DOI: 10.29132/ijpas.1811247 ISSN: 2149-0910

Deep Learning-Based Prediction of Tribological Properties in Mg-Cu and Mg-Zn Alloys for Biomedical Applications

Rukiye Tekin Ünver, Cihan Bayraktar, Bilge Demir, Kürşat Mustafa Karaoğlan
Show PDF Cite
Incorporating Cu and Zn at low concentrations into a magnesium matrix provides antibacterial properties and biodegradability advantages for biomedical applications. However, the complex effects of these alloying additions on microstructural and tribological behaviours and their optimal ratios have not yet been fully elucidated. This study developed a deep learning-based regression model to predict the wear properties of Mg-Cu and Mg-Zn alloys. Experimental data were obtained from wear tests conducted under 5–20 N loads and a sliding distance of 250 m in dry and simulated body fluid environments. Microstructural parameters, including grain size, density, hardness, crystallite size, microstrain, and dislocation density, were used as model inputs. In contrast, volume loss, coefficient of friction, and specific wear rate were used as output variables. The developed Deep Multilayer Perceptron model with optimised hyperparameters demonstrated high predictive performance on both the training and test datasets, preventing overfitting. Model evaluation yielded R² values of 0.9912 for volume loss, 0.9935 for coefficient of friction, and 0.9545 for specific wear rate, with RMSE values of 0.58, 0.011, and 0.61, respectively, confirming model reliability. Error analyses proved a narrow distribution and high correlation between predicted and experimental data. Sensitivity analysis results indicated that load, environment, hardness, and grain size are the most critical factors determining wear behaviour. This work presents a machine learning framework for tribological prediction in biodegradable Mg alloys, which may contribute to alloy design and optimisation of ortho-paedic implant materials.

More from our Archive