Search academic texts

Information × Registration Number 0224U031495, 0122U001858 , R & D reports Title Modelling of functional and structural properties of shape memory alloys by machine learning methods popup.stage_title Head Yasnii Oleh P., Доктор технічних наук Registration Date 02-05-2024 Organization Ternopil National Technical University named after Ivan Puluj popup.description2  Shape memory alloys (SMAs) retain their original shape under the influence of thermomechanical loading. SMAs are widely used due to their unique properties, for example, in structural elements, automotive engineering, aerospace engineering, microelectromechanical systems, etc. It is extremely important to study their structural properties, in particular, the dependence of stresses and strains on the number of loading cycles. Since experimental procedures are often quite expensive, it is advisable to use machine learning methods. Thus, it is important to model the dependence of stresses and strains on the number of load cycles for a shape memory nickel-titanium alloy utilizing machine learning methods.Dependences of the stress and strain ranges on the number of loading cycles for four samples taken from available literature data are predicted by machine learning methods in a specialized software environment for data analysis named Orange. In general, two models were built for each of four samples. The dependence of the corresponding physical quantity on the number of load cycles was applied to the input of each model. The number of load cycles was considered as an independent variable, and the corresponding physical quantity was interpreted as a dependent variable. To increase the accuracy of the simulation results, the data set was further increased by interpolating the original experimental data with a one-dimensional Akima spline.For each sample, the data set was divided into two unequal parts - the training sample and the test sample. Regression dependencies were built by the method of random forests, neural networks, gradient boosting, the method of support vectors, AdaBoost, and the method of k nearest neighbors. Each of the obtained models was additionally checked ten times by the method of cross-validation. Product Description popup.authors Baran Denys Ya. Boiko Andrii R. Havrysh Roman M. Homon Olha O. Marushchak Olena V. Chornomaz Nataliia Yu. popup.nrat_date 2024-05-02 Close