove damages from basic components of steel frames. PBDS begins with defining a set of performance objectives using some metrics, followed by preliminary design. Ideally, the building performance should then be assessed by using generated engineering demand parameters (EDPs) from nonlinear time history analysis, and earthquake-induced impacts should be eva‎luated from the analysis results. Conducting nonlinear time history analysis iteratively for SC-MRFs and revisiting the design would be computationally expensive and labor intensive. To overcome this issue, boosting-based ensemble learning models are used to estimate the sesismic drift demand of SC-MRFs based on a generated dataset of important features. Also, interpretable machine learning (ML) models are used to interpret important parameters on resulted estimations. Based on the five different metrics, Catboost model had the best prediction result amog three ML models. SHAP Values are also used to make interpretation of estimations results. Finally, compression between seismic performance of SC-MRFs with welded steel moment frames shows great reduction in amount of resudail drifts in different stories of the frame.

طراحي بر مبناي عملكرد , پارامترهاي تقاضاي مهندسي , قاب‌ هاي خمشي فولادي مركزگرا , يادگيري ماشين , مدل هاي تفسيري

Performance-based design , Engineering demand parameters , Self-Centering moment resisting frames , Machine learning , Interpretable models

alireza farzinpour

vahid broujerdian