Complete Stribeck curve prediction by applying machine learning to acoustic emission data from a lubricated sliding contact

Song Yang, Robert Gutierrez, Thomas Kirkby, Hafedh Bouassida, Marc Hilbert, Min Yu, Tom Reddyhoff,
Complete Stribeck curve prediction by applying machine learning to acoustic emission data from a lubricated sliding contact,
Mechanical Systems and Signal Processing,
Volume 241,
2025,
113544,
ISSN 0888-3270,
https://doi.org/10.1016/j.ymssp.2025.113544.
(https://www.sciencedirect.com/science/article/pii/S0888327025012452)
Abstract: The Stribeck curve for a sliding contact characterizes its lubrication behavior and enables friction to be predicted across various regimes. It is crucial for optimizing the energy efficiency and durability of mechanical systems. However, current methods of obtaining Stribeck curves are confined to laboratory settings and often involve destructive testing, which precludes real-time monitoring. This study shows how to predict the Stribeck curve for a reciprocating, sliding contact between a steel ball and flat by applying machine learning (ML) models to recorded acoustic emission (AE) signals. These models were trained and validated against experimental Stribeck curves obtained by directly measuring the coefficient of friction while varying the Stribeck number through temperature-induced viscosity adjustments. AE signals were captured at a high sampling rate and processed using Short-Time Fourier Transform (STFT) and Short-Time Histogram (STHG) techniques to compress the data and extract relevant features. Machine learning models, consisting of artificial neural networks (ANNs) optimized with Particle Swarm Optimization (PSO) or Genetic Algorithm (GA), were utilized to predict both the Stribeck Number and the Coefficient of Friction. The STFT＋GA-ANN combination provided the highest accuracy in predicting the Stribeck Number (MSE = 2.60, MAPE = 28.84%), while the STFT＋PSO-ANN combination excelled in predicting the Coefficient of Friction (MSE = 0.0001, MAPE = 4.94%). Combining these methods enables the complete Stribeck curve to be successfully predicted based only on the emitted sound. This demonstration of in-situ, non-destructive AE-ML-based prediction of comprehensive tribological behavior paves the way for real-time lubrication monitoring in machine applications without disrupting system operation.
Keywords: High-frequency reciprocating rig; Stribeck curve; Friction; Acoustic emission; Machine learning