Sirvi Autor "Tamuri, Kaarel" järgi

Nüüd näidatakse 1 - 1 1

Masinõppe meetodite võrdlus vardas esinevate pragude iseloomustamiseks
(Tartu Ülikool, 2025) Tamuri, Kaarel; Jaanuska, Ljubov, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Rods are common components in engineering, where the presence of cracks can significantly reduce load-bearing capacity and pose safety risks. This bachelor’s thesis evaluates the suitability of different machine learning methods for crack identification in rods. Three machine learning models, linear regression, random forest and XGBoost, are compared in their ability to predict crack depth and location using input features from three different domains: natural frequencies, 16 Haar wavelet coefficients and 32 Haar wavelet coefficients. The models are assessed using standard regression metrics and noise is introduced to simulate real-world measurement uncertainty. The results show that the Haar wavelet coefficients outperform natural frequencies, especially in predicting crack location. Among the models, XGBoost consistently delivers the highest accuracy, achieving R2 up to 0.896 in the combined prediction task using the dataset of 32 Haar wavelet coefficients. Random forest also performs well, while linear regression provides fast but less accurate results. The study concludes that XGBoost trained on 32 Haar wavelet coefficients is the most effective approach for crack identification in rods.