PSI - Issue 81

Available online at www.sciencedirect.com

ScienceDirect

Procedia Structural Integrity 81 (2026) 35–40

© 2026 The Authors. Copy from the contract: Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of DMDP 2025 organizers Keywords: SMA; artificial intelligence; machine learning; frequency; Voting model 1. Introduction Shape memory alloys (SMAs) are characterized by unique functional properties such as the shape memory effect and superelasticity, which result from reversible martensitic to austenitic phase transformations. Owing to these properties, SMAs have been widely applied in various areas of engineering and science (Chaudhary et al., 2024; Sharma and Srinivas, 2020; Hartl and Lagoudas, 2007; Niu et al., 2025; Schmelter et al., 2024; Riccio et al., 2021; Zhang et al., 2024). Due to superelasticity, the material can withstand large deformations and fully recover its original shape after the load is removed. The hysteresis observed Abstract Shape memory alloys (SMAs) are characterized by complex hysteresis behavior resulting from phase transformations between martensite and austenite, which is significantly affected by the frequency of cyclic loading. Accurate prediction of these processes is essential to ensure the reliability of structural components in various engineering applications. This paper presents an approach for predicting the hysteresis behavior of SMAs based on an ensemble Voting machine learning model. The ensemble included Random Forest, Gradient Boosting, Extra Trees, Support Vector Regressor, K-Nearest Neighbors, and a Multilayer Perceptron. The model weights were determined as the inverse of the mean squared error, which ensured a balanced contribution of each algorithm to the final prediction. The model performance was evaluated using the MAE, MSE, R², and MAPE metrics. The results demonstrated high prediction accuracy (R 2 > 0.998, MAE < 0.022, MSE < 0.0008, and MAPE < 0.008) and confirmed the ability of the model to generalize across independent cycles, including extrapolated ones (251 and 300). The predicted hysteresis loops showed good agreement with the experimental curves. The obtained results confirm the effectiveness of the ensemble approach for modeling the behavior of SMAs and predicting their functional properties. VIII International Conference “In - service Damage of Materials: Diagnostics and Prediction“ (DMDP 2025) Modeling the Hysteresis Behavior of SMA by an Ensemble Voting Machine Learning Model Dmytro Tymoshchuk a, *, Oleh Yasniy a , Yuri Lapusta b , Iryna Didych a , Iaroslav Pasternak c , Volodymyr Iasnii a a Ternopil Ivan Puluj National Technical University, Ruska str. 56, Ternopil, 46001, Ukraine b Université Clermont Auvergne, CNRS, Clermont Auvergne INP, Institut Pascal, F-63000 Clermont-Ferrand, France c Lesya Ukrainka Volyn National University, 13 Voli Ave., Lutsk, 43025, Ukraine

* Corresponding author. Tel.: +380352519700. E-mail address: dmytro.tymoshchuk@gmail.com

2452-3216 © 2026 The Authors. Copy from the contract: Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of DMDP 2025 organizers 10.1016/j.prostr.2026.03.007