EXP 4 AM Digital tool for Microstructure

Deposits of M4 and TA6V by laser cladding to develop a validated modeling via finite element simulations linked or not to Machine Learning in order to predict the microstructure.

We would like to have deposits of M4 and TA6V by the laser cladding process to develop an efficient and validated modeling via 2D finite element simulations. This FE model will be exploited to generate a database. This database will interact with “surrogate models” in a machine learning machine approach with the ultimate aim of predicting microstructures. My previous work has illustrated the difficulty of having efficient finite element simulations to predict microstructures in additive processes and more particularly in Laser Cladding.

The need for a precise thermal history: maximum temperature, rate of heating and cooling, numbers of melting

of the material ... in short a complete history of temperature is imperative. This means a FE model with a refined mesh to simulate the strong temperature gradients around the melt pool and to monitor the position of the laser at any time and not modeling by layers. This approach leads to very long simulations. For massive parts, a 2D approach is however possible. It will be affected by the out of plane heat leak. To compensate for this leak, the use of virtual "iddle times" and laser power but nevertheless in connection with physics is necessary. The required experiments (different geometries, different materials) will allow us to develop a method of calibrating these parameters. However, there are many other concerns: the thermo-physical properties are to be measured on the material deposit because the microstructure plays an important role on these properties. The microstructures are strongly out of equilibrium, the kinetic models of phase transformation are to be revisited. In this context, exploring the possibilities of Machine Learning isanoption that should not be overlooked. This is the path we want to explore.