Machine Learning Prediction of Bleeding of Bored Concrete Piles Based on Centrifuge Tests

Bleeding in fresh concrete refers to the rise of excessive water as a result of sedimentation and consolidation of the aggregates. Various machine learning models are applied in this study to predict the amount of bleeding water based on centrifuge model tests on concrete bored piles. The soil type, its consolidation, concrete age and exposition class, pile geometry and amount of mixing water are the model test variables. Together with derived features at the prototype scale and geometric ratios, the degree of dewatering is predicted with three algorithms, namely the linear and decision tree regressions and artificial neural networks. The predictions match the observed data with a coefficient of determination up to 0.882. Given that the data are retrieved from centrifuge model tests, a reasonable agreement with field measurements is expected.