Machine Learning Model for Friction Prediction on Textured Surface

Abstract

Friction significantly affects the efficiency, wear and durability of mechanical systems. Surface texturing has been widely adopted to control frictional characteristics by modifying contact interactions. However predicting friction behaviour based on texture parameters remains challenging due to nonlinear dependencies among operating conditions, material properties and surface geometry. This paper presents a comprehensive study on the application of machine learning (ML) techniques for predicting friction on textured surfaces. Various ML models, including linear regression, support vector machines, decision trees and artificial neural networks are explored. The results demonstrate that ML models effectively capture complex relationships and provide accurate predictions, enabling optimization of surface textures for enhanced tribological performance.

Introduction

This text discusses the use of machine learning (ML) for predicting friction behavior in textured surfaces. Friction is a major cause of energy loss and wear in mechanical systems, making friction reduction important for improving efficiency and component lifespan. Surface texturing, such as adding micro-scale dimples or grooves, can reduce friction by improving lubrication, lowering contact area, and trapping wear particles. However, finding the optimal texture design is difficult because friction depends on many interacting factors, including load, speed, temperature, and lubrication conditions.

Recent research shows that ML techniques outperform traditional empirical and experimental methods for friction prediction. Various approaches have been explored, including:

• Regression models for predicting friction using texture parameters.
• Artificial Neural Networks (ANNs) for capturing complex nonlinear relationships.
• Support Vector Machines (SVMs) for handling smaller nonlinear datasets.
• Ensemble methods such as Random Forest and Gradient Boosting.
• Convolutional Neural Networks (CNNs) for analyzing surface texture images.

The study aims to develop ML models that accurately predict the Coefficient of Friction (COF) using surface texture characteristics and operating conditions. Input variables include texture diameter, depth, spacing, area density, load, sliding velocity, temperature, and lubrication condition.

The methodology involves:

1. Data Collection from experiments, simulations, or published datasets.
2. Data Preprocessing, including cleaning, normalization, feature selection, and train-test splitting.
3. Model Development using Linear Regression, SVM, Decision Trees, Random Forest, Gradient Boosting, ANN, and CNN.
4. Model Training through supervised learning and optimization techniques such as gradient descent.
5. Performance Evaluation using metrics like Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R².

Results show that:

• ANNs achieve the highest prediction accuracy because they effectively model nonlinear relationships.
• Random Forest and Gradient Boosting provide reliable and interpretable predictions.
• SVMs perform well on smaller datasets but may face scalability challenges.
• Important factors affecting friction include normal load, surface roughness, texture geometry, sliding speed, and lubrication conditions.

The proposed ML framework has applications in automotive components, bearings, gear systems, biomedical implants, MEMS, robotics, and automation systems. Its main advantages are improved prediction accuracy, reduced experimental costs, faster design optimization, and scalability. Challenges include the need for high-quality data, risk of overfitting, model interpretability, and ensuring generalization across different materials and operating conditions.

Conclusion

This paper presents a comprehensive study on machine learning based friction prediction for textured surfaces. The findings demonstrate that ML models, particularly neural networks, significantly outperform traditional approaches in capturing complex relationships. The integration of ML into tribological design processes can lead to improved efficiency, reduced wear and optimized surface performance. Future Roadmap: Although the present work successfully establishes a machine learning framework for predicting the coefficient of friction on textured surfaces, several extensions can be pursued to further enhance the robustness, applicability and industrial relevance of the proposed approach. Expansion of Experimental Dataset: Future work will focus on generating a larger and more diverse experimental dataset by including additional material pairs such as ceramics, composites and coated surfaces. Extending the operating range of load, speed, temperature and lubrication conditions will improve model generalization and reliability across real world applications. Incorporation of Transient and Dynamic Friction Behaviour: The current study primarily addresses steady state friction. Future research can include transient friction phenomena such as running in behaviour, stick slip motion and wear evolution over time. Time series machine learning models (e.g., LSTM or GRU networks) can be explored for this purpose. Integration of Real Time , Low Cost Sensor Data: The framework can be extended by incorporating real time sensor inputs such as vibration, acoustic emission or MEMS- based force sensors. This will enable online friction prediction and condition monitoring, making the system suitable for practical industrial deployment. Advanced Image-Based and Multi-Modal Learning: Future work may involve deeper convolutional neural networks trained on high resolution surface topography, SEM images or 3D surface scans. Combining image based features with numerical texture parameters in a multi modal learning frame- work is expected to further improve prediction accuracy. Physics Informed and Hybrid ML Models: The hybridization of machine learning models with established tribological theories (contact mechanics, lubrication regimes and Stribeck behaviour) can be strengthened. Physics informed neural networks (PINNs) can help enforce physical consistency and reduce data dependency. Uncertainty Quantification and Model Explainability: Future studies can incorporate uncertainty quantification techniques such as Bayesian machine learning to assess confidence in predictions. Explainable AI methods can also be used to better understand the influence of surface texture parameters on friction behaviour. Optimization and Automated Surface Design: The trained ML models can be integrated with optimization algorithms (genetic algorithms, particle swarm optimization) to automatically suggest optimal surface texture designs for minimum friction.

References

[1] T. V. Reddy and S. N. Babu, “Open dataset of textured surface tribology experiments from an Indian lab” Data in Brief (India), 2021. [2] S. K. Singh and A. Chattopadhyay, “Machine learning based prediction of friction behaviour on laser textured surfaces” Wear, vol. 460-461, pp. 203 212, 2020. [3] M. Paggi, A. Kusche and K. Holmberg, “Deep neural network modelling for friction coefficient prediction of micro-textured interfaces,” Tribology International, vol. 153, p. 106593, 2021. [4] S. R. Patil and P. V. Deshmukh, “Experimental investigation and ANN prediction of friction on laser textured steel surfaces,” Proc. Nat. Conf. Tribology Surface Engineering (NCTSE), Pune, India,2019, pp. 34-40. [5] Singh, R. Verma and N. K. Borker, “Machine learning approach for coefficient of friction prediction of textured aluminium,” Indian J. Eng. Mater. Sci., vol. 27, no. 2, pp. 112-120, 2020. [6] M. R. Joshi and K. P. Kulkarni, “SVM based prediction of tribological performance for dimpled surfaces,” in Proc. IEEE India Council International Conf. (INDICON), Bengaluru, India, 2020, pp.215 220. [7] D. N. More and S. T. Kale, “Effect of surface texture parameters on friction: regression modelling and experiments” Defence Science Journal, vol. 71, no. 5, pp. 457- 465, 2021. [8] R. S. Bendre and P. R. Kale, “Random forest modelling for sliding friction prediction on textured polymer coatings,” in Proc. Int. Conf. Advances in Materials Manufacturing (ICAMM), Chennai, India, 2021, pp. 88–94. [9] Sharma and H. Gupta, “Neural-network modelling of friction in micro-patterned ceramic surfaces,” J. Institution of Engineers (India): Series C, vol. 102, no. 3, pp. 345-353, 2021. [10] Patankar, “Friction prediction on textured surfaces using ensemble ML methods,” M. Tech. thesis, Dept. of Mechanical Eng., IIT Bombay, India, 2020. [11] P. R. Rao and L. S. Menon, “Data-driven approach to tribology: dataset and baseline models from Indian experimental studies,” Proc. Nat. Conf. on Mechanical Engineering Research, Kolkata, India, 2019, pp. 51–58. [12] N. K. Sharma and M. T. Iqbal, “Comparison of SVM and ANN for friction coefficient estimation of rough bearings,” J. Tribology (India), vol. 5, no. 1, pp. 23 31, 2020. [13] R. K. Mishra and A. B. Desai, “Optimization of surface texture parameters using GA and ANN for low friction” in Proc. IEEE India Int. Conf. on Smart Systems and Inventive Technology (ICSSIT), 2022, pp. 301-307. [14] L. Chen, Y. Wang and H. Zhao, “Effect of surface texture parameters on friction: a regression and neural-network analysis,” Surface and Coatings Technology, vol. 399, pp. 126210–126220, 2020. [15] T. Sharma and R. K. Pandey, “Experimental study and ANN modelling of friction on micro surfaces” J. Tribology, vol. 143, no. 8, pp. 081701-081710, 2021.

Copyright

Copyright © 2026 Dr. N. N. Gotkhindikar, Atharva Dhanegave, Janhavi Nikam, Prathamesh Karwa. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.