Mix Proportion Optimization of Ultra-High Performance Concrete Based on Machine Learning

Abstract:

In recent years， ultra-high performance concrete （UHPC） has become one of the hot research directions due to its excellent mechanical properties and durability， but its high cost has always limited its application in engineering. In order to reduce the cost of UHPC， this paper proposes a method based on machine learning to optimize the mix proportion of UHPC. In order to achieve this goal， the prediction model of a 28-day compressive strength and expansion of UHPC was first established by using artificial neural network （ANN）， which was taken as the constraint condition， taking into account the constraints of UHPC component content， component proportion and absolute volume， The cost of UHPC was reduced by using genetic algorithm （GA）. The research results show that the error between the prediction results of ANN model and the experimental results is within 10 %， which has good prediction accuracy. The cost of UHPC optimized by GA is reduced to $838.8，which is lower than the cost of $1000 mentioned in the literature.