Machine learning techniques for prediction of failure loads and fracture characteristics of high and ultra?high strength concrete beams

摘要

The construction sector benefits from the development of artificial intelligence (AI) systems because they can more accuratelyforecast the mechanical properties of concrete, which saves time, money, and effort. However, there has not been enoughresearch done using AI on ultra-high strength concrete (UHSC). Therefore, this work examines the use of sophisticated AIapproaches to forecast the failure loads and fracture characteristics of high strength concrete (HSC) and ultra-high strengthconcrete (UHSC) beams. In order to develop machine learning (ML) models, a dataset of 87 experimental tests with seveninput variables was collected from the published literature. Four ML models, namely, the extreme gradient boosting (XGB),the random forest (RF), the convolutional neural network (CNN), and K-nearest neighbors (KNN), were developed. Theprediction effectiveness and generalizability of the created models are validated using a wide range of performance parametersonce a regression model has been adjusted. In both the stages of training and testing, the employed ML models showa significant correlation (correlation coefficient > 0.97) between the experimental and predicted values for four differentoutputs, namely, failure loads (P_(max)), fracture energy (G_F), critical stress intensity factor (K_(IC)), and critical crack tip openingdisplacement (CTOD_C). The proposed XGB model outperformed CNN, KNN, and RF during the testing phase, achievingthe best accurate prediction. Based on the findings of the paper, the developed XGB model can be used as an alternate toolfor estimating the failure loads and fracture characteristics of HSC and UHSC beams in civil engineering projects.