Adam Obrusnik, Krystof Mrozek, Martin Kubecka, Kristina Tomankova, Petr Zikan, PlasmaSolve s.r.o., Brno, Czech Republic
Process modelling and the concept of digital twin tools is recently gaining traction in the coating community. The main driver behind this are long turn times and low predictability, especially when transferring processes between coaters or upscaling the process. Very recently, process modelling also starts to play a role in reducing the energy requirements of coaters.
In previous contributions, we have presented a Coater Digital Twin model implemented within the MatSight simulation suite developed by PlasmaSolve. The model was able to accurately predict coating composition as well as performance properties (hardness and residual stress) based on only 8 characterized samples as a training dataset. This was enabled by developing a hybrid process model which combines physics simulation, chemistry simulation and machine learning algorithms. The model is aware of the physical constraints relevant to the coater, which drastically reduces the number of training experiments required.
In this contribution, we take one step further and solve the inverse task – we use the Digital Twin model to design a coating recipe based on material specification. The material specification includes especially the number of layers in the stack, thickness of each sub-layer, composition of each sub-layer and the desired hardness/residual stress of the entire stack. The Digital Twin model produces a recipe file that can directly be loaded to the coater and the model predictions can be verified. We also demonstrate how the Digital Twin model can be interfaced with other PVD simulation tools to get a full picture of the process. Specifically, we show how the MatSight Coating Uniformity App can provide insight into the spatial distribution of coating thickness and spatial variations in ion bombardment and coating composition on a 3D part.