Machine Learning Applications of Two-Phase Flow Data in Polymer Electrolyte Fuel Cell Reactant Channels

Multiphase flow is a primary concern in many energy conversion and storage technologies including fuel cells and flow batteries which utilize intricate flow-field patterns to facilitate reactant and byproduct transport. Due to complex physics and the limitations of numerical methods, new techniques of collecting and evaluating two-phase behavior in these reactant channels are needed. This paper demonstrates a novel method for two-phase data collection, processing, and its use in a machine learning algorithm. Decision tree (DT) regressions were used to correlate liquid distributions in reactant channels with the two-phase flow pressure drop along the channel. A transparent 3.0 mm x 2.4 mm rectangular channel was used to simulate the two-phase flow conditions of a polymer-electrolyte fuel cell (PEFC) reactant channel by injecting water through a gas-diffusion layer (GDL) while air was flowed through the channel. A synchronized camera and pressure transducer setup collected images at 5 Hz of the liquid distribution and the two-phase flow pressure drop. By training using the liquid distributions as inputs and the corresponding pressure drop data as outputs, the DT models achieved pressure drop prediction accuracies in most cases exceeding 90%. Applications for flow-field design and liquid saturation estimation based on pressure are discussed.

Automated summary

This study demonstrates a novel method for collecting and evaluating two-phase behavior in fuel cell reactant channels, achieving pressure drop prediction accuracies exceeding 90% with decision tree regressions. Applications for flow-field design and liquid saturation estimation based on pressure are discussed.