Sim-to-Real Transfer in Deep Learning for Agitation Evaluation of Biogas Power Plants

[摘要] Biogas is an important driver in carbon-neutral energy sources. Many biogas digester setups, however, are not well optimized and waste energy or fail to maximize their gas output potential. To optimize these systems, a framework was developed to measure and predict digester systems’ efficiencies by closely monitoring fluid movements. This framework includes a numerical calculation of fluid behavior (Computational Fluid Dynamics (CFD)), and Deep Learning to estimate the fluid shear-rates introduced by the agitator’s action. Additionally, a novel measurement system is presented that can measure the same metrics, as simulated, in real-world environments. Lastly, an outlook is given that presents the options and extensions of the presented setup to reduce prediction error, minimize measuring efforts further, and recommend optimization approaches to the operator.

[关键词] ANN;artificial neural networks;CNN;convolutional neural networks;deep learning;CFD;computational fluid dynamics;agitation performance prediction;shear-rate