A newly developed bidirectional ultrasonic sensing method allows simultaneous, colocated measurement of dry-gas hydrogen concentration and recirculation flow within proton-exchange membrane fuel-cell anode loops. In these systems, hydrogen concentration and recirculation rates jointly govern fuel utilization and dictate when accumulated inert gases must be purged from the stack. Rather than using separate physical sensors or model-based observers, the technique extracts both parameters from a single pair of opposed acoustic transducers. The formulation separates composition-dependent mixture sound speed from path-averaged axial flow velocity along an oblique transit-time acoustic path. A thermophysical gas model constrained by synchronous temperature, pressure, and relative humidity measurements calculates dry-gas hydrogen content, while three-dimensional computational fluid dynamics simulations resolve flow distortion caused by pipe bends and intrusive sensor mountings. Experimental benchtop tests using hydrogen, nitrogen, and equimolar blends demonstrate that integrating acoustic inversion, velocity-profile conversion, and localized geometric calibration establishes a reliable framework for online fuel-cell anode monitoring.
Proton-exchange membrane fuel cells offer rapid low-temperature startup, high power density, and zero point-of-use emissions, making them attractive for heavy-duty transport subjected to wide load variations. Beyond electrochemical performance, operating durability requires balanced reactant distribution, membrane hydration, and thermal control over extended service lifetimes. At high current densities, strong multiphysics interactions link mass, heat, and charge transport, turning system observability into a necessity for robust stack control. The anode recirculation loop recovers unreacted hydrogen from the stack exhaust, raising fuel utilization and stabilizing stoichiometry. However, recirculation pumps and ejectors exhibit shifting entrainment ratios under dynamic backpressures, meaning actuator commands cannot reveal actual hydrogen delivery. Furthermore, membrane nitrogen crossover and internal water transport gradually transform the recycled gas into a humid, variable hydrogen-nitrogen mixture. Nitrogen accumulation lowers hydrogen partial pressure, while unnecessary purging wastes fuel. Direct physical measurements of composition and flow are therefore needed to distinguish inert-gas dilution from inadequate mechanical circulation.
Transit-time ultrasound addresses this diagnostic challenge because a single acoustic beam carries two independent sets of physical information. In the study’s sensing geometry, two transducers face each other across a circular pipe of inner diameter D along an acoustic path of length L inclined at an angle α to the centerline. For low-Mach-number gas flows where acoustic refraction is negligible, sound waves travel along straight paths. Downstream and upstream transit times depend directly on the mixture sound speed combined with the projected axial gas velocity. Summing reciprocal transit times suppresses convective flow effects to first order, isolating the gas sound speed. Conversely, taking the difference of reciprocal transit times cancels the sound speed, yielding the line-averaged velocity along the acoustic beam. Volumetric flow rate then follows from cross-sectional area integration and an empirical velocity-profile correction factor. Colocated temperature, pressure, and humidity sensors acquire synchronized thermodynamic states to constrain the subsequent composition inversion.
Extracting dry-gas hydrogen concentration from sound speed relies on an explicit thermophysical model under ideal-gas assumptions. Because the acoustic frequency is high relative to thermal diffusion rates, acoustic perturbations occur adiabatically. Sound speed equals the square root of the heat-capacity ratio multiplied by the gas constant and absolute temperature, divided by mixture molar mass. In the anode loop, the gas is treated as a single-phase ternary mixture of hydrogen, nitrogen, and water vapor, with trace carbon oxides neglected. Because hydrogen possesses a far lower molar mass than nitrogen or water, composition changes induce large shifts in acoustic velocity. Mixture heat capacities are derived through mole-fraction-weighted summations of temperature-dependent pure-gas polynomial equations. Because water vapor acts as a third thermodynamic variable, synchronous temperature and relative humidity inputs determine saturation pressure via the corrected Buck correlation. Once the vapor mole fraction is constrained, the remaining dry-gas fraction is shared between hydrogen and nitrogen, enabling direct mathematical inversion of dry-gas hydrogen concentration.
To examine real-gas departures under humid conditions, the authors performed a thermodynamic screening using compressibility factors and Kay-type pseudo-critical mixing rules. At an operating condition of 150 kilopascals absolute, 368.15 kelvin, and nominal vapor saturation, the corrected Buck relation yields a water-vapor partial pressure of 84.55 kilopascals. This corresponds to a wet mixture containing approximately 56.4 percent water vapor and 21.8 percent each of hydrogen and nitrogen. Mixing rules establish pseudo-critical values of 399.5 kelvin and 13,457.9 kilopascals, indicating a reduced temperature of 0.92 and a reduced pressure of 0.011. A generalized compressibility chart indicates a compressibility factor of roughly 0.995, which implies an acoustic velocity shift of only 0.25 percent via square-root scaling. The authors note that this calculation provides only a preliminary screening estimate. Because sound speed involves an isentropic derivative, a simple compressibility factor does not rigorously validate caloric or acoustic equations of state across all operating regimes.
Translating acoustic line-averaged velocity into cross-sectional volumetric flow requires compensating for non-uniform pipe velocity profiles. For fully developed laminar flow, the Hagen-Poiseuille parabolic profile yields a theoretical diametric profile factor of three-quarters, whereas turbulent flows exhibit flatter profiles represented by empirical power laws linked to Reynolds number. To bridge operating conditions, the researchers applied a finite-Reynolds-number profile correction. However, fuel-cell packaging involves tight pipe bends that distort the velocity distribution. The sensor was positioned downstream of the water separator and upstream of the recirculation device to ensure stable single-phase gas without device-induced pulsations. Computational fluid dynamics simulations conducted in COMSOL Multiphysics 6.3 evaluated flow through an upstream bend, recovery piping, an eight-diameter measurement section, downstream piping, and a second bend. Downstream of the initial bend, the high-velocity core shifts toward the outer radius while secondary Dean vortices develop. While the profile approaches axisymmetry around ten pipe diameters, downstream turning effects also propagate upstream into the measurement section.
To optimize straight piping lengths within constrained packaging spaces, the researchers examined eighty-one combinations of upstream and downstream lengths between two and eighteen pipe diameters. At a design condition of 230 kilopascals absolute, 20 volume percent hydrogen, 363.15 kelvin, and 20 meters per second, upstream pipe extension proved far more effective at reducing velocity error than downstream extension. A configuration with sixteen diameters upstream, ten diameters downstream, and a 45-degree acoustic angle was selected for implementation. For a 19.1-millimeter pipe, this layout requires 305.6 millimeters upstream and 191.0 millimeters downstream, totaling 496.6 millimeters of straight pipe and 648.6 millimeters including the 152-millimeter sensor chamber. Across sixteen orthogonal validation cases, this bend-only layout showed a maximum absolute velocity error of 1.143 percent, with four cases exceeding the one-percent design threshold. Introducing realistic intrusive transducer housings generated localized flow acceleration and downstream separation zones, demonstrating that bend corrections alone cannot account for physical sensor intrusion.
To compensate for transducer-induced flow disruption, the authors formulated a geometry-specific local correction factor, K-two, defined as a function of Reynolds number. Simulations employed a hybrid mesh combining swept hexahedral elements in straight sections with free tetrahedra around bends and recesses, refined by a ten-layer near-wall boundary mesh. Mesh-independence evaluations across seven resolutions confirmed numerical convergence. The local factor was calibrated across sixteen CFD cases spanning Reynolds numbers from 2395 to 30,913, producing a double-exponential curve fit with a correlation coefficient of 0.9946. Incorporating this factor lowered maximum calibration fitting residuals to 0.228 percent. The formulation was then evaluated against an independent sixteen-case validation set covering Reynolds numbers from 2532 to 24,406. Across these independent cases, absolute velocity-estimation residuals remained below 0.35 percent, with signed deviations ranging between negative 0.342 percent and positive 0.297 percent. These results verify that combining generic profile conversions with local geometric factors eliminates installation biases.
Experimental validation was conducted on a laboratory flow bench using pure hydrogen, pure nitrogen, and an equimolar dry blend under controlled temperature and pressure conditions. Gas conditioning maintained an absolute pressure of 265 kilopascals and a gas temperature and dew point of 64 degrees Celsius upstream of the 16D/10D ultrasonic test section, with a downstream reference meter providing benchmark data. Data were logged across three independent experiments per condition at a ten-hertz sampling rate for thirty seconds after flow stabilization. Flow estimations were converted to normal liters per minute and dry-gas hydrogen concentrations were tracked in percentage points. Experimental flow deviations satisfied a pointwise tolerance envelope defined by the greater of 5 normal liters per minute or 5 percent of reference flow, which meet at 100 normal liters per minute. The absolute 5-NLPM threshold successfully bound errors at low and zero flow rates. Colocated state sensors enabled accurate hydrogen concentration inversion across all test mixtures.
Integrating the coordinated ultrasonic sensor into functional fuel-cell systems involves specific engineering boundaries, packaging trade-offs, and error-propagation considerations. Because acoustic transit times and state measurements are evaluated simultaneously within the inversion algorithm, measurement uncertainties propagate jointly into the calculated states. A first-order covariance model using a unified Jacobian matrix describes how timing jitter, dimensional tolerances, and sensor errors influence final concentration and flow outputs. Furthermore, the acoustic model strictly assumes single-phase gas flow. In operating fuel cells, avoiding liquid-water condensation requires maintaining gas temperatures above the local dew point, reinforcing the necessity of installing the sensor immediately downstream of the water separator. Packaging the 648.6-millimeter straight piping run and accommodating transducer pressure losses represent important practical constraints in compact vehicle powertrains. While prospective applications include closed-loop ejector control and demand-driven anode purging, adapting the sensor to different pipe sizes or transducer geometries will require recalibration of the local correction factor.
Subject of Research: Ultrasonic measurement of hydrogen concentration and flow in fuel-cell anode loops
Article Title: Coordinated ultrasonic measurement of hydrogen concentration and recirculation flow in PEM fuel-cell anode loops
Article References: Zhou, Z., Yi, W., Ye, Z., & Lin, W. (2026). Coordinated ultrasonic measurement of hydrogen concentration and recirculation flow in PEM fuel-cell anode loops. Ionics. https://doi.org/10.1007/s11581-026-07598-6
Image Credits: AI Generated
DOI: 10.1007/s11581-026-07598-6
Keywords: PEM fuel cells, ultrasonic flowmeter, hydrogen concentration, anode recirculation, sound speed, computational fluid dynamics, sensor integration, fuel cell diagnostics, Coordinated, ultrasonic, measurement, hydrogen
