An Investigation of Learning Curves and Their Use in Simulation

In 1995, the C-17 Factory Simulation Model (FSM) was developed to enable analysts to address \u27what-if\u27 questions about the resources required to build future aircraft, and is based on learning curve models that are used to both portray and simulate future aircraft production. In this thesis, we examine and develop alternate learning curve models that also utilize a small amount of initial production data to portray the relationship between the number of aircraft built and the resources required to build them. The goal is to identify a model which not only provides a good fit and forecast based on a small amount of data but is also intuitive and reasonably simple to apply. We also propose and evaluate the use of Autoregressive Moving Average (ARMA) models for modeling the effects of learning. These models are exercised in fitting simulated log-linear data, as well as in fitting and forecasting historical F-102 manufacturing data and notional C-17 manufacturing data. The results are somewhat inconclusive since they do not identify any one model as the best. They do, however, suggest that ARMA models are a promising alternative to the standard log-linear learning curve. The thesis concludes with an examination of the effects of explicitly accounting for uncertainty in parameter estimation when simulating future performance based on the traditional log-linear learning curve model. The results show that the approach employed in the FSM is viable even though it does not directly account for this uncertainty