Machine Learning Enhances Tribological Coating Performance with High Accuracy Predictions

By Soham NandiReviewed by Bethan DaviesJan 30 2025

A recent study published in Polymers explores the use of machine learning (ML) to assess the tribological properties of sericite/epoxy composite coatings (SEC).

Researchers tested six different ML algorithms to analyze how filler content and mesh size influence friction and wear under varying loads. They found that gradient boosting regression (GBR) provided the most accurate and stable predictions, with 93.7 % accuracy for friction coefficient and 85.7 % for wear rate. Their findings offer valuable insights for optimizing material performance in engineering applications.

Background

Friction and wear are inevitable in industries like machinery, energy, and transportation, leading to reduced efficiency, increased energy use, and higher maintenance costs. To address these challenges, researchers have developed coatings with enhanced tribological properties. Traditional methods for evaluating these coatings, such as wear and tensile tests, can be time-consuming, destructive, and resource-intensive. Machine learning presents a promising alternative by enabling efficient, non-destructive assessments.

Previous studies have applied ML to predict tribological performance based on factors like load, temperature, and sliding velocity. While these efforts have advanced understanding, they have largely overlooked the role of inherent filler properties, an aspect this study aims to address. This study has attempted to fill that gap by evaluating six ML algorithms, including GBR and random forest (RF), to determine the best predictive model and identify key factors influencing performance.

Experimental Methodology

The researchers developed and tested SEC formulations to enhance tribological performance. Epoxy resin was combined with sericite powder of varying mesh sizes (325, 500, 800) and weight percentages. After thorough mixing and vacuum treatment to remove air bubbles, the mixture was sprayed onto steel templates and cured at room temperature for 48 hours. The final coating thickness measured approximately 35 micrometers (µm).

To assess tribological properties, they conducted tests using a high-speed reciprocating friction machine, with a GCr15 steel ball as the friction counterpart. This setup was chosen due to its ability to simulate real-world contact conditions, providing reliable and reproducible measurements of friction and wear behaviors. Tests were performed under loads of 5, 10, and 15 Newtons (N).

The researchers then analyzed surface morphology and wear characteristics using scanning electron microscopy (SEM) and white light interferometry, while elemental distribution was examined through energy-dispersive spectroscopy (EDS).

For ML-based predictions, six models—RF, GBR, Gaussian process regression (GPR), artificial neural networks (ANN), support vector regression (SVR), and extreme gradient boosting (XGB)—were trained on data from 45 friction tests. The dataset was normalized and split into 80 % for training and 20 % for testing. Researchers optimized model hyperparameters using grid search and cross-validation, evaluating performance based on R-squared values, mean squared error, and mean absolute error.

Results and Discussion

The study examined how sericite content, load, and mesh size affected the coefficient of friction (COF) and wear rate. EDS and SEM analyses confirmed a uniform distribution of sericite powder in the coatings. Results showed that at 20 weight percentage (wt%) sericite, the COF and wear rate peaked due to adhesive wear mechanisms. However, at 30 wt%, wear debris accumulation led to lower COF and wear rates. Feature importance analysis identified sericite content as the primary factor affecting friction and wear, with load and mesh size playing secondary roles.

Among the ML models tested, GBR provided the highest predictive accuracy, achieving R-squared values of 0.937 for COF and 0.857 for wear rate, with minimal errors. Its ability to iteratively optimize residuals and adjust parameters ensured a strong balance between model fit and complexity.

Building on these results, the findings suggest that GBR is a powerful tool for predicting tribological behavior, helping reduce the time and cost associated with traditional testing methods. By providing reliable insights into how filler content and testing conditions influence SEC performance, this model supports the development of advanced tribological coatings for engineering applications.

Conclusion

This study demonstrated the effectiveness of ML in predicting the tribological properties of SEC. Among six tested models, GBR emerged as the most accurate, achieving 93.7 % accuracy for friction coefficient and 85.7 % for wear rate. The results highlighted sericite content as the most influential factor in tribological performance. By offering a cost-effective and nondestructive alternative to traditional testing, GBR can help optimize SEC formulations, ultimately improving the durability and efficiency of engineering coatings.

Journal Reference

Yan et al., 2025. Machine Learning-Based Prediction of Tribological Properties of Epoxy Composite Coating. Polymers, 17(3), 282. DOI:10.3390/polym17030282 ‌https://www.mdpi.com/2073-4360/17/3/28

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.