Results indicate that support vector regression (SVR) provides the most consistent overall performance across all regimes and offers a strong balance between accuracy and computational efficiency, and a Bayesian neural network (BNN) achieves competitive predictive performance while additionally enabling uncertainty estimation.
Abstract
Accurate prediction of creep rupture life is essential for ensuring the long-term reliability of high-temperature components in power generation and petrochemical industries. Selecting appropriate data-driven models for limited and heterogeneous creep datasets remains a critical challenge, as conventional accuracy-based comparisons do not fully capture model behavior under varying service conditions. This study presents a unified evaluation framework for systematically comparing multiple machine learning and deep learning models for creep rupture life prediction of 2.25Cr–1Mo steel. The framework integrates predictive accuracy, prediction reliability, regime-specific error analysis, and computational efficiency, enabling a comprehensive assessment beyond global error metrics. The input feature space is reduced from seventeen to eight physically meaningful variables without loss of predictive performance. To further assess model robustness, prediction errors are analyzed across four distinct rupture life regimes, revealing significant variations in model behavior that are not reflected in aggregate metrics. Results indicate that support vector regression (SVR) provides the most consistent overall performance across all regimes and offers a strong balance between accuracy and computational efficiency. Among deep learning models, a Bayesian neural network (BNN) achieves competitive predictive performance while additionally enabling uncertainty estimation. These findings demonstrate that, for small tabular creep datasets, appropriately regularized models outperform complex neural network architectures, highlighting the importance of matching model complexity to dataset characteristics. This study is limited to a single steel grade, moderate dataset size, and extrapolation beyond trained stress and temperature ranges, which are key directions for future work.
The aim of structural reliability assessment (SRA) is to guarantee the safety, durability, and performance of structures; however, traditional methods like stochastic finite element analysis (SFEA) can be computationally prohibitive to use in practical situations. This paper introduces a novel framework for SRA utilizing deep neural networks (DNNs) implemented in an open-source program called TRA-DNN, replacing the resource-intensive finite element (FE) analysis with a DNN model. The DNN is trained using 6874 FE column models, including factors like geometric imperfections, resulting in a training database with 419,314 data records. It accurately predicts axial load-deformation curves for corroded steel CHS columns, enabling the determination of the ultimate capacities for columns with varying properties. The model’s accuracy is confirmed through rigorous quantitative and qualitative validation, including various failure modes. TRA-DNN employs the DNN model to perform SRA via Crude Monte Carlo Simulation (MCS), yielding results that are in high agreement with conventional SFEA, yet with significantly reduced computational time (1,388,250 times faster). In addition, TRA-DNN can be used to estimate the service life of CHS columns considering both corrosion propagation and load increase with time. Future research can utilize TRA-DNN to optimize column design and maintenance to minimize both risk and cost.
Assem Atif Farag, Alaa Elsisi, A. Eraky et al.· Applied Sciences· 0 citations
ABSTRACT An attention‐based multimodal deep learning framework is developed to fuse processing parameters with microstructural micrographs for predicting creep rupture life of IN718 under a fixed creep testing condition. Under a predefined composition‐stratified, sample‐level split, the framework achieved a mean test‐set R2 of 0.917 ± 0.014 across 50 random‐seed training repetitions. The corresponding mean RMSE and MAPE were 0.14% and 6.0%, respectively. Interpretability analyses suggest that the predictions are consistent with established metallurgical understanding, particularly the important role of δ‐phase characteristics. Furthermore, uncertainty quantification endows the model with self‐assessment capabilities, allowing it to reliably quantify the confidence of its predictions. This study establishes a methodological framework that unifies predictive accuracy, physical interpretability, and model confidence, providing a validated paradigm for developing trustworthy AI models for materials design under data‐limited conditions.
Haopeng Lv, Dayong Wu, Z. Rao et al.· Advancement of science· 0 citations
Remaining Useful Life (RUL) prediction for turbofan engines is critical for balancing operational safety against maintenance costs and environmental impact from premature replacements. Models must reliably extrapolate beyond training data, yet no single method performs optimally across all operational contexts, making model selection fundamentally heuristic. Rather than demonstrating individual model superiority, this work recognizes that reliable prognostic performance emerges from adaptive model contribution. This work introduces an ensemble framework integrating three components: (1) legacy-representative data splitting that mirrors realistic deployment where models predict for newer assets using historical data, (2) median absolute deviation-based outlier filtering for stability, and (3) WTA³ weighting that dynamically adjusts model influence based on cycle-by-cycle performance. This treats model coordination as a time-varying optimization problem adapting to evolving degradation patterns. The framework is validated on NASA CMAPSS data using six diverse models spanning traditional machine learning (Random Forest, XGBoost, Support Vector Regression) and deep learning (LSTM, CNN, Transformer). We investigate how prediction robustness changes during temporal extrapolation, whether adaptive weighting provides more stable forecasts than individual models or fixed combinations, and how uncertainty quantification supports safer maintenance decisions. Initial validation demonstrates that the WTA³ meta-ensemble achieves approximately 3 cycles RMSE and under 3 cycles MAE, representing over 30% improvement over the best individual model, with particularly strong performance in the critical late-life phase. The ensemble maintains highly stable predictions with well-calibrated confidence intervals and substantially improved coverage compared to individual models. This work reframes prognostics from competitive model selection to adaptive coordination, demonstrating that ensemble stability under extrapolation can be systematically achieved. The approach provides actionable confidence bounds for aerospace maintenance programs, enabling cost-efficient scheduling while reducing environmental waste, directly supporting both economic and sustainability objectives where data scarcity and safety criticality are paramount.
Ayushi Bharti, N. Kim, Hee-Cheol Kim· e-Journal of Nondestructive...· 0 citations
A physics-guided machine learning framework that integrates domain-informed feature engineering, conditional synthetic data augmentation, and stacking ensemble learning to predict the chloride diffusion coefficient of concrete from Rapid Chloride Migration (RCM) test data is presented.
Moutaman M. Abbas· Journal of Composites Scienc...· 0 citations
Accurate prediction of the residual drift ratio of reinforced concrete bridge piers is challenging because conventional methods are computationally expensive, time-consuming, and unable to effectively capture complex nonlinear interactions among multiple influencing factors. To address these limitations, this study proposes an interpretable machine learning framework for predicting the residual drift ratio of reinforced concrete bridge piers. A comprehensive database containing 261 quasi-static experimental datasets was established, incorporating key structural and material parameters, including axial com-pression ratio, shear span ratio, stirrup ratio, longitudinal reinforcement ratio, material strengths, and geometric dimensions. Based on this database, six representative machine learning models were developed and systematically compared. Their predictive performance, robustness, and generalization capability were evaluated using multiple statistical metrics and Monte Carlo simulations. The results show that the CatBoost model consistently outperformed the other models, achieving an R2 value of 0.9629 on the test set while maintaining excellent stability under random data partitions. Furthermore, SHAP analysis was employed to interpret the trained model and quantify the contributions of individual input variables. Eight key factors influencing the residual drift ratio were identified, with the loading displacement ratio (θ) exhibiting the greatest influence. These findings demonstrate that the proposed framework provides an accurate, reliable, and interpretable tool for predicting the post-earthquake residual drift ratio of reinforced concrete bridge piers, offering valuable support for performance-based seismic design, post-earthquake damage assessment, and resilience-based bridge engineering.
Min Zhang, Xuefeng Zhang, Liang-Jun Li et al.· Buildings· 0 citations
A novel hybrid method integrating physics-informed neural network (PINN) and Gaussian process regression (GPR) that enforces the B4 creep model as a physics-informed constraint by embedding its governing equations into the loss function, effectively incorporating physical knowledge into data-driven training.
Zhiren Tao, Jianxin Peng, Shijie Liao et al.· Journal of materials in civi...· 0 citations