صبا دعائي

Nowadays, with growing use of surrogate modeling in various fields and the ability to simulate complex objective functions with high speed and precision, the application of these models has become essential in structural design processes. Although surrogate models have become crucial tools for the design and analysis of structural systems, optimizing real-world structures based on modern design standards remains a complex and computationally expensive task. This research introduces, for the first time, a surrogate model using a proposed active machine learning algorithm to predict the structural responses of 3D Steel Building Frames (such as story drift, maximum roof displacement, and Demand/Capacity (D/C) ratio), aiming to eliminate the need for time-consuming and complex structural analysis and design processes for engineers. The main objective of this study is to introduce the most effective surrogate model for predicting various structural responses of 3D Steel Building Frames during the design process. A total of 2,000 samples were generated using the discrete Latin Hypercube method for 4-, 8-, and 12- story frames. After sampling, 1,000 samples were used to develop surrogate models with XGBoost, CatBoost, and RF, utilizing both the proposed active machine learning method and the conventional approach. The results indicate that the XGBoost surrogate model, in combination with the proposed active machine learning algorithm, generally outperforms other approaches. The use of the proposed active algorithm not only improves the model’s accuracy but also reduces the number of required samples. This reduction allows for the use of more accurate and computationally intensive sampling methods in future studies. The surrogate models eva‎luation reveals an R² of over 0.85 for the D/C ratio and over 0.75 for the other two responses, indicating a good alignment with the target functions. However, the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) values suggest that further improvements are needed before these models can fully replace traditional structural design processes.

مدل‌سازي جايگزين , پيش‌بيني پاسخ قاب خمشي فولادي , يادگيري ماشين فعال

Surrogate modeling , Prediction of 3D Steel Building Frame responses , Active machine learning

Saba Doaei

Dr. Majid Ilchi Ghazaan- Dr. Asghar Habibnejad korayem