A Multi-Technique Machine Learning Workflow for Optimizing the Manufacturing Process of Functional Layers in Electrochemical Energy Devices

Abstract In this presentation we will share our latest developments revolving around the utilization of Artificial Intelligence/Machine Learning techniques to optimize the manufacturing process of functional layers in lithium ion batteries (electrodes) and polymer electrolyte membrane fuel cells (gas diffusion layers -GDL-). We have used several machine learning techniques (including unsupervised, supervised and deep learning techniques) to predict the influence of manufacturing process parameters (e.g. slurry formulation, drying temperature, electrode calendering degree, GDL fiber diameter) on the properties of the functional layers (e.g. conductivity, tortuosity factor, porosity). We have also applied deep learning techniques in order to derive surrogate models mimicking the behavior of physics-based numerical models simulating the manufacturing process of the electrodes. The so-derived machine learning models are then used in combination with optimization algorithms in order to perform inverse design of the manufacturing processes, i.e. predicting which manufacturing parameters we need to adopt in order to maximize and/or minimize given properties. In this presentation, we review these developments, including the different workflows that we designed and discuss our manufacturing digitalization vision towards accelerated design of functional layers for electrochemical energy device applications. We believe that this approach gives the promise to reduce costs and save time in the industrialization of these devices towards a more sustainable world.