Soft-computing models for predicting plastic viscosity and interface yield stress of fresh concrete

Interface yield stress and plastic viscosity of fresh concrete significantly influences its pumping ability. The accurate determination of these properties needs extensive testing on-site which results in time and resource wastage. Thus, to speed up the process of accurately determining these concrete properties, this study tends to use four machine learning (ML) algorithms including Random Forest Regression (RFR), Gene Expression Programming (GEP), K-nearest Neighbor (KNN), Extreme Gradient Boosting (XGB) and a statistical technique Multi Linear Regression (MLR) to develop predictive models for plastic viscosity and interface yield stress of concrete. Out of all employed algorithms, only GEP expressed its output in the form of an empirical equation. The models were developed using data from published literature having six input parameters including cement, water, time after mixing etc. and two output parameters i.e., plastic viscosity and interface yield stress. The performance of the developed algorithms was assessed using several error metrices, k-fold validation, and residual assessment etc. The comparison of results revealed that XGB is the most accurate algorithm to predict plastic viscosity (training, testing) and interface yield stress (training, testing). To get increased insights into the model prediction process, shapely and individual conditional expectation analyses were carried out on the XGB algorithm which highlighted that water, cement, and time after mixing are the most influential parameters to estimate both fresh properties of concrete. In addition, a graphical user interface has been made to efficiently implement the findings of this study in the civil engineering industry.