Estimation of capacity of eccentrically loaded single angle struts with decision trees

Özet

Single angle struts are used as compression members for many structures including roof trusses and transmission towers. The exact analysis and design of such members is challenging due to various uncertainties such as the end fixity or eccentricity of the applied loads. The design standards provide guidelines that have been found inaccu-rate towards the conservative side. Artificial Neural Networks (ANN) have been ob-served to perform better than the design standards, when trained with experimental data and this has been reported literature. However, practical implementation of ANN poses problem as the trained network as well as the knowhow regarding the application should be accessible to practitioners. In another data-driven tool, the De-cision Trees (DT), the practical application is easier as decision based rules are gen-erated, which are readily comprehended and implemented by designers. Hence, in this paper, DT was explored for the evaluation of capacity of eccentrically loaded sin-gle angle struts and was found to be robust and yielded comparable accuracy as ANN, and better than design code (AISC). This has enormous potential for easy and straight-forward implementation by practicing engineers through the logic based decision rules, which would be easily programmable on computer. For this application, use of dimensionless ratios as inputs for the development of DT was found to yield better results when compared to the approach of using the original variables as inputs.