Speaker
Description
In the automotive industry hydrodynamic lubrication is a concern in most of the machinery where there are two contacting surfaces in relative motion. The optimization of these surfaces through micro-texturing represents a very promising way to reduce friction, in order to achieve enhanced lifetimes and reduced lubricant use. However, it is very difficult to predict these optimal texturing patterns.
We present a very efficient neural network capable of predict the Stribeck curves based only on the micro-texturing pattern while also being able to reproduce the reverse process, in which it generates a set of micro-texturing patterns that reproduce the desired input Stribeck curve.
In order to train this machine learning application, we solved the Reynolds equation to obtain the pressure and cavitation profiles by using an implementation of the iterative inexact Newton implementation (INE) in a finite element method (FEM) framework. The solver code, written in MATLAB, was developed and validated in comparison to known cases in literature and tested with complex problems such as lubricated contacts with arbitrarily generated patterns of dimples/defects, thus enabling the efficient generation of a training data set.
Our approach will allow for fast and accurate predictions of the tribological behavior of micro-textured lubricated contacts, at a fraction of the computational cost required for a direct solution method.
