top of page


Mr. John Smith

Job title



Research and/or engineering questions/objective : The presence of surface defects (roughness, surface defects, profiles, etc.) within a contact undeniably leads to a modification of its local properties such as the coefficient of friction. On the operations of wheel chock on railway axles, this surface condition is essential because it conditions the good design in fatigue during its final use. Nevertheless, these local phenomena are not well understood and require a real step back. The objective is then to propose a multi-scale numerical strategy to better understand the phenomena. Methodology : The multi-scale strategy is divided into two steps. First, in order to model the surface defects, a DEM type modeling is used. In a second step, the results from the discrete element method (DEM) are injected in a finite element method calculation considering implicitly the defects. In view of the great diversity of surface defaults (size, dispersion, height, etc.) that can be obtained, a large number of simulations DEM is proposed coupled with a deep learning architecture. Finally, these results will be introduced into a finite element model simulating the assembly operation at a more local scale. Results : The work carried out has made it possible to set up, after multiple tests, a simulation with results comparable to those obtained in industry, and therefore usable. An automation of the creation of the domains, the calibration of the numerical parameters and the simulation was then set up for the generation of the database. A Deep Learning-type multi-input artificial intelligence model was also developed. This one manages to determine from the topology of two rough surfaces, the coefficient of friction resulting from their friction. By exploiting this AI, and by iterating over all the finite elements defining the contact surface of the wheel and the axle in the modeling of the wedging operation, to introduce a local friction coefficient, it is possible to numerically find results similar to those from experimental tests. Limitations of this study : Nowadays it is only possible with the calculation software used to model purely elastic domains or to represent perfect plasticity, which is not perfectly representative of reality. A criterion of rupture will then have to be defined to model a pseudo-plasticity. Moreover, modeling by discrete elements requires a large calculation time and a large calculation power. The use of artificial intelligence for the generalization of data will overcome this problem, but will only apply to our study. If the material and/or the geometry were to be modified, the creation of a new database would be necessary. Conclusion : Even if the approach here is applied to the assembly of a railway axle, it can be generalized to any type of application (braking etc.). The contribution of AI is undeniable because it allows almost instantly to obtain local information that is introduced into a structural calculation. The other advantage of AI is to be able to establish links between the most critical situations which will lead to a more relevant design of axles with respect to fatigue and, more precisely,fretting-fatigue.

Mr. Victor Lalleman, PhD Student, LaMcube; Dr. Vincent Magnier, Senior lecturer, Université de Lille; Dr. Pierre Gosselet, DR CNRS (research professor), CNRS; Dr. Cédric Hubert, Lecturer, UPHF; Ing. Stéphane Salengro, Head of Research and Development Department, MG-Valdunes

Multi-scale friction coefficient : from roughness to system computation using deep learning

EB2012-IBC-004 • Paper • EuroBrake 2012 • IBC


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page