PSI - Issue 72

Oleh Yasniy et al. / Procedia Structural Integrity 72 (2025) 188–194

189

1. Introduction Shape memory alloys (SMAs) are unique materials that can restore their previous shape under certain conditions, such as temperature changes or mechanical stress. These properties allow SMAs to be used in many fields, including aerospace by Bettini et al. (2021), construction by Noruzi et al. (2024), and medicine by Kim et al. (2020). Shape memory effect (SME) is the property of an alloy to return to its original shape when heated after being deformed in the cold state. Superelasticity (SE) is the ability of an alloy to withstand large deformations and return to its original shape after the loading is removed without heating. The frequency of applied loading can significantly affect the functional properties of SMA by Iasnii et al. (2024). This is particularly important for applications where alloys are subjected to cyclic loads, such as vibration dampers. The frequency of loading can affect thermal processes in the material. Studying the effect of loading frequency on the functional properties of SMAs is critical to their successful application in real-world structures and devices. This helps developers and engineers choose the optimal parameters for using these materials, ensuring their reliability and durability. Machine learning is utilized in various fields of science and technology. In materials science, its methods are used to predict the properties of alloys and composites by Yasniy et al. (2023), by Maruschak et al. (2024), by Tymoshchuk et al. (2024). In cybersecurity, machine learning helps to detect and prevent attacks by analyzing network traffic and behavioral patterns, which allows for the effective detection of DDoS attacks and other threats by Klots et al. (2024) by Tymoshchuk et al. (2024). In the automotive industry, it is used to develop autonomous vehicles and predict maintenance by Lyashuk et al. (2024). Non-linear dependencies of SMA parameters arise due to the complex interaction between phase transformations, temperature, stresses, and cyclic loading frequency. This makes it difficult to model them using traditional mathematical approaches. Therefore, machine learning methods are often used to more accurately predict the behavior of SMAs, which can consider these nonlinear dependencies and complex interactions between parameters by Yasniy et al. (2020) by Yasniy et al. (2024). 2. Material and methods This paper investigates the properties of SMA based on nickel and titanium (NiTi), namely the superelasticity effect. This property is caused by phase transformations between the martensitic and austenitic phases, which occur under mechanical stress. The superelasticity allows the alloy to withstand significant deformations and return to its original shape after removing the loading without residual deformation. Artificial neural networks (ANNs) can effectively predict the hysteresis behavior of SMAs and their behavior under different loading frequencies. This approach is an effective tool for analyzing the impact of different loading frequencies on the behavior of alloys. The study main objective is to create models that can accurately predict hysteresis loops for different frequencies, considering the complexity of the processes occurring during deformation and phase transformation. To train the artificial neural network, the experimentally obtained hysteresis loops were utilized for four loading frequencies of 0.1, 0.5, 1, and 5 Hz by Iasnii et al. (2022). The input data consisted of the following parameters: stress  (MPa), cycle number N , and loading frequency f (Hz). By analyzing these functional dependencies, the neural network learned to recognize patterns, which allowed it to predict the material strain  (%) at different loading cycle frequencies. The hysteresis behavior of SMA is modeled by an artificial neural network of the multilayer perceptron (MLP) type by Haykin (2009). MLP is a classical architecture of artificial neural networks that consists of one or more Nomenclature  the stress the strain  N number of the loading and unloading cycle f the frequency