Genetic programming control of self-excited thermoacoustic oscillations in a turbulent hydrogen–methane combustor

Thermoacoustic instabilities are a key challenge in developing sustainable combustion systems. In this experimental study, we present the first application of a data-driven machine learning algorithm based on genetic programming (GP) to suppress self-excited thermoacoustic oscillations in a turbulent premixed combustor operating with hydrogen-enriched fuels. The GP algorithm evolves model-free control laws via genetic operations such as replication, mutation, and crossover. Its performance is optimized through a cost function that balances the thermoacoustic amplitude reduction against the actuator power consumption. We evaluate GP in both closed-loop and open-loop configurations, benchmarking these against traditional open-loop time-periodic actuation. We find that GP closed-loop control proves superior in every metric evaluated, achieving the highest amplitude reduction with the lowest power consumption. This efficient suppression is physically achieved via synchronous quenching without resonant amplification, where GP actuation synchronizes the acoustic field and disrupts its coupling with the heat-release-rate (HRR) fluctuations of the flame. This disruption inhibits the formation of large-scale coherent vortices, resulting in a steadier HRR field decoupled from the acoustics, as evidenced by phase drifting and reduced Rayleigh index values. We also find that the GP algorithm is robust across varying reactant flow velocities, combustor lengths, and hydrogen concentrations, consistently yielding thermoacoustic amplitude reductions of 80%–94%. These findings establish GP as an effective, efficient and robust data-driven strategy for controlling thermoacoustic instabilities in turbulent combustion systems, including those fueled with hydrogen-enriched mixtures, advancing the development of sustainable energy technology. Novelty and significance statement: This experimental study is the first to apply genetic programming (GP) in both closed-loop and open-loop forms to suppress self-excited thermoacoustic oscillations in a turbulent combustor fueled by hydrogen- enriched mixtures. The GP algorithm discovers model-free control laws that achieve synchronous quenching (SQ) of the thermoacoustic mode by disrupting the flame–acoustic coupling, without resonant amplification of the actuation signal. GP closed-loop control outperforms both GP open-loop and conventional time-periodic forcing, achieving 80%–94% amplitude reduction across a range of Reynolds numbers, combustor lengths, and hydrogen power fractions while minimizing the actuation power. These results establish GP as an effective, efficient and robust strategy for active control of thermoacoustic instabilities in turbulent combustion systems, advancing sustainable energy technology.