This website requires JavaScript.

Deep learning-based fast time-resolved flame emission spectroscopy in high-pressure combustion environment

Taekeun YoonSeon Woong KimHosung ByunYounsik KimCampbell D. CarterHyungrok Do
Jul 2022
A novel deep learning strategy is developed for fast and accurate gasproperty measurements using flame emission spectroscopy (FES). Particularly,the short-gated fast FES is essential to resolve fast-evolving combustionbehaviors. However, as the exposure time for capturing the flame emissionspectrum gets shorter, the signal-to-noise ratio (SNR) decreases, andcharacteristic spectral features indicating the gas properties becomerelatively weaker. Then, the property estimation based on the short-gatedspectrum is difficult and inaccurate. Denoising convolutional neural networks(CNN) can enhance the SNR of the short-gated spectrum. A new CNN architectureincluding a reversible down- and up-sampling (DU) operator and a loss functionbased on proper orthogonal decomposition (POD) coefficients is proposed. Fortraining and testing the CNN, flame chemiluminescence spectra were capturedfrom a stable methane-air flat flame using a portable spectrometer (spectralrange: 250-850 nm, resolution: 0.5 nm) with varied equivalence ratio (0.8-1.2),pressure (1-10 bar), and exposure time (0.05, 0.2, 0.4, and 2 s). The longexposure (2 s) spectra were used as the ground truth when training thedenoising CNN. A kriging model with POD is trained by the long-gated spectrafor calibration and then prediction of the gas properties taking the denoisedshort-gated spectrum as the input. The measurement or property predictionerrors of pressure and equivalence ratio using the new technique were estimatedto be 5.7% and 1.5% with 0.2 s exposure, which are exceptionally good andtypically not achievable with such low SNR spectrum signals without a signalamplifier.