In friction brakes, the contact interface is highly scalable with changing contact locations during braking. This evolution is crucial to the performance of the system such as particle emission, noise generation, squeal, etc. The prediction of the contact evolution is then a challenge that it is essential to take up to better understand the phenomena and thus to better control them. A state-of-the-art review considers some models exist to predict this evolution [1,2,3]. Nevertheless, these models are often not predictive with respect to the experimental truth due to the difficulty of contact access. In this paper, a model for predicting the localization is proposed with the experiment as a starting point. One signature of the contact state is the temperature field. To this end, a pin-on-disc test [4] was heavily instrumented with 6 thermocouples that were uniformly distributed in the brake lining near the surface. More precisely, a campaign of nearly 600 tests was carried out with variable input parameters (disc speed, braking time etc.). All of this data are used to establish a prediction model of the contact evolution regardless of the configuration, even if it does not exist in the initial database. In this work, we propose an artificial intelligence approach via a Deep Recurrent Convolutional Neural Network (DRCNN) architecture, and by considering a memory effect over a short period. At the end of this model and according to the first instants of contact (less than one second), a prediction of the evolution of the various thermocouples is proposed on a contact which can go beyond 30s. The methodology shows its effectiveness regarding configurations for tests that were not included in the learning base (20% of the 600 tests). Even very local phenomena on each thermocouple are predicted with results of less than 5% error. Based on this prediction of the contact location for all instants, another artificial intelligence scheme is proposed downstream to predict the dynamic behaviour on target squeal frequencies. Again, the results are convincing with a good correspondence of the model results with the experimental results. In conclusion, the proposed strategy shows an indisputable link between the localization that can now be predicted and the vibratory behaviour. Moreover, by inverse method, the most influential factors have been identified allowing to foresee the possibility of medium-term power for the industrialists to optimize the system by reworking the initial surface condition for example. Finally, the obtained results show, the contact mechanism can be clearly explained, and this problem can be solved. Also, noise pollution can be greatly reduced, and the environment condition can be improved. [1] Y. Waddad, V. Magnier, P. Dufrénoy, G. De Saxcé, A multiscale method for frictionless contact mechanics of rough surfaces, Tribology International, Volume 96, 2016, Pages 109-121, ISSN 0301-679X, DOI: https://doi.org/10.1016/j.triboint.2015.12.023. [2] M. Mueller, G.P. Ostermeyer, Cellular automata method for macroscopic surface and friction dynamics in brake systems, Tribology International, Volume 40, Issue 6, 2007, Pages 942-952, ISSN 0301-679X, DOI: https://doi.org/10.1016/j.triboint.2006.02.045. [3] M. Stender, M. Tiedemann, D. Spieler, D. Schoepflin, N. Hoffmann, and S. Oberst, “Deep learning for brake squeal: Vibration detection, characterization and prediction,” arXiv, 2020. [4] M. Duboc, V. Magnier, J-F. Brunel, P. Dufrénoy, Experimental set-up and the associated model for squeal analysis Mechanics & Industry, 21 2 (2020) 204, DOI:https://doi.org/10.1051/meca/2019083

University of Lille: Mr. Nikzad Motamedi, Mr. Vincent Magnier, Mr. Van-Vuong Lai, Dr. Jean-François Brunel, Dr. Philippe Dufrénoy, Mr. Hazem Wannous