Artificial Intelligence-Based Identification of Lubrication Regimes Using Multimodal Tribological Signals
International Conference on Tribology and Surface Science (TRiboSurf’26), Trabzon, Türkiye, 4 - 07 Kasım 2026, (Yayınlanmadı)
- Yayın Türü: Bildiri / Yayınlanmadı
- Basıldığı Şehir: Trabzon
- Basıldığı Ülke: Türkiye
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
Lubrication regime transitions strongly influence friction, wear, and the stability of tribological contacts. Detecting these transitions directly from measured signals is difficult because the relationship between frictional behaviour and the underlying contact state is nonlinear and may change with speed, load, lubricant condition, and surface characteristics. This study proposes an artificial intelligence framework for automatic identification of lubrication regimes using synchronized tribological signals. The framework considers measurable signals such as friction force, normal force, vibration, acoustic emission, and temperature as multidimensional indicators of contact behaviour. Signal processing is first used to extract physically interpretable time- and frequency-domain descriptors, including statistical moments, spectral energy, characteristic frequencies, and transient-event indicators. These descriptors are then supplied to supervised and unsupervised machine learning algorithms for classification of boundary, mixed, and hydrodynamic or elastohydrodynamic lubrication states, depending on the investigated contact system. A key aspect of the proposed approach is the use of temporal information and physically meaningful features rather than relying solely on raw signal values. Model performance is assessed using independent test data and confusion-matrix-based metrics, while feature-importance analysis is used to determine which measurements provide the strongest information about regime transitions. The resulting framework is intended to provide an objective and scalable alternative to manual interpretation of tribological signals. In addition to identifying the current lubrication state, the methodology can provide an early indication of transitions toward less stable regimes associated with increased friction or wear risk. The study demonstrates how artificial intelligence can be integrated with conventional tribological measurements to convert complex sensor data into practical information for monitoring and diagnosis of lubricated contacts.