This paper investigates the use of machine learning methods to predict the loading frequency of shape memory alloys (SMAs) based on experimental data. SMAs, in particular nickel-titanium (NiTi) alloys, have unique properties that restore the original shape after significant deformation. The frequency of loading significantly affects the functional characteristics of SMAs. Experimental data were obtained from cyclic tensile tests of a 1.5 mm diameter Ni_55.8Ti_44.2 wire at different loading frequencies (0.1, 0.5, 1.0, and 5.0 Hz). Various machine learning methods were used to predict the loading frequency f (Hz) based on input parameters such as stress σ (MPa), number of cycles N, strain ε (%), and loading–unloading stage: boosted trees, random forest, support vector machines, k-nearest neighbors, and artificial neural networks of the MLP type. Experimental data of 100–140 load–unload cycles for four load frequencies were used for training. The dataset contained 13,365 elements. The results showed that the MLP neural network model demonstrated the highest accuracy in load frequency classification. The boosted trees and random forest models also performed well, although slightly below MLP. The SVM method also performed quite well. The KNN method showed the worst results among all models. Additional testing of the MLP model on cycles that were not included in the training data (200th, 300th, and 1035th cycles) showed that the model retains high efficiency in predicting load frequency, although the accuracy gradually decreases on later cycles due to the accumulation of structural changes in the material.

The thermal conductivity coefficient of epoxy composites for aircraft, which are reinforced with glass fiber and filled with aerosil, γ-aminopropylaerosil, aluminum oxide, chromium oxide, respectively, was simulated. To this end, various machine learning methods were used, in particular, neural networks and boosted trees. The results obtained were found to be in good agreement with the experimental data. In particular, the correlation coefficient in the test sample was 0.99%. The prediction error of neural networks in the test samples was 0.5; 0.3; 0.2%, while that of boosted trees was 1.5; 0.9%.

The paper presents a comparative analysis of various machine learning algorithms for the classification of epoxy composites reinforced with basalt fiber and modified with inorganic fillers. The classification is based on key thermophysical characteristics, in particular, the mass fraction of the filler, temperature, and thermal conductivity coefficient. A dataset of 16,056 interpolated samples was used to train and evaluate more than a dozen models. Among the tested algorithms, the MLP neural network model showed the highest accuracy of 99.7% and balanced classification metrics F1-measure and G-Mean. Ensemble methods, including XGBoost, CatBoost, ExtraTrees, and HistGradientBoosting, also showed high classification accuracy. To interpret the results of the MLP model, SHAP analysis was applied, which confirmed the predominant influence of the mass fraction of the filler on decision-making for all classes. The results of the study confirm the high effectiveness of machine learning methods for recognizing filler type in composite materials, as well as the potential of interpretable AI in materials science tasks.

The study presents an approach to predicting the hysteresis behavior of shape memory alloys (SMAs) using recurrent neural networks, including SimpleRNN, LSTM, and GRU architectures. The experimental dataset was constructed from 100 to 250 loading–unloading cycles collected at seven loading frequencies (0.1, 0.3, 0.5, 1, 3, 5, and 10 Hz). The input features included the applied stress σ (MPa), the cycle number N (the Cycle parameter), and the indicator of the loading–unloading stage (UpDown). The output variable was the material strain ε (%). Data for training, validation, and testing were split according to the group-based principle using the Cycle parameter. Eighty percent of cycles were used for model training, while the remaining 20% were reserved for independent assessment of generalization performance. Additionally, 10% of the training portion was reserved for internal validation during training. Model accuracy was evaluated using MAE, MSE, MAPE, and the coefficient of determination R2. All architectures achieved R2 > 0.999 on the test sets. Generalization capability was further assessed on fully independent cycles 251, 260, 300, 350, 400, 450, and 500. Among all architectures, the LSTM network showed the highest accuracy and the most stable extrapolation, consistently reproducing hysteresis loops across frequencies 0.1–3 Hz and 10 Hz, whereas the GRU network showed the best performance at 5 Hz. Model interpretability using the Integrated Gradient (IG) method revealed that Stress is the dominant factor influencing the predicted strain, contributing the largest proportion to the overall feature importance. The UpDown parameter has a stable but secondary role, reflecting transitions between loading and unloading phases. The influence of the Cycle feature gradually increases with the cycle number, indicating the model’s ability to account for the accumulation of material fatigue effects. The obtained interpretability results confirm the physical plausibility of the model and enhance confidence in its predictions.

Estimation de l'intégrité structurelle et de la durée de vie d’ importants éléments structurels: Ce travail a été réalisé sous co-tutelle entre l’Université Technique Nationale de Ternopil (Ukraine) et l’Université Clermont Auvergne, CNRS, SIGMA Clermont, l’Institut Pascal à Clermont-Ferrand (France). La thèse porte sur la solution d’une tâche scientifique réelle d’évaluer la résistance et la durabilité des éléments responsables des structures. L’objectif de l’étude est d’évaluer la résistance et la durabilité résiduelle des éléments structurels par des méthodes d’apprentissage automatique. La plupart des parties des machines et des composants des structures pendant l'exploitation sont influencés par des charges de nature différente. Ces forces sont soit directement attachées à l’élément, soit transmises par des éléments adjacents qui y sont reliés. Pour le fonctionnement normal des parties responsables des structures,chaque élément doit être d’une taille et d’une forme qui lui permettent de résister aux charges. En particulier, il doit être solide, pas avoir de déformation significative sous tension, rigide et conserver sa forme d’origine. La durée de vie estimée des machines et des structures peut être prédite à l’aide des diagrammes de la croissance des fissures de fatigue des matériaux. Dans la plupart des cas, les données expérimentales présentent certaines variations dont il faut tenir compte. L’expérimentation prend souvent beaucoup de temps et de ressources humaines. Par conséquent, il est conseillé d'apprendre à calculer la durabilité par des méthodes d'apprentissage automatique, en particulier les réseaux de neurones, les arbres renforcés, les forêts aléatoires, les machines à vecteurs de support et les k-plus proches voisins.

Fatigue crack growth rate is modelled by neural network and compared with analytical models, namely, the Paris' law and polynomial log regressions of the second to fifth degree. To assess the accuracy of the modelling, the determination coefficient R², mean absolute error (MAE) and mean absolute percentage error (MAPE) are used to assess the accuracy of the modelling. Results show that both approaches, in particular, the classical analytical models and the neural network, provide the high accuracy of approximation of the experimental data. At the same time, the neural network demonstrates an advantage in the region of high values of DK, where a nonlinear acceleration of crack growth is observed, reaching a coefficient of determination R² equal to 0.9994. The presented approach confirms the effectiveness of applying machine learning methods, in particular neural networks, in the problems of fracture mechanics and fatigue analysis.