Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 87 of the 87 references it cites were found in a registry or an index; 10 of the 18 sources its claims cite were checked for support.

Predicting EHL film thickness parameters by machine learning approaches

Open access

Abstract

Non-dimensional similarity groups and analytically solvable proximity equations can be used to estimate integral fluid film parameters of elastohydrodynamically lubricated (EHL) contacts. In this contribution, we demonstrate that machine learning (ML) and artificial intelligence (AI) approaches (support vector machines, Gaussian process regressions, and artificial neural networks) can predict relevant film parameters more efficiently and with higher accuracy and flexibility compared to sophisticated EHL simulations and analytically solvable proximity equations, respectively. For this purpose, we use data from EHL simulations based upon the full-system finite element (FE) solution and a Latin hypercube sampling. We verify that the original input data are required to train ML approaches to achieve coefficients of determination above 0.99. It is revealed that the architecture of artificial neural networks (neurons per layer and number of hidden layers) and activation functions influence the prediction accuracy. The impact of the number of training data is exemplified, and recommendations for a minimum database size are given. We ultimately demonstrate that artificial neural networks can predict the locally-resolved film thickness values over the contact domain 25-times faster than FE-based EHL simulations ( R 2 values above 0.999). We assume that this will boost the use of ML approaches to predict EHL parameters and traction losses in multibody system dynamics simulations.

Overview

CoChat Summary

TL;DR

Machine learning models trained on EHL simulation data predict central and minimum film thickness parameters more accurately than analytical proximity equations, with R² above 0.99 when using original EHL inputs. An ANN also predicts locally resolved film thickness over the contact 25 times faster than FE-based EHL simulation with R² above 0.999.

Key findings

  • With original EHL inputs, SVM and GPR predicted central and minimum film thicknesses with R² > 0.99; GPR was slightly superior to SVM. ANN configurations also reached R² > 0.99, with 12 neurons in two hidden layers performing best for the 3D circular point contact.
  • Using non-dimensional groups U, G, W or M, L instead of original inputs substantially reduced accuracy: R² fell well below 0.9 and almost to zero in some cases (e.g., GPR for 2D U, G, W gave h_c R² = 0.100 and h_min R² = 0.091; GPR for 3D U, G, W gave h_c R² = 0.013).
  • For the 2D line contact, 600 training data points were sufficient for 12 input parameters, giving R² = 1.000 for h_c and h_min; 300 and 100 points reduced accuracy but stayed above 0.9 (300: 0.961/0.963; 100: 0.922/0.927). The original 1,249-point set gave 0.988/0.988, which the authors suggest may imply some overfitting.
  • The ANN predicted the local lubricant film distribution over the contact length with an overall R² of 1.000 against testing data and was over 25 times faster than the FEM-based EHL simulation (19 s vs 0.74 s), with prediction coefficients above 0.999.

Methods

  • Data were generated from steady-state isothermal EHL simulations using a full-system finite element solution, with Latin hypercube sampling. The 2D infinite line and 3D circular point contacts were sampled with 1,500 and 1,000 data sets, respectively; 1,499 and 764 converged data sets were ultimately obtained.
  • SVM, GPR, and multi-hidden-layer feedforward backpropagation ANNs were implemented in MATLAB. Data were normalized to [−1, 1], split into 85% training and 15% testing, and evaluated by R². ANN architecture varied 10–20 neurons per layer and zero to three hidden layers; activation functions tested were hyperbolic tangent, sigmoid, and identity.
  • Predictions were compared with analytically solvable proximity equations from Dowson/Toyoda/Higginson for 2D line contacts and Hamrock/Dowson for 3D circular point contacts. For local film thickness prediction, position along the contact length was added as an input, using 86 uniformly distributed points and 128,914 total data sets.

Limitations

  • Stated: the trained ML approaches were not physics-informed and cannot provide physical explanations; predictions must be considered carefully because negative film thicknesses are possible. The authors suggest future use of a strictly positive linear function and output normalization to avoid unphysical results.
  • Stated: predictions were meaningful only for operating conditions covered by the training data. Future work should consider elliptical contacts, thermal effects, other rheological models, time-dependent squeeze, starvation effects, surface roughness, and solid asperity contact.
  • Stated and evident: the proximity equations have a limited validity range, and the design of experiments slightly exceeded it; thermal effects, limited oil supply, and shear-thinning fluid behavior were not considered in those correlations. Evident from the text: the ML models were trained on isothermal EHL simulation data and were not experimentally validated.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s40544-022-0641-6

Cites

87 of 87 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Schematic of an EHL infinite 2D line contact with relevant input variables. Reproduced with permission from Ref. [4], © The authors 2020.
Fig. 1 Schematic of an EHL infinite 2D line contact with relevant input variables. Reproduced with permission from Ref. [4], © The authors 2020.From the PDF
Image 2
From the PDF
Fig. 3 Flowchart of the ML methodology proposed in the present contribution.
Fig. 3 Flowchart of the ML methodology proposed in the present contribution.From the PDF
Fig. 4 (a) Schematic of the numerical solution of the EHL model and computational domains with meshing and boundary conditions for (b) infinite 2D line and (c) 3D circular point contacts.
Fig. 4 (a) Schematic of the numerical solution of the EHL model and computational domains with meshing and boundary conditions for (b) infinite 2D line and (c) 3D circular point contacts.From the PDF
Image 5
From the PDF
Image 6
From the PDF
Fig. 6 Central and minimum film thickness predicted from the proximity equations (vertical axis) versus the calculated ones from EHL simulations (horizontal axis) for the (a, b) infinite 2D line and (c, d) 3D circular point contacts.
Fig. 6 Central and minimum film thickness predicted from the proximity equations (vertical axis) versus the calculated ones from EHL simulations (horizontal axis) for the (a, b) infinite 2D line and (c, d) 3D circular point contacts.From the PDF
Image 8
From the PDF
Fig. 7 Predicted versus calculated values (testing data) of the central and minimum film thicknesses for the infinite 2D line and 3D circular point contacts using (a-d, i-l, q-t) SVM and (e-h, m-p, u-x) GPR trained with (a-h) the original EHL inputs, (i-p) the non-dimensional parameters U , G , W , or (q-x) M , L .
Fig. 7 Predicted versus calculated values (testing data) of the central and minimum film thicknesses for the infinite 2D line and 3D circular point contacts using (a-d, i-l, q-t) SVM and (e-h, m-p, u-x) GPR trained with (a-h) the original EHL inputs, (i-p) the non-dimensional parameters U , G , W , or (q-x) M , L .From the PDF
Image 10
From the PDF
Image 11
From the PDF
Fig. 10 Comparison of calculated (blue) and predicted (orange) lubricant film distribution over the contact length for a representative infinite 2D line contact using FEM and an ANN with 12 neurons in each of the two hidden layers.
Fig. 10 Comparison of calculated (blue) and predicted (orange) lubricant film distribution over the contact length for a representative infinite 2D line contact using FEM and an ANN with 12 neurons in each of the two hidden layers.From the PDF
Image 13
From the PDF

Verification

Claims the paper makes

8 Valid · 0 Not valid · 5 Not checked · 63 Own results and interpretations

41 more claims

Create a free account to see all 76 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 16
  • It may state two separate claims as one. 17
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

3 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

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.

4 of 4

Citations

Cites

Show all 80 references

Research works printed but not linked: 10

  1. Moes H. Discussion on Paper D1 by D. Dowson. Proc Instn Mech Engrs 180: 244–245 (1966)
  2. Moes H. Communications. In Proc. of the Symposium on Elastohydrodynamic Lubrication, 1965: 244–245.
  3. Dowson D. Elastohydrodynamics. Proc Inst Mech Eng 182: 151–157 (1968)
  4. Moes H. Lubrication and Beyond—University of Twente Lecture Notes code 115531. Enschede, Netherlands: University of Twente, 2000.
  5. Wiśniewski M. Einfluß eines begrenzten Ölangebotes auf die elastohydrodynamische Schmierung von Zahnrädern. Tribologie und Schmierungstechnik 30: 270–277 (1983)
  6. Müller AC, Guido S. Introduction to machine learning with Python: A guide for data scientist. Beijing: O’Reilly, 2016.
  7. Gohar R. Elastohydrodynamics. Chichester: Halsted Press, 1988.
  8. Roelands C. Correlational aspects of the viscosity-temperature-pressure relationship of lubricating oils. Ph.D. Thesis. Delft University of Technology, 1966.
  9. Glantz SA, Slinker BK. Primer of Applied Regression and Analysis of Variance. New York, NY: McGraw-Hill, 1990.
  10. Montgomery D C, Runger G C. Applied Statistics and Probability for Engineers 6th Edition. Wiley, 2014.

Cited by

See all 76

Metadata

Authors
Marian, M., Mursak, J., Bartz, M., Profito, F. J., Rosenkranz, A., & Wartzack, S.
Affiliations
  • Pontificia Universidad Católica de Chile (CL)
  • Friedrich-Alexander-Universität Erlangen-Nürnberg (DE)
  • Universidade de São Paulo (BR)
  • University of Chile (CL)
Corresponding author
Marian, Max
Venue
Friction
Published
2022-06-12
Volume
11
Issue
6
Pages
992-1013
ISSN
  • 2223-7690 (ISSN-L)
  • 2223-7704
OpenAlex
W4282599518
Cited by
76
References
80
Citations by year
2026: 21 · 2025: 16 · 2024: 21 · 2023: 13 · 2022: 5
FWCI
5.19
Type
journal article
Language
en
Open access
Licence
Locations
  • Friction (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • Universidad de Chile (submittedVersion, cc-by-nc-nd)
Funders
  • Universidad de Chile (UM‐04/19)
  • Pontificia Universidad Católica de Chile
  • Agencia Nacional de Investigación y Desarrollo (11180121, EQM190057)
Topics
  • Physical Sciences › Engineering › Mechanical Engineering › Tribology and Lubrication Engineering
  • Physical Sciences › Engineering › Mechanical Engineering › Gear and Bearing Dynamics Analysis
  • Physical Sciences › Engineering › Mechanical Engineering › Lubricants and Their Additives
Keywords
elastohydrodynamic lubrication, film thickness prediction, machine learning, artificial neural networks, support vector machine, Gaussian process regression, Latin hypercube sampling, training data size, activation functions, multibody dynamics, traction loss

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: Predicting EHL film thickness parameters by machine learning approaches. https://app.cochat.ai/papers/10.1007/s40544-022-0641-6
The paper
Marian, M., Mursak, J., Bartz, M., Profito, F. J., Rosenkranz, A., & Wartzack, S. (2023). Predicting EHL film thickness parameters by machine learning approaches. Friction. https://doi.org/10.1007/s40544-022-0641-6

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.