The five networks developed in this study were trained and tested using a collection of existing cablestayed bridges. Excellent agreement between the outputs obtained by the present neural networks and the actual data is generally confirmed. This signifies that the trained networks have captured the relationships embedded in the input and output data, and in turn confirms the sufficiency of the input parameters to characterize their relationships with the required output.
The developed networks were then applied to an actual cable-stayed bridge under construction in Bangkok, the Rama VIII bridge. All outputs from the networks show a high degree of accuracy, except the specific network that predicts the height of the pylon and the depth of the bridge deck. This is due to the fact that the depth of the bridge deck has been restricted at the outset. After the network was modified to include the deck depth as an additional input parameter, the predicted height of the pylon was determined to be much closer to the actual design pylon height.
With these results, it is confirmed that ANN can be conveniently used as a tool to assist engineers in the conceptualization of cable-stayed bridges. This tool is also handy for preparing the conceptual design of a cable-stayed bridge so that a cost estimation can be carried out for the bidding under the design-build concept.
đang được dịch, vui lòng đợi..
