Neural networks surrogate models for simulating payment risk in pavement construction

Abstract

A common provision in quality control/quality assurance (QC/QA) highway pavement construction contracts is the adjustment of the pay that a contractor receives on the basis of the quality of the construction. It is important to both the contractor and the contracting agency to examine the amount of pay that the contractor can expect to receive for a given level of construction quality. Previous studies have shown that computer simulations can provide a better, more de- tailed examination of the pay schedule than is possible by simply determining the expected pay. In particular, the simula- tion process can provide an indication of the variability of pay at various quality levels and can identify the factors most responsible for pay adjustments. Stochastic simulation models are very useful in estimating and analyzing payment risk in highway pavement construction. However, such models are constrained by their computational requirements, and it is of- ten necessary to couple them with simpler models to speed up the process of decision-making. This paper investigates the use of Neural Networks (NN) to build surrogate models for a pavement construction payment-risk prediction model. The results show that although the average error associated with the NN predictions are acceptable; in some particular cases the errors may be unacceptably high.