Hydrogen leak detection in refuelling stations requires sensing elements that combine high sensitivity at low concentration, fast response, low power and long-term baseline stability, together with a measurement chain that delivers a trustworthy signal. We describe a micro-electromechanical system (MEMS) hydrogen sensor whose sensitive layer is a Pd–Ag alloy film, and we set out both the data-driven procedure that fixed its design and the measurement procedure by which it was evaluated. Sensitivity, response/recovery time and drift are defined as competing objectives under process and power constraints from a multi-physics description of the sensing process—surface adsorption and dissociation at Pd sites, diffusion into the alloy lattice, hydride formation and the resulting increase of electron scattering. The design space is explored by Latin hypercube sampling, evaluated with a surrogate model and searched by a multi-objective evolutionary algorithm to obtain a Pareto-optimal design set. The device is then fabricated in an eight-step MEMS process on a Si/SiO₂/Si₃N₄ micro-hotplate platform and characterized on a dynamic gas-mixing bench with synchronized data acquisition and repeated measurements at every point, so that the Type A and Type B uncertainty contributions can be combined into an expanded uncertainty with a coverage factor of two. The sensor resolves 1 ppm H₂ and, at 100 ppm, reaches T90 = 8.6 s and T10 = 18.4 s at 32 mW, with a response about 45% faster and a cycling baseline drift about 33% lower than a pure-Pd film device, and a working power more than 70% below that of a metal-oxide device of comparable sensitivity.
 

hydrogen sensor, MEMS, Pd–Ag alloy, sensor characterization, data acquisition, multi-parameter optimization, multi-objective evolutionary algorithm, measurement uncertainty, data analytics