Given the numerous parameters influencing the tribological behavior of brake linings, including material properties and braking conditions, this study aims to identify the parameters with the most significant impact on this behavior. This information will allow for the selection of representative test conditions and a better understanding of the resulting tribological responses. Three gradient boosting models (CatBoost, LightGBM, and XGBoost) were used to predict the responses in terms of friction and wear. The SHAP algorithm was then applied to interpret the models and determine the parameters most influential on the coefficient of friction and wear. An experimental study was conducted on an industrial brake lining to validate the machine learning predictions. Among the models, CatBoost achieved the best prediction accuracy compared to experimental measurements, as confirmed by regression metrics. The SHAP analysis revealed that sliding speed is the predominant factor influencing wear, while material properties primarily determine the coefficient of friction. Experimental results confirm these observations, showing that an increase in sliding speed has a limited effect on the coefficient of friction and a significant effect on wear.
KunaroopNRimdusitSMoraP, et al.Car-bonized hemp hurd powder for eco-friendly polybenzoxazine composite brake material: excellent friction property and high mechanical performance. Arab J Chem2024; 17: 105769. https://www.sciencedirect.com/science/article/pii/S1878535224001710.
5.
ShindeDBulsaraMPatilJ. Wear analysis of eco-friendly non-asbestos friction-lining material applied in an automotive drum brake: experimental and finite-element analysis. Proc Inst Mech Eng, J: J Eng Tribol2021; 236: 552–562.
6.
IlieFCristescuAC. Tribological behavior of friction materials of a disk-brake pad braking system affected by structural changes-A review. Materials (Basel). 2022; 15: 4745. PMID: 35888213; PMCID: PMC9323244.
7.
BaskarPShuklaK. Tribological and performance assessment of two wheeler brake pads using dynamometer testing. Sci Rep2025; 15: 33653.
8.
BenmedakheneTOuelaaNFissahB, et al.Investigation on the tribological properties of an organic matrix composite material designed for brake applications. Tribol Ind2025; 47: 100–111.
9.
Laguna-CamachoJRJuárez-MoralesGCalderón-RamónC, et al.A study of the wear mechanisms of disk and shoe brake pads. Eng Fail Anal, 2015; 56: 348–359, ISSN 1350-6307,.
10.
ShindeDBulsaraMMistryK. Tribological performance of non-asbestos brake friction material in contact with brake disc of varying topography. Ind Lubric Tribol2020; 72: 1277–1283.
11.
PrajapatiPMThakorSJainP, et al.Investigation of dielectric studies of paracetamol-diethylamine solutions: experimental and machine learning approach. Spectrochim Acta, Part A, 2026; 345: 126767, ISSN 1386-1425,.
RathoreMSSinghAPMadhaniV, et al.Influence of KMnO4 concentration on the physical and dielectric properties of potassium ion conducting PVdF-HFP polymer electrolytes. J Electron Mater2025; 54: 8664–8678.
TouzaniSGrandersonJFernandesS. Gradient boosting machine for modeling the energy consumption of commercial buildings. Energy Build2018; 158: 1533–1543.
26.
ChamundeswariGDornalaRRReddyMCK, et al.Machine learning-driven design of broadband, polarization-insensitive terahertz metamaterial absorber. J Electron Mater2025; 55: 1989–2000. doi: 10.1007/s11664-025-12587-y
27.
Shuqin ZhangXSWangZ. A study on the interpretability of network attack prediction models based on light gradient boosting machine (lgbm) and Shapley additive explanations (shap). Comp Mater Cont2025; 83: 5781–5809.
28.
AlizamirMWangMIkramRMA, et al.Quamaproperties in waste foundry sand-based eco-friendly concrete. Res Eng2025; 25: 104307.
29.
AmerMSajjadUHamidK, et al.Reliable aprediction of solar photovoltaic power and module efficiency using Bayesian surrogate assisted explainable data-driven model. Res Eng2024; 24: 103226.
30.
BakloutiMCristolADesplanquesY, et al.Impact of the glass fibers addition on tribological behavior and braking perfor-mances of organic matrix composites for brake lining. Wear2015; 330-331: 507–514, 20th International Conference on Wear of Materials.
31.
SellamiA. Dataset for the prediction of coefficient of friction and wear [Data set]. Zenodo. 2025, 10.5281/zenodo.18047461
32.
EzugwuAEGreeffJHoY-S. A comprehensive study of ground-breaking machine learning research: analyzing highly cited and im-pactful publications across six decades. J Eng Res2025; 13: 371–383.
ThakkarKSinghVSharmaP, et al.Bifunctional phyto-synthesized nano silver for mitigating salinity-induced dormancy and associated fungal infections during seed germination in Brassica Juncea with integration of machine learning-based predictive modeling. Appl Biochem Biotechnol2025; 197: 5956–5991.
ZhangLJánošíkD. Enhanced short-term load forecasting with hybrid machine learning models: catboost and xgboost approaches. Expert Syst Appl2024; 241: 122686.
38.
HussainSMustafaMWAl-ShqeeratKHA, et al.A novel feature-engineered–ngboost machine-learning framework for fraud detection in electric power consumption data. Sensors2021; 21: 8423–8446. doi: 10.3390/s21248423
