Deep learning-driven rapid hydrophobicity prediction for high-efficiency fog collection on laser-textured brass surfaces

Water scarcity remains a critical global challenge, and fog collection has emerged as a promising solution to harness alternative freshwater sources. However, conventional fabrication of fog collection devices, typically relying on chemical methods, suffers from complex procedures and environmental risks. This study introduces a deep-learning strategy to predict the hydrophobic contact angle of laser-processed square-groove arrays on brass surfaces directly from microscopic images, aiming to optimize superhydrophobic/hydrophilic hybrid surfaces for efficient fog collection. By training a convolutional neural network (CNN) on 1000+ microstructure images with corresponding contact angle data, the method achieves 92.3% prediction accuracy and reduces surface functionalization time by 80% compared to conventional approaches. This enables real-time optimization of laser parameters (e.g., spacing, power) to tailor wettability for fog harvesting. Experiments show that brass surfaces fabricated via this method exhibit stable superhydrophobicity (water contact angle >150°) after heat treatment, eliminating the need for time-consuming posttreatment drying processes. The integration of deep learning with laser processing facilitates the construction of highperformance hybrid wettability surfaces, with potential to enhance fog collection efficiency - supporting sustainable freshwater generation in arid and water-scarce regions.