Machine learning design of R/C columns

Abstract

In this work, various functions developed with Machine Learning techniques are proposed for the rapid and accurate design of R/C columns and bridge piers. Both rectangular and circular as well as solid and hollow sections are examined. Using powerful modern-day hardware and software, it is found that large Artificial Neural Networks (ANNs), in tandem with carefully assembled large training sets, can yield models with adequate accuracy for design, clearly surpassing that of traditional design charts and practically equivalent to iterative section analysis procedures. The error estimation of each function is described in detail based on extensive test sets while auxiliary ANNs eliminate extrapolation issues. A computational performance comparison is also carried out, indicating that the proposed approach outperforms classic design algorithms by orders of magnitude while being naturally immune to numerical instabilities.