Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{210861,
        author = {Appalanaidu Yalla and Dr. Sharmila Sujatha},
        title = {Structural Reliability Analysis using Machine Learning Techniques},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {6},
        pages = {2346-2355},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=210861},
        abstract = {Structural reliability analysis estimates the probability of failure Pf of a component whose material, geometric and loading parameters are uncertain. Monte Carlo simulation (MCS), the first- and second-order reliability methods (FORM and SORM) and importance sampling (IS) become costly or inaccurate when limit-state evaluations are expensive or strongly nonlinear. This paper develops a MATLAB framework in which machine-learning surrogates, namely the artificial neural network (ANN), support vector machine (SVM), support vector regression (SVR), Gaussian process regression (GPR/Kriging), active-learning Kriging (AK) and random forest (RF), replace the limit-state function in MCS. It is applied to two mechanical components: a beam in bending with ten uncertain variables (a 10,000-sample dataset with 1.46% failures) and a shaft under combined bending and torsion governed by the von Mises criterion (100,000 samples). For the beam, surrogates trained on 7,000 samples evaluated 100,000 Monte Carlo samples; the ANN, SVR and GPR estimates of Pf (0.01388, 0.01449 and 0.01448) lie within the sampling uncertainty of the dataset value 0.0146, and GPR fits best (coefficient of determination R2 = 0.99998), whereas RF underestimates Pf (0.00517) despite R2 = 0.92252. For the shaft, MCS gives Pf = 0.33362, and ANN-MCS, Kriging-MCS and AK-MCS lie within 0.4% of it, while FORM and SORM differ by 4.6% and 2.7%. Because both limit states are analytical, direct MCS was faster than every surrogate; surrogates are therefore justified mainly where each evaluation needs finite-element or other costly analysis.},
        keywords = {Active learning, Artificial neural network, Gaussian process regression, Kriging, Monte Carlo simulation, Reliability index, Structural reliability, Support vector regression, Surrogate modelling},
        month = {November},
        }