Metalik malzemelerin sürtünme ve aşınma özelliklerinin makine öğrenmesi yöntemleri ile tahmin edilmesi

---

Thesis Type: Postgraduate

Institution Of The Thesis: Ataturk University, Fen Bilimleri Enstitüsü, Makine Mühendisliği Anabilim Dalı, Turkey

Approval Date: 2023

Thesis Language: Turkish

Student: ŞENAY ER

Supervisor: Halim Kovacı

Open Archive Collection: AVESIS Open Access Collection

Abstract:

The aim of this study is to predict the friction and wear behavior of metallic materials under variable experimental conditions (slip environment, time, speed and load) using machine learning techniques, one of the artificial intelligence applications. It is thought that the results obtained can enable more sensitive and effective decisions to be made, especially in engineering applications, such as material selection, design optimization, reducing the number of experiments/tests and maintenance strategies. In this study, samples of AISI 4140 low-alloy steel, which is widely used in many different industries, especially in the automotive industry, were subjected to wear tests under different test conditions by changing the parameters that affect the friction and wear properties of the material. The experiments were performed under different sliding environments (dry, salt water and oil), times (30, 60 and 120 min), speeds (8, 16 and 32 mm/s) and loads (1.3 and 5 N). The data set was split into 70% of the data obtained from the variable experimental conditions and 30% of the test, and these data were used in training and testing machine learning models. Six different supervised machine learning regression models were used to estimate the coefficient of friction and wear rate. At the same time, the performances of each model were evaluated by using the k-fold cross-validation method on the data set, taking into account the error metrics of each model. The results obtained were compared to the performance criteria of the models and it was determined which model achieved the best result. As a result of the analysis, the highest R2 value for the friction coefficient estimates in the Support Vector Regression (SVR) model was 0.80. This result shows that the SVR model fits the coefficient of friction estimates better than the other models. In addition, the highest R2 value for wear rate estimates, 0.95, was obtained in the linear regression model. This result shows that the linear regression model has greater success in estimating wear rate than other models. This shows that different models may perform better for different predictions, and different results may be obtained depending on the data set and variables used. The SVR model performed better on datasets with nonlinear relationships, while the linear regression model performed better on variables with linear relationships. The statistical performance measures obtained from the models used in this study demonstrate that the developed machine learning models can predict the coefficient of friction (CoF) and wear rate of AISI 4140 steel at a satisfactory level when the material and tribological variables are taken into account. It is believed that the results of the study can contribute to tribological applications related to friction and wear behavior under different load, speed and environmental conditions by using more training datasets in engineering applications.