Machine Learning Breakthrough Enables Sensor-Free Methanol Concentration Control in Direct Methanol Fuel Cells

Strategic Insights -

Researchers at the Chinese Academy of Sciences have developed a novel data-driven strategy to control methanol concentration in direct methanol fuel cells (DMFCs) without the need for traditional sensors. This innovative approach, grounded in machine learning, has been successfully validated through a series of experiments and simulations. The study, published in the journal Energy, demonstrates the potential of this sensor-less control strategy to enhance the performance and reliability of high-power DMFC systems.

Key Takeaways:

  • The research team developed a semi-empirical electrochemical model that was calibrated against experimental polarization curves to improve the voltage-performance dataset.
  • The enriched data enhanced model robustness and enabled a systematic comparison of machine-learning algorithms, ultimately leading to the identification of a Bayesian-regularised artificial neural network as the most accurate voltage predictor.
  • The integrated model, which combined the surrogate models into a dynamic Matlab/Simulink framework, reproduced the transient stack voltage with mean absolute relative errors of 2.1% and 1.6% in two one-hour random-current tests.
  • The predicted methanol concentrations deviated from measurements by only 0.12 wt % and 0.09 wt %, respectively, over a 24-hour sensor-less control experiment on a 100 W DMFC stack.
  • The sensor-less control strategy maintained the inlet methanol concentration within 1.9-2.1 wt % throughout the experiment, with deviations remaining below 0.2 wt %.
  • The research concluded that the proposed data-driven strategy offers a novel perspective for advancing sensorless methanol concentration control strategies in DMFC systems.

Statistics:

  • The Bayesian-regularised artificial neural network yielded an R2 value of 0.99838 for voltage prediction.
  • The interaction-effect linear regression captured the methanol consumption rate with an R2 value of 0.99391 and an RMSE of 0.000473.
  • The integrated model reproduced the transient stack voltage with mean absolute relative errors of 2.1% and 1.6% in two one-hour random-current tests.
  • The predicted methanol concentrations deviated from measurements by only 0.12 wt % and 0.09 wt % over a 24-hour sensor-less control experiment.

Sources:

  • NewsRx. Studies from Chinese Academy of Sciences Yield New Information about Machine Learning (Machine Learning-based Simulation and Experiment of a Concentration Sensor-free Control Strategy for Direct Methanol Fuel Cells). Information Technology Newsweekly. November 4, 2025; p 826.
  • Machine Learning-based Simulation and Experiment of a Concentration Sensor-free Control Strategy for Direct Methanol Fuel Cells. Energy, 2025;335.