An automatic spectral baseline estimation method and its application in industrial alkali-pulverized coal flames

Flame emission spectroscopy has a wide range of applications in combustion diagnostics. Processing the flame emission spectra is a key step for further spectral analysis, including both quantitative and qualitative analyses. This study examined an improved asymmetric reweighted penalized least squares (arPLS) method to accurately and automatically estimate flame emission spectral baselines without priori information for complex spectra (both continuous and discontinuous spectral) preprocessing scenarios. This method innovatively couples the original arPLS method with the radiation thermometry theory. Thus, it inherits the advantages of the original method and has increased robustness to parameter selection. The suitability of the proposed method in terms of the root mean square error (RMSE) and goodness-of-fit coefficient (GFC) was verified using the baseline estimation results of simulated spectra and comparisons with the results of the original arPLS and piecewise polynomial fitting (PPF) methods. This illustrated the feasibility and stability of the method as a key step in spectral signal processing for combustion diagnosis. Then, the temperature and emissivity of industrial alkali-pulverized coal flames were measured using spectral equipment based on the presented method. The results showed that the continuous spectral lines estimated by this method could reach a GFC value of 0.9999. Compared with the temperature measurement results based on the baselines obtained by PPF, the temperature errors calculated using the estimation results of the presented method were within 1%.