Interpretable domain knowledge enhanced machine learning framework on axial capacity prediction of circular CFST columns - Abstract & framework

Aug 3, 2026

Upload by Hazel

This study proposes an interpretable machine learning framework that embeds domain knowledge to predict the axial bearing capacity of concrete-filled steel tube (CFST) members. A Domain Knowledge Enhanced Neural Network (DKNN) is trained on 2621 experimental data points and uses feature engineering through Pearson correlation, XGBoost, and random tree methods. The DKNN reduces MAPE by over 50% versus existing models and remains accurate under noisy conditions. Sensitivity and SHAP analyses quantify parameter contributions and support design recommendations for diameter, material strength range, and material combinations, advancing CFST predictive modeling.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages33

Reading Time1 h 23 min

AccessFree to download

Introduction

Motivation and background

Limitations of black-box neural networks and data-driven ML

Domain knowledge-aided neural network methods

Methods and modeling approach

Domain Knowledge Enhanced Neural Network (DKNN)

Feature engineering strategy

Results and validation

Prediction accuracy improvement and robustness

Sensitivity analysis and SHAP interpretation

Design recommendations

Interpretable domain knowledge enhanced machine learning framework on axial capacity prediction of circular CFST columns Dian Wanga, Zhigang Rena *, Gen Kondob * a School of Civil Engineering and Architecture, Wuhan University of Technology, No. 122 Luoshi Road, Wuhan 430070, China b Department of Civil and Environmental Engineering, University of California, Berkeley, CA 94720, USA * Corresponding authors. E-mail address: [renzg@whut.edu.cn](renzg@whut.edu.cn) (Z. Ren), [kondogen@berkeley.edu](kondogen@berkeley.edu) (G. Kondo) Abstract: This study introduces a novel machine learning framework, integrating domain knowledge, to accurately predict the bearing capacity of CFSTs, bridging the gap between traditional engineering and machine learning techniques. Utilizing a comprehensive database of 2621 experimental data points on CFSTs, we developed a Domain Knowledge Enhanced Neural Network (DKNN) model. This model incorporates advanced feature engineering techniques, including Pearson correlation, XGBoost, and Random tree algorithms. The DKNN model demonstrated a marked improvement in prediction accuracy, with a Mean Absolute Percentage Error (MAPE) reduction of over 50% compared to existing models. Its robustness was confirmed through extensive performance assessments, maintaining high accuracy even in noisy environments. Furthermore, sensitivity and SHAP analysis were conducted to assess the contribution of each effective parameter to axial load capacity and propose design recommendations for the diameter of cross-section, material strength range and material combination. This research advances CFST predictive modelling, showcasing the potential of integrating machine learning with domain expertise in structural engineering. The DKNN model sets a new benchmark for accuracy and reliability in the field. Keywords: Concrete-filled steel tube, Machine learning, Axial compression capacity, Domain knowledge, Neural Network, SHAP 1. Introduction Concrete-filled steel tube (CFST) structures are extensively utilized in engineering for their high strength, stability, and durability. However, predicting the bearing capacity of CFST members under load accurately is challenging, influenced by factors like material properties, structural forms, and environmental conditions. Conventional methods, based on empirical formulas and experimental data, struggle with complex non-linear material behavior and intricate structural configurations, limiting their accuracy and applicability [1]. Advanced analytical techniques, such as neural networks (NNs), have gained attention as a preferred method for the prediction of bearing capacity. NNs' ability to learn complex nonlinear relationships from large datasets makes them a compelling choice for accurately calculating CFSTs' bearing capacity, drawing significant research interest [2–4] . While NNs exhibit remarkable adaptability and predictive capabilities, they have certain limitations. Firstly, NN models are considered black-box models, making it difficult to interpret their decision-making process. This lack of transparency is problematic in engineering, where practical applications demand explanations and validation of predictions [5] . Secondly, traditional data-driven machine learning (ML) methods may yield unreliable outputs in data spaces lacking information, affecting generalization ability due to data distribution and sample size limitations [6]. A summary of recent papers related to NN models in the field of CFST structures is presented in Table 1. Addressing these issues to advance ML in CFST structures requires incorporating domain knowledge and practical application requirements before implementing ML, ensuring meaningful interpretation and utilization of prediction results. In recent years, domain knowledge-based optimization and prediction methods for NNs, such as Physics-Informed NNs (PINN) and Physics Model-b

FAQs

What problem does the DKNN framework address in CFST axial capacity prediction?

It aims to accurately predict axial bearing capacity of circular CFST columns while bridging the gap between traditional engineering approaches and machine learning, by embedding domain knowledge into the learning process.

How is the DKNN model built and trained?

The model is trained using a comprehensive dataset of 2621 experimental data points and incorporates feature engineering based on Pearson correlation, XGBoost, and random tree algorithms.

What methods are used to interpret the model and guide design recommendations?

Sensitivity analysis and SHAP are used to evaluate how each effective parameter contributes to axial load capacity, and the findings are used to propose recommendations for cross-section diameter, material strength range, and material combinations.

Aug 3, 2026

Upload by Hazel