An Experimental and Numerical Assessment of Nonreactive Precombustion Hydrogen–Natural Gas Mixing Dynamics in Gaseous Fuel Delivery Systems

Abstract

As hydrogen (H2) continues to gain traction globally as a potential low-carbon energy carrier, gas turbines (GTs) must accommodate an increasingly wide range of H2 and natural gas (NG) blend compositions and flow rates, raising concerns about blend uniformity and consistency. This work addresses these concerns with an experimental and computational study of nonreactive precombustion H2–NG mixing in a GT-relevant T-junction fuel supply line geometry. Experimental spatially-resolved species concentration measurements were obtained using an equal-area traversing method in two sampling planes located at x = 60.4 mm and x = 86.9 mm, exactly one pipe diameter apart, downstream of the junction for a single 40 vol% H2 / 60 vol% NG blend. Numerical results were recorded continuously along the vertical and horizontal traverses within the same downstream plane locations. The spatial coefficient of variance (CoVϕH2 ) was used to assess blend uniformity within each sampling plane. Experimental results showed a CoVϕH2 of 0.119 and 0.0521 for the first and second plane, respectively, indicating a 56 % increase in mixedness over one diameter. In contrast, numerical results based on the steady state Reynolds-averaged Navier–Stokes (RANS) using standard k–ε and SST k–ω closure models dramatically underpredicted the mixing behavior. The k–ε model yielded CoVϕH2 values of 0.469 and 0.326, corresponding to CoVϕH2 discrepancies of 294 % and 526 % relative to the experiment, yielding a 46 % disparity in the relative mixing rate between planes. The SST k–ω model performed even less favorably. Overall, the results demonstrate that, for the present geometry and flow conditions, the RANS models systematically and significantly underestimate mixing and fail to capture the detailed species transport behavior observed experimentally. These findings dispel concerns about incomplete fuel line mixing raised by prior unvalidated computational studies and provide benchmark experimental results to aid future model development.