همایش بین المللی هوش مصنوعی

صفحه اصلی / اولین همایش بین المللی هوش مصنوعی

Application of machine learning algorithms in the prediction of the reliability of post-tensioned concrete members

نویسندگان :

Pooria Poorahad A.¹ Mahmoud R. Shiravand² Mahtab Ebadati³

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی 3- دانشگاه شهید بهشتی

کلمات کلیدی :

artificial intelligence،machine learning،supervised learning،reliability index،Monte Carlo simulation

چکیده :

The recent progress in artificial intelligence is driven by the advancements in new machine learning (ML) algorithms. The vast application of data-intensive ML methods throughout structural reliability analysis problems has paved the way for researchers to predict structural responses and seismic performances of different structures with minimal human intervention. Supervised ML method is used when labelled data are available. In this study, labelled data are created through finite element method, and reliability indices are calculated based on the Monte Carlo simulation method. The numerical models are built with different aspect and ED bar ratios, and the key parameters are randomly chosen with specified distributions. Hence, supervised learning algorithms are employed to predict the reliability index of self-centering post-tensioned piers. Reliability of post-tensioned concrete members decreases with time due to the prestress loss phenomenon. Therefore, these members cannot fulfill the performance objectives that they were initially designed for. Five ML algorithms are utilized in this paper; (i) linear regression, (ii) random forest, (iii) artificial neural network, (iv) k-nearest neighbors, (v) extreme gradient boosting. The database is divided into testing and training sets R-squared and root mean squared error are considered as the metrics used for the comparison of the ML models. Bayesian search is used for hyperparameter optimization of algorithms. The results indicate that extreme gradient boosting has the finest accuracy. The closeness of performances of testing and training sets indicates that overfitting is avoided.