The DOI matches its Crossref record; 43 of the 44 references it cites were found in a registry or an index; 6 of the 25 sources its claims cite were checked for support.
Application of Tribological Artificial Neural Networks in Machine Elements
Open accessAbstract
Traditionally, analytical equations used in tribo-dynamic modelling, such as those used for predicting central film thickness within elastohydrodynamic lubricated contacts, have led to timely computations, but tend to lack the accuracy of numerical solvers. However, it can be shown that data-driven solutions, such as machine learning, can significantly improve computational efficiency of tribo-dynamic simulations of machine elements without comprising accuracy relative to the numerical solution. During this study, artificial neural networks (ANNs) are trained using data produced via numerical solutions, which are constrained by the regimes of lubrication to ensure the quality of the training data set. Multiple ANNs are then implemented to predict EHL central film thickness, as well as viscous and boundary friction, in multiple commonly used machine elements, such as a rolling element bearing and a spur gear. The viscous and boundary friction ANN prediction are compared directly against ball-on-disc experimental measurements to validate its accuracy.
Overview
CoChat Summary
TL;DR
The paper trains artificial neural networks on numerical EHL solutions to predict central film thickness, viscous friction, and boundary friction in a cylindrical roller bearing and spur gear, with friction validated against ball-on-disc MTM experiments. The ANNs closely match numerical results and are about 1500× faster for film thickness and about 1300× faster for friction, though extrapolation beyond training ranges remains unresolved.
Key findings
- Film-thickness ANN matched the numerical solution with reported MSE of 1.46E-8 for the bearing and 2.46E-6 for the spur gear, versus analytical MSE of 1.32E-3 and 1.87E-2; the ANN was ~1500× faster than the numerical solution, while the analytical equation was ~75× faster than the ANN but less accurate.
- Separate viscous and boundary friction ANNs achieved R² values of 0.9985 and 0.9993; combined predictions were compared against MTM ball-on-disc tests at 75 N, 50% SRR, and entrainment speeds from 0.6 to 2.9 m/s, with both models overpredicting friction at higher entrainment speeds, possibly due to shear thinning.
- Training databases used 600–5000 points; average numerical data-point generation took 5.88 s, giving single-core database wall times of 0.98–8.16 h; 1000–2000 points typically sufficed, optimum hidden layers were 2–3, and ReLU performed worse than tanh or logistic sigmoid.
- The spur gear friction ANN was ~1300× faster than the numerical solution and captured the minimum friction coefficient near the pitch point.
Methods
- Numerical EHL line-contact models (Reynolds equation, Roelands viscosity, Dowson-Higginson density, Greenwood-Tripp asperity friction) generated training data; Latin Hypercube Sampling was constrained by the Greenwood regime chart to steel-steel contacts in the Piezo-Rigid and Piezo-Elastic regimes and high-speed ranges.
- ANNs were developed in MATLAB 2022a with min-max normalization, 70/15/15 training/validation/test splits, MSE and R² evaluation, early stopping, and regularization; a sensitivity study tested over 500 structures with 1–4 hidden layers, 10–20 neurons, and tanh, logistic sigmoid, or ReLU activation.
- Case studies used a cylindrical roller bearing (12 rollers, 2500 N radial force, 10,000 rpm) and a spur gear (50 teeth, 200 Nm torque, 2000 rpm); friction validation used an MTM ball-on-disc tribometer with Group III base oil and AISI 52100 steel.
Limitations
- Stated: ANNs lack physical understanding; extrapolation beyond the training input range must be addressed for wider MBD adoption; training data were constrained to steel-steel contacts, PE/PR regimes, and high-speed ranges; experimental data are expensive and time-consuming to generate.
- Stated: friction validation used a fixed 50% slide-to-roll ratio, one load, five entrainment speeds, and one lubricant/contact pair; overprediction at higher speeds may be due to shear thinning.
- Evident from the text: no extrapolation performance is demonstrated, and the case studies are limited to two machine elements.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 8 Not valid · 13 Not checked · 66 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Previous implementations of ANNs in tribology and film thickness prediction have been limited to between one or three hidden layers [39].”
What the sources say
- Disputedcites [39] Marian, M., Tremmel, S.: Current trends and applications of machine learning in tribology—a review. Lubricants 9(9), 86 (Sep.2021). https://doi.org/10.3390/lubricants9090086 sourceExperimental — accuracy not yet measured
Section 3.1.1 (page unavailable) reports a tribological wear-prediction study that tested ANNs with one to four hidden layers, contrary to the claimed three-layer limit. The passages found do not establish a corresponding limit for film-thickness prediction. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The optimum number of layers was between two and three similar to tribological applications of ANNs [39].”
What the sources say
- Partly supportedcites [39] Marian, M., Tremmel, S.: Current trends and applications of machine learning in tribology—a review. Lubricants 9(9), 86 (Sep.2021). https://doi.org/10.3390/lubricants9090086 sourceExperimental — accuracy not yet measured
Section 3.4.1 (Coatings; page unavailable) reports a suitable two-hidden-layer ANN, and section 3.1.2 (Thermoplastic Matrix Composites; page unavailable) describes an ANN with three hidden layers for one output. The review does not identify two to three hidden layers as a general optimum; section 3.2.4 (Brakes and Clutches; page unavailable) also reports best results from a single-hidden-layer ANN. (LLM, conf=0.83)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Marian et al . used Latin Hypercube Sampling (LHS) for EHL film prediction as an effective method to generate a training database to cover a vast design space with many factors [29].”
What the sources say
- Supportedcites [29] Marian, M., Mursak, J., Bartz, M., Profito, F.J., Rosenkranz, A., Wartzack, S.: Predicting EHL film thickness parameters by machine learning approaches. Friction (2022). https://doi.org/10.1007/s40544-022-0641-6 source
Section 2.2 (Design of experiments) describes LHS as suitable for filling a factor space evenly with relatively few simulations despite many factors. It then says the study used that design to sample EHL parameter ranges and generate data sets; Section 2.4 (Machine learning) confirms those data sets were used for training film-parameter prediction models. (LLM, conf=0.94)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Interestingly in the study by Marian et al . [29], grouped non-dimensionalised parameters, denoted in Eq. 2 of the current study, were used as inputs and it was found to significantly impeded the ANNs ability to predict the central film thickness”
What the sources say
- Supportedcites [29] Marian, M., Mursak, J., Bartz, M., Profito, F.J., Rosenkranz, A., Wartzack, S.: Predicting EHL film thickness parameters by machine learning approaches. Friction (2022). https://doi.org/10.1007/s40544-022-0641-6 source
Section 3.2.2.1 (p. 14) states that training the ANN with dimensionless groups rather than the original inputs considerably reduced prediction quality. The section discusses prediction of central and minimum film thicknesses, so the cited claim about central film thickness is supported. (LLM, conf=0.93)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“One previously developed EHL film thickness database constructed via a Finite Element Method (FEM) solver covered a vast range of lubricants and material properties for relatively low speed conditions ( < 0.4 m ∕ s of 2D line contacts) [29].”
What the sources say
- Partly supportedcites [29] Marian, M., Mursak, J., Bartz, M., Profito, F.J., Rosenkranz, A., Wartzack, S.: Predicting EHL film thickness parameters by machine learning approaches. Friction (2022). https://doi.org/10.1007/s40544-022-0641-6 sourceExperimental — accuracy not yet measured
Section 2.2 (pp. 6–7) describes a simulation-generated database for 2D line contacts spanning specified material and lubricant parameters; Section 2.3 (p. 8) says the simulations used FEM. The example 2D cases in Section 3.2.2.1 (p. 14) have speeds below 0.4 m/s, but the retrieved passages do not establish that 0.4 m/s was the database’s upper speed limit. (LLM, conf=0.78)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
36 more claims
Create a free account to see all 92 claims and every source verdict.
- It may state two separate claims as one. 19
- Probably not a substantial claim. 36
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6.3 Case Study: EHL Film Prediction for Bearings and Gears · p. 12
“The computation time per data point was also documented for each of the case studies in Table 12”
§ 6.1 ANN Structure Performance · p. 12
“Table 12 Training times of ANN structures using 5000 data points, a LogSig activation function 1-4 hidden layers and 10-20 neurons within each layer (Color table online)”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6.1 ANN Structure Performance · p. 12
“Figure 8a-d and Tables 9, 10, 11, and 12 demonstrate the influence of increasing the data points from 600 to 1000, 2000, and 5000 data points for a logistic sigmoid function.”
§ 6.1 ANN Structure Performance · p. 12
“Fig. 8 R 2 performance of ANN structure 10 -[ 14 -14 -14 ] 3 -1 using Hyperbolic Tangent activation function a 600 points, b 1000, c 2000, and d 5000”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6.1 ANN Structure Performance · p. 10
“Values for R 2 using 600 data points and different activation functions, number of layers, and neurons are presented in Tables 6, 7, 8, 9, 10, 11, and 12.”
§ 6.1 ANN Structure Performance · p. 11
“Table 9 Training times of ANN structures using 600 data points, a LogSig activation function 1 -4 hidden layers, and 10 -20 neurons within each layer (Color table online)”
Citations
Cites
The Contact of Two Nominally Flat Rough Surfaces
Greenwood, J. A., & Tripp, J. H. (1970) · Proceedings of the Institution of Mechanical Engineers · doi:10.1243/pime_proc_1970_185_069_02 · Cited by 1,754 · cited 2×
“These equations provide quick estimates of certain key parameters, such as central [5] and minimum film thickness [6] and viscous [7] and boundary friction [8].”
BackgroundMethodBarbedo, J. G. A. (2018) · Computers and Electronics in Agriculture · doi:10.1016/j.compag.2018.08.013 · Cited by 750 · cited 1×
“To develop more complex ANNs and achieve greater accuracy, it is necessary to have a sufficiently large dataset [25, 26].”
Holmberg, K., & Erdemir, A. (2019) · Tribology International · doi:10.1016/j.triboint.2019.03.024 · Cited by 677
A numerical solution to the elasto-hydrodynamic problem
Dowson, D., & Higginson, G. R. (1959) · Journal of Mechanical Engineering Science · doi:10.1243/jmes_jour_1959_001_004_02 · Cited by 663
Elasto-Hydrodynamic Lubrication
(1977) · Elsevier eBooks · doi:10.1016/c2013-0-05764-7 · Cited by 591 · cited 1×
“Lubricant density changes with pressure due to the compressibility of the lubricant, hence the Dowson–Higginson model [33]:”
MethodCorrelational aspects of the viscosity-temperature-pressure relationship of lubricating oils
Roelands, C. J. A. (1966) · Research Repository (Delft University of Technology) · Cited by 565
Asiltürk, İ., & Çunkaş, M. (2011) · Expert Systems with Applications · doi:10.1016/j.eswa.2010.11.041 · Cited by 391 · cited 2×
“Additional work has been completed to model wear in manufacturing processes [22] including specific applications like the estimation of flank wear in drilling [23] and modelling of the surface roughness of machined parts [24].”
BackgroundGohar, R., & Rahnejat, H. (2018) · doi:10.1142/q0152 · Cited by 251
The Rheological Properties of Elastohydrodynamic Lubricants
Evans, C. R., & Johnson, K. L. (1986) · Proceedings of the Institution of Mechanical Engineers, Part C: Mechanical Engineering Science · doi:10.1243/pime_proc_1986_200_134_02 · Cited by 243 · cited 1×
“These equations provide quick estimates of certain key parameters, such as central [5] and minimum film thickness [6] and viscous [7] and boundary friction [8].”
BackgroundCurrent Trends and Applications of Machine Learning in Tribology—A Review
Marian, M., & Tremmel, S. (2021) · Lubricants · doi:10.3390/lubricants9090086 · Cited by 238 · cited 2×
“The optimum number of layers was between two and three similar to tribological applications of ANNs [39].”
BackgroundMethod
Research works printed but not linked: 6
- ESDU, “Film thicknesses in lubricated Hertzian contacts (EHL). Part 1: two-dimensional contacts (line contacts),” Eng. Data Sheets, vol. 85027, no. October 1985.
- Dowson, D., Toyoda, S.: A central film thickness formula for elastohydrodynamic line contacts. In: Proceedings of the Society of Photo Optical Instrumentation Engineering, pp. 60–65 (1979)
- Ripa, M., Frangu, L.: A survey of artificial neural networks applications in wear and manufacturing processes. J. Tribol. 8, 35–42 (2004)
- Barbedo, J.G.A., Linjordet, T., Balog, K.: Impact of training dataset size on neural answer selection models BT—advances in information retrieval. Comput. Electron. Agric. 153(March), 828–835 (2019). https://doi.org/10.1016/j.compag.2018.08.013 · DOI · Paper page
- Gohar, R.: Elastohydrodynamics. Imperial College Press, London (1988)
- Weber, C., Banaschek, K., Niemann, G.: Formänderung und Profilrücknahme bei-gerad und schrägverzahnten Rädern. F Vieweg, Paris (1955)
Cited by
AI for tribology: Present and future
Yin, N., Yang, P., Liu, S., Pan, S., & Zhang, Z. (2024) · Friction · doi:10.1007/s40544-024-0879-2 · Cited by 109
“Among them, bearings are one of the most complex and widely used basic components in tribology, and AI technology is also the most widely used in bearing research [298, 299].”
Physics-Informed Machine Learning—An Emerging Trend in Tribology
Marian, M., & Tremmel, S. (2023) · Lubricants · doi:10.3390/lubricants11110463 · Cited by 88
Singh, A., Wolf, M., Jacobs, G., & König, F. (2023) · Tribology International · doi:10.1016/j.triboint.2022.108166 · Cited by 38
Shah, R., Jaramillo, R., Thomas, G., Rayhan, M. T., Hossain, N., Kchaou, M., Profito, F. J., & Rosenkranz, A. (2025) · Advanced Engineering Materials · doi:10.1002/adem.202401944 · Cited by 48
Machine Learning for Film Thickness Prediction in Elastohydrodynamic Lubricated Elliptical Contacts
Issa, J., El Hajj, A., Vergne, P., & Habchi, W. (2023) · Lubricants · doi:10.3390/lubricants11120497 · Cited by 30
“More recently, Walker et al. [35] worked on predicting the central film thickness and viscous and boundary friction in EHD line contacts using ANNs.”
Method
Metadata
- Authors
- Walker, J., Questa, H., Raman, A., Ahmed, M., Mohammadpour, M., Bewsher, S. R., & Offner, G.
- Affiliations
- Loughborough University (GB)
- Anstalt für Verbrennungskraftmaschinen List (Austria) (AT)
- Corresponding author
- Walker, J.
- Venue
- Tribology Letters
- Published
- 2022-11-23
- Volume
- 71
- Issue
- 1
- ISSN
- 1023-8883 (ISSN-L)
- 1573-2711
- OpenAlex
- W4309925999
- Cited by
- 39
- References
- 40
- Citations by year
- 2026: 7 · 2025: 11 · 2024: 12 · 2023: 8 · 2022: 1
- FWCI
- 3.02
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Tribology Letters (publishedVersion, cc-by)
- Loughborough University Research Repository (Loughborough University) (submittedVersion, cc-by)
- Funders
- AVL List
- Topics
- Physical Sciences › Engineering › Mechanical Engineering › Tribology and Lubrication Engineering
- Physical Sciences › Engineering › Mechanical Engineering › Lubricants and Their Additives
- Physical Sciences › Engineering › Mechanical Engineering › Gear and Bearing Dynamics Analysis
- Keywords
- artificial neural networks, elastohydrodynamic lubrication, machine elements, rolling bearings, spur gears, viscous friction, boundary friction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Application of Tribological Artificial Neural Networks in Machine Elements. https://app.cochat.ai/papers/10.1007/s11249-022-01673-5
- The paper
- Walker, J., Questa, H., Raman, A., Ahmed, M., Mohammadpour, M., Bewsher, S. R., & Offner, G. (2023). Application of Tribological Artificial Neural Networks in Machine Elements. Tribology Letters. https://doi.org/10.1007/s11249-022-01673-5
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.