Optimizing adhesive rheology for stencil printing of fuel cell sealings using supervised machine learning

Abstract

In a single low-temperature proton exchange membrane fuel cell (LT-PEMFC) stack, the sealings are responsible for preventing leakages, providing electrical insulation and compensating assembly tolerances. This study investigates stencil printing to produce fuel cell sealings and aims to identify reliable conditions to print layer thicknesses of at least 500 μm, typically used for this application and substantially exceeding the current state-of-the-art. First, rheological and print experiments were conducted with 25 distinct adhesives, including ultraviolet (UV) curable acrylics and epoxies, to determine their printability. An adhesive was considered printable when it detached from the stencil without forming bubbles, a print defect that significantly diminishes the process stability. This had to be accomplished at two different squeegee speeds (40 and 160 mm/s), confirming the robustness of the adhesive for the print process. In addition, the impact of the aperture orientation, aperture aspect ratio (AR) and separation speed of the substrate from the stencil were evaluated. Here, it was detected that a multivariable criterion was necessary to reliably predict printability. Thus, classification models based on binary logistic regression, a simple supervised machine learning technique, were proposed using the thixotropy index TI (indicator for viscosity recovery ability), steady-state viscosity at 100 s−1, adhesive surface tension and aperture AR as predictors. The derived models achieved a prediction accuracy ≥88.2%, demonstrating adequate generalisation. Moreover, it was identified that increasing TI is the most effective approach to achieve printability. Finally, the driving physical mechanisms influencing the obtained print results were assessed, and it was shown that a TI value of about 50 Pa s/s is required to reach the minimum layer thickness.