Machine-Learning-Based Consumption Estimation of Prestressed Steel for Prestressed Concrete Bridge Construction

Accurate prediction of the prestressed steel amount is essential for a concrete-road bridge's successful design, construction, and long-term performance. Predicting the amount of steel required can help optimize the design and construction process, and also help project managers and engineers estimate the overall cost of the project more accurately. The prediction model was developed using data from 74 constructed bridges along Serbia's Corridor X. The study examined operationally applicable models that do not require indepth modeling expertise to be used in practice. Neural networks (NN) models based on regression trees (RT) and genetic programming (GP) models were analyzed. In this work, for the first time, the method of multicriteria compromise ranking was applied to find the optimal model for the prediction of prestressed steel in prestressed concrete bridges. The optival model based on GP was determined using the VIKOR method of multicriteria optimization; the accuracy of which is expressed through the MAPE criterion is 9.16%. A significant average share of 46.11% of the costs related to steelworks, in relation to the total costs, indicates that the model developed in the paper can also be used for the implicit estimation of construction costs.

Automated summary

Accurately predicting the amount of prestressed steel is crucial for the successful design, construction, and performance of concrete-road bridges. It can optimize the design and construction process and help estimate project costs more accurately. This study developed a prediction model based on data from 74 constructed bridges in Serbia's Corridor X, analyzing neural networks (NN) models and genetic programming (GP) models that do not require extensive modeling expertise.