Neural network-based prediction of stability coefficients for pultruded Fiber-Reinforced Polymer columns

Pultruded fiber-reinforced polymer (FRP) columns are widely used in infrastructure due to their excellent mechanical properties. However, their anisotropic properties and low transverse stiffness pose significant challenges for accurate stability coefficient prediction. Traditional theories, such as Euler’s and Perry’s formulas, rely on idealized assumptions that neglect material heterogeneity and initial imperfections, leading to inconsistent results. While artificial neural network (ANN) offers improved accuracy, their “black-box” nature limits engineering applicability. To address these limitations, this study proposes a novel single-layer ANN-based explicit solution for predicting stability coefficients of FRP columns. A database of 348 samples was constructed to establish the model. By simplifying the ANN architecture and extracting weight coefficients, an explicit formula (ANN formula) was derived. Comparisons with traditional methods demonstrated the ANN formula’s superior performance, achieving R² values of 0.937, outperforming Euler’s (R² = 0.753) and Perry’s (R² = 0.862) formulas. The MLP model exhibited exceptional accuracy (R² = 0.971). The proposed explicit solution uniquely bridges data-driven precision and engineering transparency, eliminating reliance on restrictive theoretical assumptions. This work advances FRP design by providing a mechanics-guided, interpretable tool for stability prediction, enabling safer and more efficient structural optimization. The methodology also establishes a framework for integrating machine learning into engineering standards for complex composite systems.