Hottel—Broughton-corrected visible-near-infrared spectral pyrometry of high-temperature sooting flames using partial least squares regression

Broadband emission pyrometry enables non-intrusive temperature measurements in high-temperature sooting flames, but non-gray soot radiation, detector noise, and wavelength-dependent attenuation distort the measured spectrum. We developed a partial least squares (PLS) thermometry method that combines Hottel–Broughton emissivity correction with regression over 600–1000 nm. The baseline PLS model was trained on normalized Planck spectra from 1073 to 4273 K, and a compact four-factor model was selected using Monte Carlo simulations with 1% and 2% Gaussian perturbations. Validation against blackbody-furnace spectra at 1173 and 1273 K gave relative deviations of −1.25% and −0.25%, respectively. For fuel-rich oxy-acetylene flames, the emissivity correction was applied before normalization and PLS inversion; the mean retrieved temperature increased from 1997.8 to 3011.1 K with increasing oxygen flow. Varying the assumed dispersion index from 1.20 to 1.50 produced signed temperature deviations from −4.44% to +2.46%. Under 2% distributed Gaussian noise, PLS reduced the largest relative error to below approximately 5%, compared with approximately 8% for normalized Planck fitting and 6% for two-color pyrometry. Its median timing advantage over normalized Planck fitting was only 0.0096 ms (9.6 μs) per spectrum and is therefore secondary to the observed random-noise tolerance. The PLS inversion remained sensitive to spectral-slope errors, wavelength shifts, emissivity assumptions, and window fouling.