Towards the identification of the link between the contact roughness and the friction-induced vibration: Use of deep learning

To date, there is no general criterion to determine in advance the friction-induced-vibrations such as squeal, chatter, etc. Indeed, these different noises depend on the conditions of contact between the moving parts (pad-disc, etc.). Indeed, it is well known that these different noises depend on the contact conditions of the moving parts (pad-disc, etc.). These contact conditions are by nature multi-scale and depend on geometries (roughness, contact plates, system, etc.), materials, and their evolution within the contact (wear, debris, etc.). From a predictive point of view, it is then necessary to establish complex modal analysis type computations for the squeal. This method has the drawback of being useable only for an established contact configuration and is therefore not adapted to the general case. This paper attempts to generalize the determination of frequencies leading to squeal for any surface. Thus to generalize, an artificial intelligence program has been developed to predict the state of the dynamical behavior and the potential squeal frequencies. The input are surface roughness data. Supervised learning starts from a multi-scale contact computation. The scheme has an accuracy rate of 99% for near-instantaneous response times. In the end, an inverse analysis will show the influential zones in obtaining these instabilities allowing in the longer term to establish a criterion. •A model was presented that is able to predict the behavior of the call system without the need to contact process modelization.•The presented method is very user-friendly and it does not require expertise in artificial intelligence or complex mechanism of tribology to use it in the industry.•The presented method is based on artificial intelligence and more precisely deep learning algorithms such as two-dimensional convolutional neural networks and supervised learning.