Sensitivity analysis on a construction operations simulation model using neural networks

This paper addresses how to perform sensitivity analysis on simulation models for large, complex, resource-constrained, and technology-driven construction operations, with particular focus on how to quantify the effects of each input factor upon the output measures of performance on a precast viaduct construction operations simulation model. We first briefly reviewed existing techniques for sensitivity analysis on simulation models and identified their respective limitations. Then we introduced and applied a neural network (NN)-based technique to facilitate sensitivity analysis on construction operations simulation models. The technique defined input sensitivity in undistorted, practically accurate terms and permitted relating a set of input factors to multiple outputs. In the case study on precast viaduct construction operations, we investigated the effects of four relevant factors - related to tractor resource provision, precast segment delivery logistics, and site layout - upon the average cycle time as required for erecting one span of the viaduct. It is concluded that a valid simulation complemented with the NN-based sensitivity analysis contributes to gaining insights and deriving new knowledge on the real system, which ultimately leads to improved cost-effectiveness and enhanced efficiency on the real system.