Physics-guided fuel-switching neural networks for stable combustion of low calorific industrial gas.

Dual fuel combustion of low calorific industrial gas has been widely used in various types of burners for industrial production needs. Fuel switching rule is purely empirical lacking theoretic basis, potentially leading to flame extinction or excessive waste of ignition fuel. This study focuses on exploring physics-guided fuel-switching strategy for stable combustion of coke oven gas (COG) and blast furnace gas (BFG) in representative swirl burner. Stable operating regimes for COG and BFG are firstly identified with chemical eigen-analysis. The intrinsic propensity of flame extinction during fuel switching is revealed by positive chemical eigenvalues, and the key species and elementary reactions are identified. A clustered neural network with flag controller (CNNF) is then constructed to achieve stable combustion during fuel switching from COG to BFG as early as possible to reduce the use of COG. Results show that COG must be used to ignite and increase the ambient temperature to at least 673 K to avoid flame extinction. The chemical eigen-guided CNNF controller can adjust the switch speed to advance the fuel switching time by 14 % and can be further expanded to other fuel switching applications. • Fuel switching properties are analyzed with PSR model. • Eigen-analysis is used to find key species and reactions. • An active neural network controller is constructed. [ABSTRACT FROM AUTHOR]