Bayesian optimization for the direct optimization of temperature-programmed separations in ultra-fast gas...

Ultra-Fast gas chromatography electronic nose (UFGC-E-nose) systems are essential for the rapid volatile fingerprinting of complex food and beverage matrices. Despite their efficiency, optimizing separation in these instruments is technically challenging. The minimal thermal mass of micro-columns, coupled with extreme ramp rates reaching 5°Cs^-1, induces significant thermal hysteresis that often invalidates traditional physical retention models. This study introduces an automated optimization framework based on Bayesian optimization (BO) to systematically improve the separation and discrimination efficiency of UFGC-E-nose systems. By employing a multi-objective chromatographic response function, the framework balances separation quality against analysis duration through a dynamic weighting strategy that shifts the optimization priority from resolution toward throughput. Within only 35 experimental trials, the BO framework identified optimal gradient parameters that increased the number of detected peaks by 28.3% and improved average resolution by 62.5% relative to default settings. Most target volatiles reached baseline separation (R_s≥1.5), with substantial improvements in the reproducibility of retention times and peak areas, as evidenced by relative standard deviation (RSD) reductions of 75% and 60%, respectively. The practical utility of the optimized method was validated through the discriminant analysis of Baijiu samples with varying mixing ratios. The resulting model captured 82.4% of the total variance and exhibited a clear linear trajectory along the first discriminant factor (DF1) that accurately reflected compositional gradients. This data-efficient strategy bypasses the limitations of physicochemical modeling under thermal non-equilibrium conditions, providing a robust solution for automated method development in high-speed GC systems.