Open Access

ARTICLE

A Data Science Framework for Predicting the Creep Rupture Life of 1.25Cr- 0.5Mo Steel for Elevated Temperature Applications

Muhammad Ishtiaq, Yeonwoo Kim, Sung-Gyu Kang^*, Nagireddy Gari Subba Reddy^*

Department of Materials Engineering and Convergence Technology, Gyeongsang National University, 501, Jinju-daero, Jinju, Gyeongsangnam-do, Republic of Korea

* Corresponding Authors: Sung-Gyu Kang. Email: ; Nagireddy Gari Subba Reddy. Email:

(This article belongs to the Special Issue: Machine Learning Methods in Materials Science)

Computers, Materials & Continua 2026, 87(3), 34 https://doi.org/10.32604/cmc.2026.077416

Received 09 December 2025; Accepted 20 February 2026; Issue published 09 April 2026

Abstract

The long-term reliability of 1.25Cr-0.5Mo steels in high-temperature service critically depends on their creep rupture behavior, which is strongly influenced by alloy composition, microstructural characteristics, and testing conditions. In this study, an advanced Artificial Neural Network (ANN) model was developed to accurately predict the creep-rupture life of 1.25Cr-0.5Mo steels, offering a data-driven framework for alloy design and service-life assessment. The model incorporated eleven compositional variables (C, Si, Mn, P, S, Ni, Cr, Mo, Cu, Al, N), average grain size, non-metallic inclusions (NMI), steel properties including hardness measured on the Rockwell B scale (HRB) yield strength (MPa), ultimate tensile strength (MPa), elongation (%), reduction in area (%), and test conditions including temperature (°C) and stress (MPa) as input features, with rupture time as the output. A total of 276 experimental datasets were compiled, of which 219 were used for training and 57 for testing. To optimize predictive performance, a systematic hyperparameter evaluation was performed. Network architectures with one to three hidden layers and 10–30 neurons per layer were examined. The optimal configuration—three hidden layers with 21 neurons—achieved outstanding predictive accuracy, yielding an RMSE of 0.00007, an adjusted R² of 0.9930, a Pearson’s r of 0.9965, and a minimum MAE of 0.0504. Further optimization of training parameters showed that a momentum coefficient of 0.6 and a learning rate of 0.7 provided the most stable convergence behavior, while 9000 training iterations produced the lowest RMSE (0.000021). Five-fold cross-validation was employed to further assess the model’s predictive reliability and generalization. The predictive performance of the developed ANN model was further compared with multiple established machine-learning approaches to demonstrate its relative accuracy and generalization capability. The optimized ANN model was deployed in a user-friendly graphical interface (GUI) to facilitate practical implementation. Sensitivity analyses using both single-variable and two-variable approaches revealed the dominant role of key alloying elements, as well as the strong effects of test temperature and rupture stress on rupture life. The developed data science framework provides a powerful and reliable tool for predicting creep-rupture life across a broad compositional and testing window, enabling accelerated design and optimization of 1.25Cr-0.5Mo steels for high-temperature applications.

---

Keywords

---

Supplementary Material

Supplementary Material File

---