Real-time sludge moisture monitoring via jet imaging and deep learning

Waste activated sludge from wastewater treatment plants poses a major environmental challenge, with its high moisture content complicating disposal and resource recovery processes across global industries. Efficient sludge management requires precise moisture monitoring to optimize treatment methods, reduce costs, and enhance outcomes such as anaerobic digestion and composting. Traditional approaches for moisture measurement are time-intensive and batch-based, while emerging techniques, such as infrared or nuclear magnetic resonance methods, suffer from inaccuracies, high costs, or limitations in real-time applications. Here we show that sludge jet characteristics, reflecting its non-Newtonian fluid properties, can be captured via high-speed imaging and analyzed with deep learning to accurately predict moisture content within 20 s. By developing a laboratory-scale system of instantaneous capturing of activated sludge jet expansion images (iCASJEI), we acquired over 11,000 jet images across 79–94 % moisture ranges and trained convolutional neural networks, with VGG-16 outperforming AlexNet and LeNet under optimized conditions (0.2 MPa pressure, 4 mm aperture), achieving 93.5 % validation accuracy at 2 % precision and 87.6 % at 1 % precision. These findings show that incorporating iCASJEI to extract non-Newtonian fluid characteristics from sludge jets with deep learning algorithms can substantially reduce testing time for sludge moisture content. This approach could also be applicable to other sectors where non-Newtonian fluid characteristics enable real-time moisture detection in viscous liquids.