Physics-informed TVAE workflow for data augmentation, mechanical validation, optimization of CFRP-strengthened CFST beams

TL;DR

This paper introduces a physics-informed data augmentation workflow for designing CFRP-strengthened concrete-filled steel tube beams.

Contribution

A novel physics-informed workflow that integrates data augmentation, mechanical validation, and machine learning for structural engineering.

Findings

The workflow generates synthetic datasets that preserve statistical fidelity while enforcing physical constraints.

An open-source software tool is provided to support reproducibility and practical engineering decision-making.

The method enables multi-objective optimization of CFRP configurations using ensemble machine learning.

Abstract

This article presents a reproducible, physics-informed workflow designed to address data scarcity in the analysis and design of carbon fiber-reinforced polymer (CFRP)-strengthened concrete-filled steel tube (CFST) members. Conventional tabular generative models can reproduce statistical trends but cannot ensure that generated samples satisfy the physical, geometric, and material constraints required for structural engineering applications. To overcome this limitation, the proposed method embeds physics-based constraints directly into the data augmentation process and incorporates mechanics-based validation using nonlinear moment-curvature analysis to verify the physical admissibility of synthetic samples. The resulting datasets are suitable for downstream predictive modeling, design optimization, and decision support in data-scarce structural applications. The workflow further integrates ensemble machine learning models, explainable analysis, multi-objective optimization, and uncertainty quantification into a unified methodological framework. To support reproducibility and practical adoption, the complete methodology is implemented and released as an open-source, Streamlit-based software tool.•Generates synthetic structural datasets that preserve statistical fidelity while enforcing fundamental physical, geometric, and material constraints for CFRP-strengthened CFST members.•Enables design-oriented prediction, interpretation, and multi-objective optimization of CFRP configurations through an integrated ensemble machine learning framework.•Provides an open-source, deployable software implementation to support reproducibility and practical engineering decision-making for flexural design exploration of CFRP-strengthened CFST beams.

Generates synthetic structural datasets that preserve statistical fidelity while enforcing fundamental physical, geometric, and material constraints for CFRP-strengthened CFST members.

Enables design-oriented prediction, interpretation, and multi-objective optimization of CFRP configurations through an integrated ensemble machine learning framework.

Provides an open-source, deployable software implementation to support reproducibility and practical engineering decision-making for flexural design exploration of CFRP-strengthened CFST beams.

Image, graphical abstract

Figures