Autonomous, transferable, and mode-aware wear prediction from on-line wear-debris concentration monitoring of gearboxes

A 2015 full-life gearbox test coupled an on-line visual ferrograph (OLVF) debris-concentration stream to a physically motivated three-stage wear model, but the model was calibrated retrospectively: the transition point from steady-state to severe wear was picked by visually inspecting the same concentration curve it was compared against, and the sedimentation and stage-transition coefficients were fitted to a single test. The open problem is to turn this framework into an autonomous prediction method: detect or predict the onset of severe wear in the monitoring data stream without retrospective calibration, quantify how the model coefficients depend on gear material, surface roughness, lubricant, speed and load so the model transfers across designs and operating conditions, and incorporate wear-mode information from debris morphology, including the concurrent wear modes observed in practice. Later literature delivers data-driven wear-stage identification and on-line particle typing on bearings, four-ball rigs and aero-engine particle samples, but no method yet feeds an autonomously detected transition into the physical concentration model, characterizes the coefficients' dependence on design and operating variables, or integrates morphology-based mode discrimination with concentration-based prediction for a gearbox.

Background

Wear debris released into lubricating oil carries direct information about the wear of tribological components, so gearboxes can be monitored by analyzing particles in the oil instead of disassembling the transmission. Ferrography is a technique that uses a magnetic field to deposit ferrous wear particles onto a transparent substrate (a ferrogram) for imaging; an on-line visual ferrograph (OLVF) automates this by sampling oil at high frequency from a fixed position in the lubrication circuit and capturing a digital image of the deposited debris. From each image one computes the index of particle coverage area (IPCA), the total projected area of debris normalized by the image area, which serves as a proxy for the mass concentration of wear debris in the oil.

Classical wear modeling goes back to Kragelsky's formulation, in which the wear depth of a gear tooth flank grows in proportion to a wear coefficient $I_h$; with a constant $I_h$ and a sedimentation term removing debris from the oil at rate $f$ (the sedimentation coefficient), the predicted debris concentration either grows linearly or saturates, which can only describe mild, steady wear. Real gearboxes pass through distinct stages: a running-in stage with decaying wear rate as asperities smooth out, a steady-state stage, and a severe-wear stage preceding failure. Feng, Fan, Mao and Xie (Wear 336–337, 2015) extended the Kragelsky model with a time-varying wear coefficient, exponential decay during running-in and exponential growth during severe wear, obtaining closed-form three-stage solutions for the debris concentration $S$ governed by $dS/dt = m/V - fS$, where $m$ is the debris generation rate and $V$ the oil volume. In a 37.4 h full-life test of a back-to-back spur gearbox rig (a power-circulating test rig in which two meshing gear pairs load each other; 200 Nm, 1450 rpm, oil-bath lubrication) monitored by OLVF at 2-minute intervals, the measured IPCA curve matched the three-stage simulation, early-warning signs appeared before catastrophic failure at 35.8 h, and post-test microscopy showed fatigue pitting (surface-fatigue cracks spalling small pits out of the tooth flank) and adhesive scuffing (localized adhesive damage from momentary metal-to-metal contact) — two distinct wear modes — acting simultaneously on the tooth flanks.

The framework nevertheless operated retrospectively. The transition point from steady-state to severe wear (30 h in the simulation) was chosen by inspecting the observed IPCA curve, so no genuine prediction was demonstrated; the sedimentation coefficient $f$ and the stage-transition coefficients $b$ (running-in decay) and $c$ (severe-wear growth) were fitted to the single test, leaving their dependence on gear material, surface roughness, lubricant properties, rotational speed and applied load uncharacterized; and the model used only debris concentration, discarding the visual morphology information in the ferrograms that discriminates wear modes. An oil-leakage event during the test (10–16 h) showed that lubrication disturbances, not only tooth-surface deterioration, can drive the concentration signal — a confound any autonomous detector must survive. Parallel threads existed by 2015: dynamic wear-state identification from on-line debris image features (Wu et al., Mechanical Systems and Signal Processing 42, 2014, using IPCA and the equivalent diameter of large debris), and general mappings between wear rate and debris concentration in lubrication systems (Fan et al., Tribology International 109, 2017). The unresolved target is stated in the next section.

Problem Statement

Given an oil-lubricated gearbox monitored on-line by a debris-concentration stream (such as OLVF IPCA readings), develop and validate a wear-prediction method that extends the three-stage wear-debris concentration model of Feng et al. (2015) so that it becomes autonomous, transferable, and mode-aware. Specifically: (1) the method must detect or predict the transition from steady-state wear to severe wear from the monitoring data stream itself, in real time and without retrospective calibration of the transition point on the same curve it is evaluated against, and remain correct under lubrication disturbances such as oil loss and top-up; (2) it must render the model's fitted parameters — the sedimentation coefficient $f$ and the stage-transition coefficients $b$ and $c$ — transferable, so that the fitted parameters remain valid across gear material, surface roughness, lubricant properties, rotational speed and applied load, and are shown to hold on a gearbox design or operating condition not used for fitting; and (3) it must incorporate wear-mode information derived from debris morphology, including the simultaneous occurrence of wear modes such as fatigue pitting and adhesive scuffing within a single stage, into the concentration-based prediction of wear progression and remaining life. These three components jointly define one objective: predicting gearbox wear on-line, before failure, from debris monitoring alone, without per-machine retrospective calibration.

References

1. Song Feng, Bin Fan, Junhong Mao, Youbai Xie, 'Prediction on wear of a spur gearbox by on-line wear debris concentration monitoring', Wear 336–337 (2015) 1–8, DOI 10.1016/j.wear.2015.04.007, https://doi.org/10.1016/j.wear.2015.04.007
2. Tonghai Wu, Yeping Peng, Hongkun Wu, Xiaogang Zhang, Junqun Wang, 'Full-life dynamic identification of wear state based on on-line wear debris image features', Mechanical Systems and Signal Processing 42 (2014) 404–414, DOI 10.1016/j.ymssp.2013.08.032, https://doi.org/10.1016/j.ymssp.2013.08.032
3. Bin Fan, Bo Li, Song Feng, Junhong Mao, You-Bai Xie, 'Modeling and experimental investigations on the relationship between wear debris concentration and wear rate in lubrication systems', Tribology International 109 (2017) 114–123, DOI 10.1016/j.triboint.2016.12.015, https://doi.org/10.1016/j.triboint.2016.12.015
4. Yeping Peng, Tonghai Wu, Shuo Wang, Zhongxiao Peng, 'Wear state identification using dynamic features of wear debris for on-line purpose', Wear 376–377 (2017) 1885–1891, DOI 10.1016/j.wear.2017.01.012, https://doi.org/10.1016/j.wear.2017.01.012
5. Yeping Peng, Tonghai Wu, Guangzhong Cao, Sudan Huang, Hongkun Wu, Ngaiming Kwok, Zhongxiao Peng, 'A hybrid search-tree discriminant technique for multivariate wear debris classification', Wear 392–393 (2017) 152–158, DOI 10.1016/j.wear.2017.09.022, https://doi.org/10.1016/j.wear.2017.09.022
6. Wei Cao, Guangneng Dong, You-Bai Xie, Zhongxiao Peng, 'Prediction of wear trend of engines via on-line wear debris monitoring', Tribology International 120 (2018) 510–519, DOI 10.1016/j.triboint.2018.01.015, https://doi.org/10.1016/j.triboint.2018.01.015
7. Wei Cao, Han Zhang, Ning Wang, Hai Wen Wang, Zhong Xiao Peng, 'The gearbox wears state monitoring and evaluation based on on-line wear debris features', Wear 426–427 (2019) 1719–1728, DOI 10.1016/j.wear.2018.12.068, https://doi.org/10.1016/j.wear.2018.12.068
8. Yeping Peng, Junhao Cai, Tonghai Wu, Gugangzhong Cao, Ngaiming Kwok, Shengxi Zhou, Zhongxiao Peng, 'Online wear characterisation of rolling element bearing using wear particle morphological features', Wear 430–431 (2019) 369–375, DOI 10.1016/j.wear.2019.05.005, https://doi.org/10.1016/j.wear.2019.05.005
9. Renaldy Dwi Nugraha, Shi Chen, Nian Yin, Tonghai Wu, Zhinan Zhang, 'Running-in real-time wear generation under vary working condition based on Gaussian process regression approximation', Measurement 181 (2021) 109599, DOI 10.1016/j.measurement.2021.109599, https://doi.org/10.1016/j.measurement.2021.109599
10. Wei Cao, Zhao Han, Zhuang Zhuang Yang, Ning Wang, Jin Xiu Qu, Dong Wang, 'Deterioration state diagnosis and wear evolution evaluation of planetary gearbox using vibration and wear debris analysis', Measurement 193 (2022) 110978, DOI 10.1016/j.measurement.2022.110978, https://doi.org/10.1016/j.measurement.2022.110978
11. Rob Bosman, Dirk Jan Schipper, 'Transition from mild to severe wear including running in effects', Wear 270 (2011) 472–478, DOI 10.1016/j.wear.2010.12.009, https://doi.org/10.1016/j.wear.2010.12.009
12. Yue Chen, Jian Shi, 'Modeling of Distorted Degradation Data Based on Oil Analysis', Applied Sciences 15 (2025) 6531, DOI 10.3390/app15126531, https://doi.org/10.3390/app15126531
13. Shuo Wang, Yishi Chang, Hui Wei, Miao Wan, Tonghai Wu, Ying Du, 'An integrated mechanism and data model for adaptive wear state diagnosis via moving wear particles', Wear 564–565 (2025) 205722, DOI 10.1016/j.wear.2024.205722, https://doi.org/10.1016/j.wear.2024.205722
14. Zihan Ye, Liming Wang, Peng Cui, Shenao Wang, Fangyi Li, Yanyan Nie, 'A method for identifying wear stages in mechanical equipment based on online monitoring data and wear mechanisms', Measurement 280 (2026) 121889, DOI 10.1016/j.measurement.2026.121889, https://doi.org/10.1016/j.measurement.2026.121889
15. Shuo Wang, Linyu Xia, Yaguo Lei, Yishi Chang, Tonghai Wu, Tingting Li, Xudong Zhang, Ke Feng, 'Intelligent wear state diagnosis: Wear rate and mechanism integrated model based on multi-dimensional particle informatics', Tribology International 224 (2026) 112293, DOI 10.1016/j.triboint.2026.112293, https://doi.org/10.1016/j.triboint.2026.112293