پیش بینی مولفه های زوال ستون های بتن مسلح با استفاده از روش های یادگیری ماشین

New performance-based design approaches in earthquake engineering aim to accurately and transparently assess the risk of loss of life and structural damage. Advanced analytical models are used to determine the performance of structures, with one of the key components being the deterioration of structural members under seismic loads. Multilinear backbone curves are commonly used in regulations and software to simplify the behavior of members subjected to seismic loads, including the deterioration components. This paper proposes using machine learning models to predict the deterioration components of reinforced concrete (RC) columns. A dataset of ۲۵۵ experimental data from ۱۹۷۳ to ۲۰۰۲ is used to predict the deterioration components using different machine learning methods. The RC columns have three failure modes: bending, shearing, and bending-shearing. The deterioration components predicted by the analytical relationships are compared with the results obtained from machine learning methods. The dataset includes ۱۴ features as model inputs and ۳ features as outputs. The paper examines three algorithms for predictions: AdaBoost, artificial neural network (ANN), and random forest (RF). The analysis is conducted using Python software. The results show that the random forest model has an accuracy rate of ۹۱% for the Plastic chord rotations from yield to cap , ۸۱% for Post-capping plastic-rotation capacity from the cap to point of zero strength , and ۸۸% for normalized energy dissipation capacity compared to other algorithms. Also, the results obtained from the predicting models have considerable accuracy compared to analytical relationships. Compared to analytical models, the random forest model has significantly been improved in terms of root mean square error (RMSE), mean absolute error (MAE) and coefficient determination (R۲). These improvement are ۷۹% , ۷۵.۳% and ۴۶.۵% in (R۲), ۶۳.۷% , ۴۸.۵% and ۸۶.۷ in (RMSE), ۶۴% , ۹۲% and ۸۹.۴% in (MAE).The results showed that the random forest model has been significantly improved the accuracy of determination of deterioration components compared with analytical models.

Keywords:

مولفه های زوال , یادگیری ماشین , هوش مصنوعی , شبکه های عصبی مصنوعی , جنگل تصادفی