Probing Ink Evaporation-Deposition Behavior for Scalable Catalyst-Layer Fabrication in PEM Electrolysis: Systematic...

Nonuniform catalyst-ink deposition during droplet drying, such as coffee-ring formation, remains a key bottleneck for scalable fabrication of proton exchange membrane (PEM) membrane-electrode assemblies (MEAs), as it compromises catalyst-layer uniformity, pore structure, and mass transport in PEM electrolysis. This work integrates systematic single-droplet metrology with explainable machine learning to quantify the evaporation-deposition behavior of Ir-based catalyst inks, including IrO₂ and ATO-supported formulations with PFSA ionomers. Quantitative morphology descriptors were established by combining optical microscopy with 3d-profilometry-assisted interpretation to characterize thickness nonuniformity and radial deposition distribution, including normalized variance, thickness-partitioned area fractions, and an edge-to-center grayscale ratio (ECCR). A data set spanning ionomer equivalent weight (EW = 720-1100), solid content (SC = 1-4 wt %), ionomer-to-catalyst mass ratio (I/Ir = 0.1-0.4), isopropanol volume fraction (ϕ_IPA), and substrate temperature (T_sub = 50-90 °C) was used to train a multioutput gradient-boosting decision tree model. Together with SHAP analysis, the model captures trend-level relationships between formulation/process parameters and morphology metrics (training R² = 0.74-0.88) and identifies ϕ_IPA × EW and ϕ_IPA × T_sub as dominant interaction terms governing ECCR and deposition-pattern transitions, whereas coverage and thick-region fractions are more sensitive to I/Ir and ϕ_IPA × SC. Based on these descriptors, an empirical deposition uniformity index (DUI) was further defined to rank relative deposition uniformity and guide process-window prediction. At EW = 1100, the model predicts a robust core evaporation window of ϕ_IPA = 0.568 ± 0.013 and T_sub = 60 ± 4.4 °C. Under T_sub = 60 °C, a favorable formulation region is further identified in the SC-I/Ir plane, corresponding to SC = 2.56-3.34 wt % and I/Ir = 0.26-0.34. Independent validation experiments confirmed that samples within the predicted window generally exhibited more uniform deposition and more stable electrochemical responses than out-of-window controls. In contrast, varying relative humidity from 30% to 80% RH and droplet volume from 0.04 to 0.4 μL caused only limited changes in the final deposition pattern. Overall, this work presents an integrated framework that combines droplet-scale morphology quantification, explainable learning, and process-window prediction, providing practical guidance for catalyst-ink formulation and scalable PEM catalyst-layer fabrication.