This paper investigates the role of learning curve models in estimating construction productivity. Learning curve theory is actively implemented for both the scheduling and cost estimation of complex construction projects. The purpose of the research is to assess the suitability of published learning curve models in effectively analyzing the learning phenomenon for substantially complex construction operations. The research investigates five (5) learning curve models, namely the (a) Straight-line or Wright, (b) Stanford "B", (c) Cubic, (d) Piecewise or Stepwise and (e) Exponential models. The methodology includes the comparative implementation of each one of the aforementioned models for the analysis of a large infrastructure project with the use of unit and cumulative productivity data. A two-stage investigative process for the five models was applied in order to define (a) the best-fit model for historical productivity data of completed construction activities and (b) the best predictor model of future performance. The assessment criterion for the suitability is the deviation of the real construction data from the predictions generated by each model. The research results indicate that the Cubic model dominates in terms of its predictive capability on historical data, while the Stanford "B" model is a better future performance predictor. Future research directions include the extension of the research scope with the inclusion of more learning curve models in conjunction with a populated database of historical field data.
展开▼
