Developing a Prediction model for wear and friction of Mg-based alloys using machine learning algorithms

Developing a Prediction model for wear and friction of Mg-based alloys using machine learning algorithms


Developing a Prediction model for wear and friction of Mg-based alloys using machine learning algorithms

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: Developing a Prediction model for wear and friction of Mg-based alloys using machine learning algorithms

ارائه دهنده: Provider: Negar Bagherieh

اساتید راهنما: Supervisors: Dr. Meisam Nouri

اساتید مشاور: Advisory Professors: Dr. Moslem Noori

اساتید ممتحن یا داور: Examining professors or referees: Dr. Hasan Elmkhah,Dr. Mohsen Sheykhi

زمان و تاریخ ارائه: Time and date of presentation: 21/2/2023, am 8:30

مکان ارائه: Place of presentation: Amphitheater

چکیده: Abstract: Data-driven method, including machine learning (ML), have created a novel approach in instating correlation between tribological properties and other properties of engineering materials. In the present study, ML algorithms including Artificial Neural Networks (ANNs), Random Forest (RF), Gradient Boosting Machine (GBM), Adaptive Boosting (adaboost), Decision Tree (DT), Support Vector Regressor (SVR), and k-Nearest Neighbors algorithm (KNN), by predict wear volume loss and coefficient of friction (COF) of magnesium alloys. The obtained data was extracted from the studies where in steel pins, balls, or disks were used as the counterpart. The dataset contained wear volume, COF, mechanical properties (hardness, tensile yield strength, ultimate tensile strength, ductility, and elastic modulus), chemical composition, processing procedure, heat treatment and tribological test variables (temperature, sliding speed, sliding distance, and normal load). Several steps were taken to build models for predicting the tribological properties of magnesium alloys such as pre-processing, modeling and evaluation. In the data pre-processing step, missing and outlier data were identified and managed, and qualitative data were coded. Dividing the data into training, evaluation and test datasets and normalizing the quantitative were performed. The resulting structured data were used for modeling. Machine learning models were built and validated by the training set. Then the performance of the models on new data was evaluated using the test set by the regression metrics including mean absolute error, mean squared error, root mean squared error and R squared. The developed model can predict the tribological behavior of magnesium alloys by using material properties and test. The evaluation results of the algorithms indicate the high performance of GBM with an accuracy of 87% (R squared value = 0.87) in predicting of the COF using mechanical properties and tribological test variables. Adaboost algorithm can predict COF using the processing procedure, heat treatment, alloy chemical composition and tribology test variables with 86% accuracy (R squared value = 0.86). Examining the performance evaluation results of the algorithms showing the ability of the RF algorithm in predicting the wear volume loss using mechanical properties and tribological test conditions, with the accuracy of 92% (R squared value =0.92). The best algorithm to predict wear volume loss using processing procedure, heat treatment, alloy chemical composition and test variables is adaboost meticulous, with accuracy of 94% (R squared value =0.94). A comparative analysis between parameters related to alloying elements, manufacturing process, heat treatment, mechanical properties and tribological test variables was performed using RF algorithm. The results show the importance of normal load, elastic modulus, ductility, sliding distance and content of Zn in predicting COF and the major contribution of normal load, sliding distance, sliding speed, hardness and elongation in predicting the wear volume loss

فایل: ّFile: