Cold-formed steel framing design with data-driven models

Cristopher D. Moen, Ph.D., P.E., F.SEI

December 7, 2023

Cold-formed steel framing

Our industry wins with system-level design: framing, metal buildings, joists, racks,…

Data-driven models

They are a tool for resolving the unknown.

Team effort

There is an engineering-focused group working on data-driven models.

https://github.com/EngineerWithData

Column curve

Imperfect column behavior

Chajes, A. (1974). Principles of structural stability theory.

P-M interaction

OG MIM

System behavior

Vieira Jr, L. C. M., & Schafer, B. W. (2013). Behavior and design of sheathed cold-formed steel stud walls under compression. Journal of Structural Engineering, 139(5), 772-786.

System benefits

Questions

Modern data-driven models

What is different between old school and new school data-driven models?

Regression

Prediction error

Minimize error

Networks

Decision trees

Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd international conference on knowledge discovery and data mining (pp. 785-794).

Decision trees (CFS)

Neural networks

Abraham, T. H. (2002). (Physio) logical circuits: The intellectual origins of the McCulloch–Pitts neural networks. Journal of the History of the Behavioral Sciences, 38(1), 3-25.

Neural networks

Large language models

Wolfram, Stephen (2023). What is ChatGPT Doing… and Why Does it Work?

Large language models (probability)

Questions

Data

Our industry is producing lots of it.

CFS-focused databases

The data dream

CFS engineering with ‘next level’ insight…

  • machine-readable
  • common data structures
  • high quality
  • user interfaces
  • collection automation
  • reliable models

Data structures

Moen, C. D., & Chen, B. (2022). An open-source cold-formed steel connection test database to support future data models. Cold-Formed Steel Research Consortium (CFSRC) Colloquium.

struct Source
    authors::Array{String}
    date::Date
    title::String
    bibtex::String
    units::Array{String}
    nominal_data::Vector{String}
    notes::String
end

struct Fastener
    type::Vector{String}
    details::Vector{Dict}
end

struct Ply
    type::Vector{String}
    thickness::Array{Float64}
    elastic_modulus::Array{Any}
    yield_stress::Array{Any}
    ultimate_stress::Array{Any}
end

struct Test
    name::String
    loading::String
    force::Array{Float64}
    displacement::Array{Float64}
end

struct Specimen
    source::Array{Source}
    fastener::Fastener
    ply::Ply
    test::Test
end

Machine-readable data

For example, JSON.

Data collection automation

See this flowchart.

Data user interface

CFS framing data-driven models

AISI S100-16 + LLM

Google Colab

Composite steel deck design

Database

Hollow steel column strength prediction

Meng, X., & Gardner, L. (2020). Behavior and design of normal-and high-strength steel SHS and RHS columns. Journal of Structural Engineering, 146(11)

Google Colab

Screw-fastened connection backbone prediction

Google Colab

Interior partition wall composite flexural stiffness

Interior partition wall composite flexural stiffness

Google Colab

EngineerWithData

Conclusions

  • Thanks to CFSEI (Andy, Rose, Bill, and company)
  • Try to build your own data-driven model!

Quarto

Quarto.org

Contact

Questions