| Popis: |
Earthquake damage assessment studies conducted throughout the world have already established the importance of considering the contribution of reinforced concrete infilled frames in the response of structures subjected to sudden lateral loads. Still, much clarity needs to be made on the behaviour, and failure mechanisms of RC infilled frames when subjected to such large and sudden earthquake loads. A data-driven machine learning approach to the prediction of failure modes of RC infilled frames is suggested in this paper. An exhaustive database consisting of experimental results done throughout the world was gathered. A failure mode classification system consisting of three predominant failure modes is proposed. Suitable parameters are identified for the purpose of machine learning modelling. Machine learning algorithms like AdaBoost, CatBoost, KNN, Decision Trees were used to predict the failure modes. An open-source dynamic model is created, which could be updated once new data is available from experiments. Google provides a free TensorFlow enabled Jupyter notebook for machine learning (Google Colabs). The same was used in this study as it supports remote access from different locations, and the model would always remain in the cloud, making it instantly accessible. Three performance measures were used in this study to evaluate the performance of the various machine learning models: accuracy, precision, and recall. The results obtained indicate that for complex structural interaction problems having (a number of dependent parameters) machine learning modelling techniques, in which the dataset is allowed to speak for itself, can be successfully employed. |
