QUANTRI - Research project

An artificial intelligence-based strategy was developed to predict the local dry friction from the morphology of the third body. The work was carried out in six steps:

(1)    Generation of various third body function of contact conditions coupled with rheological measurements (friction, flows);
(2)    Imaging of surface morphology & third body particles, with Scanning Electron Microscopy;
(3)    Image processing with different strategies;
(4)    Image analysis to extract quantitative descriptors (characteristics of particles and surface features) consider the whole diversity of third body particles layout and features;
(5)    Building of a consistent database grouping all those descriptors issued from the different tests conducted under various conditions, and also having hidden metadata including experimental data such as temperature, atmosphere, relative humidity, speed, roughness...
(6)    Using this database to train a random forest algorithm to predict measured local friction



 

Dry friction between two bodies in contact involves a layer of particles at the sliding interface, namely the third body layer, which is fully responsible for energy dissipation. Until now, the rheology of the third body has been characterized qualitatively by a broad description of its morphology (from powdery granular media to ductile continuous media) at the contact scale. The present work proposes to enrich this characterization through advanced image processing.
First, third bodies are generated during pin-on-disk experiments. Then they are characterized by using image acquisition and analysis. Wear tracks are observed by scanning electron microscopy. The images of third body are dispatched in two classes, (1) ejected particles out of the contact, and (2) trapped particles assimilated to texture.

The diversity of third body structures and their complexity requires the search for relevant descriptors of both morphology and texture. The necessary tools have been built to study third body rheology and get quantitative results.
Once the images are properly segmented, it is possible to search for correlations between tribological data and characteristics of the third body (morphology & texture). For this it is planned to build a database grouping all the images (particles and textures) coming from different tests conducted under various conditions by attributing different features similar to those presented here, and also having hidden metadata including experimental data such as temperature, composition of the atmosphere, relative humidity, speed, roughness... Once the database is built, the use of unsupervised machine learning algorithms would allow to better understand the mechanisms involved in dry contacts.