Rolling bearings are vital components in rotating machinery, where their premature failure can cause costly downtime or safety-critical accidents. While conventional diagnosis mainly targets obvious local defects such as spalls or cracks, subtle degradations including clearance variation and lubrication deterioration often emerge earlier in the bearing’s life cycle. These early-stage changes, resulting from wear or lubricant ageing, modify contact mechanics, load distribution, and frictional behaviour, producing measurable vibration responses even before visible damage occurs. However, traditional models typically assume ideal rolling conditions and overlook tribological effects, making it difficult to interpret these signals. In practice, weak fault features are also masked by transmission path distortion and background noise, hindering timely detection. To address these challenges, this thesis develops an integrated framework combining tribodynamic modelling, experimental investigation, entropy-based indicators, and sparsity-enhancement methods. A multi-degree-of-freedom tribo-dynamic model is established to incorporate clearance variation, lubricant viscosity, roller–raceway traction, and roller–cage collisions. The model successfully simulates load-carrying characteristics, frictional forces, and skidding behaviour under various operating conditions. Extensive experiments on roller bearings with adjustable clearance and lubrication confirm that larger clearance intensifies impact vibrations and yields spectral components similar to outer-race fault frequencies, while lubricant viscosity influences roller slippage and traction. A new frequency-domain indicator based on entropy decomposition is then proposed to quantify the sparsity and energy distribution of characteristic frequency bands, allowing clear differentiation between clearance-induced responses and genuine faults, as well as between different types of faults. Finally, two sparsity-guided blind deconvolution methods are developed to enhance weak fault features by optimising entropy-based objectives. Both methods demonstrate superior capability in extracting fault-related impulses under varying clearance and lubrication conditions. In summary, this thesis contributes (i) a comprehensive tribo-dynamic model, (ii) systematic experimental validation under combined clearance and lubrication effects, (iii) an interpretable entropy-based health indicator, and (iv) robust blind deconvolution algorithms for sparsity enhancement. The proposed framework advances the understanding of bearing friction dynamics and provides effective tools for condition monitoring.
| Date of Award | 24 Mar 2026 |
|---|
| Original language | English |
|---|
| Sponsors | China Scholarship Council |
|---|
| Supervisor | Fengshou Gu (Main Supervisor) & Helen Miao (Co-Supervisor) |
|---|