Generalizability and transferability of acoustic-emission wear-stage diagnostics across materials and operating conditions

Acoustic emission (AE) monitoring can distinguish the run-in, steady-wear, and wear-out stages of sliding systems through mappings between AE frequency bands and wear mechanisms and through cumulative-energy transition thresholds, but every published mapping and threshold — including the wavelet-based short-time-Fourier-transform and continuous-wavelet-transform (STFT+CWT) workflow and the cumulative-energy transition criterion reported for martensitic sliding plates in 2015 — was learned empirically for a single material, contact geometry, and operating condition. The field remains open: the 2020 review of AE–friction–wear relationships states that existing studies cover only isolated factors in specific contact conditions, and later work delivers per-system solutions (four-ball steel, bronze sliding bearings, aerospace liner composites, journal bearings, rotary seals, metal-matrix composites) with system-specific signatures and, at best, single-system detection statistics. The open problem is to determine whether such band-to-mechanism mappings and cumulative-energy thresholds are valid across materials, coatings, contact geometries, lubrication regimes, loads, and temperatures, and to construct transferable, statistically characterized AE wear-stage diagnostic models executable in real time — the prerequisite for certifiable industrial AE wear monitoring.

Background

Acoustic emission (AE) denotes transient elastic waves generated by the rapid release of energy from localized sources within a material, such as crack growth or asperity fracture; piezoelectric sensors coupled to a component record this activity continuously during operation, which makes AE attractive for on-line wear monitoring of sliding contacts that cannot be observed directly. The lifetime of a sliding component is conventionally divided into wear stages: run-in (rapid surface adaptation), steady or permanent wear (slow, near-linear material loss), and wear-out (accelerated damage before failure); detecting the transition out of steady wear is the central diagnostic target. Because individual AE events are short, non-stationary transients mostly in the 100 kHz–1 MHz range, time–frequency tools are standard. The short-time Fourier transform (STFT) computes a spectrum over a sliding window, here per sliding cycle, giving a per-cycle energy spectrum $E(\omega)=|X(\omega)|^{2}$ whose time integral — the cumulative AE energy — tracks damage accumulation. The continuous wavelet transform (CWT), here with the Morlet mother wavelet (a wave-shaped analysis function dilated to multiple scales), then resolves the temporal evolution of chosen frequency bands.

Earlier work established that AE frequency content carries mechanism information: Hase, Mishina and Wada (2012) associated adhesive wear with a high-frequency spectral peak near 1.1 MHz and abrasive wear with multiple peaks at 0.25–1 MHz in their metal systems. Baccar and Söffker (2015) built on this a model-free diagnostic workflow: AE recorded by a field-programmable-gate-array-based (FPGA, reconfigurable digital hardware) acquisition chain (piezoelectric disc sensor, 16-bit resolution, 4 MHz sampling) during a 20-day sliding-wear endurance test of martensitic (hardened-steel) plates was analyzed by cyclewise STFT energy accumulation plus bandwise Morlet-CWT analysis over the bands 100–300, 300–500, 500–700, and 700–900 kHz. They reported a reproducible mapping from frequency bands to wear stages (e.g., 100–300 kHz active mainly during run-in, 700–900 kHz appearing at wear-out onset), tentative associations of bands with damage mechanisms via post-test scanning electron microscopy (SEM; coating tensile cracks below about 500 kHz; abrasive and fatigue cracking at 300–900 kHz), and a cumulative-energy criterion — a normalized energy gradient of roughly 0.15 — marking the transition from run-in to permanent wear. The features and thresholds are, however, empirically learned for the tested material and condition; the paper implements no automated classifier, no detection-statistics evaluation, and no on-line decision pipeline, and the acquisition FPGA was used for data capture only.

The structural limitation is field-wide: the review of Feng, Borghesani, Smith, Randall and Peng (2020), which cites the Baccar and Söffker (2015) paper, surveys theoretical and experimental AE–friction–wear relationships and concludes that existing papers each treat only one or two factors in one specific contact condition, with no general framework linking tribological parameters (roughness, film thickness, friction coefficient), operating parameters (speed, load), and AE characteristics (amplitude, energy, frequency, count) across systems. Successor studies in 2018–2026 likewise each developed signatures for one tribosystem — AE spectral categorization for four-ball wear tests of bearing steel and for model sliding contacts (Rastegaev et al.), machine-learning wear-mode classification and long-short-term-memory (LSTM, a recurrent neural-network architecture for sequence data) wear monitoring for bronze sliding bearings (König et al.), AE monitoring of aerospace self-lubricating composite liners (Revill et al.), journal-bearing wear monitoring under varied operating conditions (Bote-Garcia et al.), rotary-seal wear prediction with Bayesian-optimized deep learning (Lu et al.), and data-driven AE damage analysis for a boron-carbide-reinforced aluminium-alloy (B4C/AA6061) composite (Kwon et al.) — so whether the published band-to-mechanism mappings and energy thresholds transfer across materials and conditions, and how to build transferable AE wear-stage models with quantified detection performance, has remained unresolved.

Problem Statement

Determine whether AE wear diagnostics derived for one sliding system — in particular the mappings between AE frequency bands and wear mechanisms and the cumulative-AE-energy criteria for wear-stage transitions, such as the run-in-to-permanent-wear transition criterion reported for the martensitic-plate system of Baccar and Söffker (2015) — remain valid across different materials, coating states, contact geometries, lubrication regimes, loads, and temperatures; and, on that basis, construct transferable AE wear-stage diagnostic models — for example physics-informed or normalized spectral features, or explicit calibration procedures — that classify the wear stage of sliding components with quantified sensitivity, specificity, false-alarm, and missed-detection rates, including confidence intervals on phase-transition markers, without extensive per-system retraining; and demonstrate that the resulting wear-stage decision logic executes on line under realistic computational constraints, with the 2015 per-cycle STFT plus bandwise Morlet-CWT workflow as the named baseline whose real-time capability was not demonstrated. These components are conjunctively required and serve one overarching objective — a validated, transferable AE-based wear-stage diagnostic methodology for sliding systems. For the transferability determination, both a demonstration that the mappings and thresholds transfer — most directly through cross-material, cross-condition experiments — and a rigorous characterization of the conditions and component classes under which they must be re-established count as answers; per-system demonstrations that neither test transfer nor quantify detection performance do not resolve the question. Because the per-system retraining budget and the computational envelope are relative to the target system, a complete answer states the calibration or retraining budget it requires and the hardware and latency class on which its wear-stage decision logic runs on line. A complete answer addresses all three components.

References

1. D. Baccar, D. Söffker, 'Wear detection by means of wavelet-based acoustic emission analysis', Mechanical Systems and Signal Processing 60-61:198-207 (2015). DOI: 10.1016/j.ymssp.2015.02.012, https://doi.org/10.1016/j.ymssp.2015.02.012
2. D. Baccar, D. Söffker, 'Wavelet-Based Online Evaluation and Classification of Wear State Using Acoustic Emission', Proc. ASME SMASIS 2012, Vol. 1, pp. 609-614 (2012). DOI: 10.1115/smasis2012-7916, https://doi.org/10.1115/smasis2012-7916
3. A. Hase, H. Mishina, M. Wada, 'Correlation between features of acoustic emission signals and mechanical wear mechanisms', Wear 292-293:144-150 (2012). DOI: 10.1016/j.wear.2012.05.019, https://doi.org/10.1016/j.wear.2012.05.019
4. D. Söffker, S. Rothe, 'New Approaches for Supervision of Systems with Sliding Wear: Fundamental Problems and Experimental Results Using Different Approaches', Applied Sciences 7(8):843 (2017). DOI: 10.3390/app7080843, https://doi.org/10.3390/app7080843
5. S. F. Wirtz, A. P. A. Cunha, M. Labusch, G. Marzun, S. Barcikowski, D. Söffker, 'Development of A Low-Cost FPGA-Based Measurement System for Real-Time Processing of Acoustic Emission Data: Proof of Concept Using Control of Pulsed Laser Ablation in Liquids', Sensors 18(6):1775 (2018). DOI: 10.3390/s18061775, https://doi.org/10.3390/s18061775
6. I. A. Rastegaev, D. L. Merson, A. V. Danyuk, M. A. Afanasyev, A. Vinogradov, 'Using acoustic emission signal categorization for reconstruction of wear development timeline in tribosystems: Case studies and application examples', Wear 410-411:83-92 (2018). DOI: 10.1016/j.wear.2018.06.006, https://doi.org/10.1016/j.wear.2018.06.006
7. P. Feng, P. Borghesani, W. A. Smith, R. B. Randall, Z. Peng, 'A Review on the Relationships Between Acoustic Emission, Friction and Wear in Mechanical Systems', Applied Mechanics Reviews 72 (2020, online 2019). DOI: 10.1115/1.4044799, https://doi.org/10.1115/1.4044799
8. I. Rastegaev, D. Merson, I. Rastegaeva, A. Vinogradov, 'A Time-Frequency Based Approach for Acoustic Emission Assessment of Sliding Wear', Lubricants 8(5):52 (2020). DOI: 10.3390/lubricants8050052, https://doi.org/10.3390/lubricants8050052
9. J.-L. Bote-Garcia, N. Mokhtari, C. Gühmann, 'Wear monitoring of journal bearings with acoustic emission under different operating conditions', PHM Society European Conference 5(1) (2020). DOI: 10.36001/phme.2020.v5i1.1202, https://doi.org/10.36001/phme.2020.v5i1.1202
10. F. König, C. Sous, A. Ouald Chaib, G. Jacobs, 'Machine learning based anomaly detection and classification of acoustic emission events for wear monitoring in sliding bearing systems', Tribology International 155:106811 (2021). DOI: 10.1016/j.triboint.2020.106811, https://doi.org/10.1016/j.triboint.2020.106811
11. F. König, J. Marheineke, G. Jacobs, C. Sous, M. Zuo, Z. Tian, 'Data-driven wear monitoring for sliding bearings using acoustic emission signals and long short-term memory neural networks', Wear 476:203616 (2021). DOI: 10.1016/j.wear.2021.203616, https://doi.org/10.1016/j.wear.2021.203616
12. P. Deshpande, V. Pandiyan, B. Meylan, K. Wasmer, 'Acoustic emission and machine learning based classification of wear generated using a pin-on-disc tribometer equipped with a digital holographic microscope', Wear 476:203622 (2021). DOI: 10.1016/j.wear.2021.203622, https://doi.org/10.1016/j.wear.2021.203622
13. P. Revill, A. Clarke, R. Pullin, G. Dennis, 'Acoustic emission monitoring of wear in aerospace self-lubricating bearing liner materials', Wear 486-487:204102 (2021). DOI: 10.1016/j.wear.2021.204102, https://doi.org/10.1016/j.wear.2021.204102
14. J. Prost, G. Boidi, M. Varga, G. Vorlaufer, S. J. Eder, 'Lifetime assessment of porous journal bearings using joint time-frequency analysis of real-time sensor data', Tribology International 169:107488 (2022). DOI: 10.1016/j.triboint.2022.107488, https://doi.org/10.1016/j.triboint.2022.107488
15. R. Gutierrez, T. Fang, R. Mainwaring, T. Reddyhoff, 'Predicting the coefficient of friction in a sliding contact by applying machine learning to acoustic emission data', Friction 12:1299-1321 (2024). DOI: 10.1007/s40544-023-0834-7, https://doi.org/10.1007/s40544-023-0834-7
16. V. Mathiyazhagan, A. Meena, 'Predictive modelling of tool wear in CFRP drilling using acoustic emission sensors under dry and cryogenic conditions', Wear 570:205930 (2025). DOI: 10.1016/j.wear.2025.205930, https://doi.org/10.1016/j.wear.2025.205930
17. F. Wirsing, F. König, G. Jacobs, 'Zustandsüberwachung und Verschleißprognose durch Kombination von Acoustic Emission und Machine Learning – Ein Ansatz zum proaktiven Lifecycle Management', Gleit- und Wälzlagerungen 2025, pp. 199-212 (2025). DOI: 10.51202/9783181024539-199, https://doi.org/10.51202/9783181024539-199
18. J. Lu, J. Lu, D. Li, S. Zheng, 'Prediction of Wear in Rotary Dynamic-Seal Friction Pairs Based on Acoustic Emission Technology', Wear 596:206729 (2026). DOI: 10.1016/j.wear.2026.206729, https://doi.org/10.1016/j.wear.2026.206729
19. J.-H. Kwon, H. W. Lee, S. Cho, D. Lee, A. R. Jo, S.-K. Hwang, H. Chae, W. Woo, D.-K. Kim, 'Data-driven acoustic emission framework with multimodal validation for resolving micromechanical damage in B4C/AA6061 composite', Composites Part B: Engineering 325:114010 (2026). DOI: 10.1016/j.compositesb.2026.114010, https://doi.org/10.1016/j.compositesb.2026.114010
20. Kanchan G. M., Gopalakrishna M. Kamath, Venkitanarayanan Parameswaran, Mohite P. M., 'Acoustic emission signal analysis using wavelet synchrosqueezing transform for incipient damage detection', Nondestructive Testing and Evaluation (2026, online). DOI: 10.1080/10589759.2026.2693062, https://doi.org/10.1080/10589759.2026.2693062