Published research / software

Cellular-beam design tools.

Connect performance screening, candidate comparison and geometry output.

Supplied cellular-beam design workflow: inputs, FE/surrogate evidence, inverse design and fabrication-oriented outputs. Publication · 10.1016/j.advengsoft.2026.104262

Role: First author · publisher-credited software contributor
Period: 2026
Methods: Surrogates · Inverse design
Status: Published research / software

Engineering software / Streamlit

A defined input.
An inspectable decision.

For steel design and computational engineering teams

Connect performance screening, candidate comparison and geometry output.

01 / INPUT

Set the engineering case

Target resistance, span, opening layout and parent-section candidates

02 / OUTPUT

Inspect the result

Ranked candidates, robustness checks, scenario metrics and CAD outputs

03 / DECISION

Compare the next options

Shortlist cellular-beam designs with explicit feasibility checks.

Technical basis

Surrogates · Inverse design

Validation evidence

Published software contribution, external FE database and numerical validation with false-feasible outcomes reported.

Current scope

Research software. The external FE dataset is attributed to its creators; screening does not replace final design verification.

Public deployment has not been verified. The supplied interface and source establish the implementation shown.

Engineering record

Scope, method and evidence.

External FE database and numerical validation; no commercial deployment or engineering certification claimed.

01From a resistance target to candidate geometry

Cellular-beam design couples parent-section selection, opening geometry, structural resistance and fabrication constraints. Starting with an arbitrary geometry can create repeated forward checks. This research framework instead starts with a prescribed resistance target and retrieves admissible parent-section alternatives.

The inverse problem has multiple possible solutions. The output is a screened set of candidates and their comparison, rather than a single geometry asserted to be uniquely correct.

02My contribution and the data basis

I am the first author of the supplied manuscript, with Felipe Piana Vendramell Ferreira, Rabee Shamass, Konstantinos Daniel Tsavdaridis, Ikram Abarkane and Musab Rabi. Manuscript, code files and validation tables document the research-software workflow.

The resistance surrogate uses an external database of 14,094 FE-derived beam cases. The database provenance is credited to the work of Degtyarev, Hicks, Ferreira and Tsavdaridis. These cases are not presented as FE simulations personally generated by me.

Status: published collaborative research in Advances in Engineering Software 221 (October 2026), 104262. Publisher record. Production deployment is not established.

03Computational workflow

Predict a resistance response
A forward surrogate maps geometric, material and opening-layout variables to FE-derived resistance. Monotonic constraints encode selected relationships during training.
Retrieve admissible alternatives
Search discrete parent sections against a prescribed target, fixed span and opening conditions. Apply geometric admissibility and the framework’s feasibility checks before ranking.
Compare retained candidates
Evaluate controlled geometric perturbations and multi-objective rankings. Carbon indicators are scenario-based comparisons, not verified project savings.
Reconstruct selected geometry
The documented workflow includes CAD-compatible geometry generation and an interface for examining candidate outputs. This is downstream representation, not a fabrication approval.

04Validation: retain the failures as well as the matches

The supplied feasibility table evaluates 6,243 candidate assessments across 459 scenarios. “Feasible” below refers to the table’s FE-labelled classification task and its chosen criteria.

FE classification outcomes: 3029 true feasible, 249 false feasible, 157 false infeasible and 2808 true infeasible. — Replotted from the supplied FE inverse-feasibility classification table. Counts are numerical-validation outcomes, not constructed assets. Open the figure for a larger view.

Scroll the table horizontally to inspect all columns.

Overall FE classification outcomes
Measure	Value	Interpretation
True positives / false positives	3,029 / 249	Predicted-feasible assessments classified feasible / infeasible by the FE label.
False negatives / true negatives	157 / 2,808	Predicted-infeasible assessments classified feasible / infeasible by the FE label.
Precision / recall	0.924 / 0.951	Classification metrics within this validation population.
False-feasible rate	7.60%	249 of 3,278 predicted-feasible assessments; not a claim about field failure probability.

A separate near-target evaluation returned 878 retained candidates for 408 of the 459 scenarios, using a ±10% target band. Target consistency must be read with its denominator: it is not a structural safety certificate.

Scroll the table horizontally to inspect all columns.

Near-target evaluation — overall results
Measure	Result	Population / condition
Scenario return rate	408 / 459 (88.89%)	Scenarios returning near-target candidates.
Returned-set FE target-band consistency	89.07%	Retained candidate set.
Top-1 FE target-band consistency	91.91%	Top-ranked returned candidates.
Median top-1 FE relative error	2.48%	Reported median in the supplied summary table.

Source: supplied feasibility and near-target validation tables associated with the research manuscript. Values are rounded for readability. Download the classification table (CSV).

05What the results do and do not establish

The evaluation demonstrates a bounded candidate-retrieval and screening workflow. False-feasible cases remain material: surrogate screening does not replace independent verification of a retained design.

Results apply to the supplied FE-derived domain and the framework’s defined beam assumptions, including laterally restrained configurations. Domain transfer, connection behaviour, other loading conditions and fabrication acceptance require separate assessment.

Robustness perturbations and sustainability scenarios inform comparison within the tool. They do not establish delivered tolerances, certified resistance, real material savings or a complete project life-cycle assessment.

06Related capability

Engineering R&D connects calculation rules, model output and decision support. Structural Analysis & FE Modelling provides the underlying mechanics and verification context.

Discuss an engineering workflow

Research supporting this project