Machine learning for combinatorial optimization of brace placement of steel frames. Issue 4 (20th August 2018)

• Record Type: Journal Article
• Authors: Tamura, Takuya; Ohsaki, Makoto; Takagi, Jiro
• Is Part Of: Japan architectural review. Volume 1:Issue 4(2018)
• Page Start: 419
• Page End: 430
• Publication Date: 2018-08-20
• Subjects: binary decision tree -- braced frame -- machine learning -- optimization -- simulated annealing -- support vector machine; Building -- Periodicals; Building -- Japan -- Periodicals; Architectural design -- Periodicals; 690
• DOI: 10.1002/2475-8876.12059
• Languages: English
• ISSNs: 2475-8876

Abstract

Abstract: A method is presented for optimal placement of braces of plane frames using machine learning. The frame is subjected to static horizontal loads representing seismic loads. We consider the process of seismic retrofit by attaching braces. Therefore, the maximum value of additional stresses in the existing beams and columns and the maximum interstory drift angle are incorporated in the optimization problem. Characteristics of approximate optimal solutions and nonoptimal solutions are extracted using machine learning based on support vector machine and binary decision tree. Convolution and pooling are used for defining the features characterizing the solutions while reducing the number of variables. Optimization is carried out using a heuristic algorithm called simulated annealing based on local search. It is shown in the numerical examples that the computational cost is successfully reduced by avoiding costly structural analysis for a solution judged by machine learning as nonoptimal, and the important features in approximate optimal and nonoptimal solutions are identified. Abstract : Machine learning is applied to reduce the computational cost for optimization of brace locations of building frames. The properties of approximate optimal solutions and non‐optimal solutions are identified by binary decision tree and support vector machine.