Adaptive Compaction Construction Simulation Based on Bayesian Field Theory

Abstract

The compaction construction process is a critical operation in civil engineering projects. By establishing a construction simulation model, the compaction duration can be predicted to assist construction management. Existing studies have achieved adaptive modelling of input parameters from a Bayesian inference perspective, but usually assume the model as parametric distribution. Few studies adopt the nonparametric distribution to achieve robust inference, but still need to manually set hyper-parameters. In addition, the condition of when the roller stops moving ignores the impact of randomness of roller movement. In this paper, a new adaptive compaction construction simulation method is presented. The Bayesian field theory is innovatively adopted for input parameter adaptive modelling. Next, whether rollers have offset enough distance is used to determine the moment of stopping. Simulation experiments of the compaction process of a high earth dam project are demonstrated. The results indicate that the Bayesian field theory performs well in terms of accuracy and efficiency. When the size of roller speed dataset is 787,490, the Bayesian field theory costs only 1.54 s. The mean absolute error of predicted compaction duration reduces significantly with improved judgment condition. The proposed method can contribute to project resource planning, particularly in a high-frequency construction monitoring environment.