Predicting costs for bridge replacement projects
Analytics
679 views ◎648 downloads ⇓
Abstract
The North Carolina Department of Transportation (NCDOT) uses historical highway bridge records to make cost-effective decisions on which maintenance, repair, or replacement action is appropriate for a deficient bridge. The current method for estimating total bridge replacement cost does not provide reliable and consistent estimates, which impairs forecasting efforts. Updating the current prediction models to include additional factors that may influence cost would theoretically improve the fidelity of the models. Prior studies on bridge cost estimation models for NCDOT and INDOT (Indiana) served as a starting point for the modeling effort detailed in this study. A dataset of recent NCDOT bridge replacement projects was compiled to serve as a foundation for the updated models. Statistical software was used to perform multivariate regression analysis to identify statistically significant predictors and to build models to predict the geometric characteristics of new, replacement bridges (such as new bridge length, width, and span length), as well as right of way costs, engineering costs, construction costs, and total replacement costs. Two approaches were explored in order to predict cost: 1) predicting new bridge characteristics from old bridge characteristics, then predicting bridge replacement costs from predicted new bridge characteristics, and 2) predicting bridge replacement costs directly from old bridge characteristics. New models developed as part of this work were compared to the previously utilized models based on how well the model fit the data (R2) and the confidence interval of the prediction. When both sets of models were used with current bridge replacement data, the new models achieved better fits and yielded narrower confidence intervals than the previously utilized models. Comparing the residual error distributions for the different modeling approaches, the models developed to predict costs directly from the bridge characteristics of the structure being replaced were found to out-perform the models developed to predict cost using forecasted characteristics of the replacement structure. Predicting replacement project costs as a total cost (instead of summing the predicted right of way, engineering cost, and construction cost amounts) avoided introducing compounded error from aggregated component cost predictions. For future work, it is recommended that changes with respect to how bridge information is logged into databases would streamline the data conditioning process and increase the usable number of entries for creation of the models.