부산대학교 도서관

In this paper, an optical-based device for online black powder detection in gas pipelines is proposed. The device uses different optical wavelengths within the infrared (IR) range and applies chemometric algorithms to quantify the actual amount of black powder. Hence, three methods based on near-infrared (NIR), mid-infrared (MIR), and Raman spectroscopy are used to acquire various spectra. The principal component regression (PCR) and the partial least square regression (PLSR) algorithms are then applied to assess the capability of each of the three methods when black powder is subject to different environmental conditions that may occur in real-life fields. The experimental results indicated that the mean squared error of prediction for the PLSR is 0.0008731, 0.0001983, and 5.04e-5 for NIR, MIR, and Raman, respectively, while for the PCR it is 0.0009065, 0.0002068, and 5.099e-5. Also, the coefficient of determination $(R^{2})$ for the PLSR was 0.9744, 0.9753, and 0.998199 for NIR, MIR, and Raman, respectively, while for the PCR it was 0.9743, 0.9744, and 0.998165. In addition, both PCR and PLSR completed the analysis very fast (in tens or hundreds of microseconds), however, PLSR accomplished the prediction analysis faster than PCR which serves as an advantage for online monitoring especially when multiprobes are used for the monitoring. Hence, while both PLSR and PCR perform equally well, the PLSR is even more robust since it can compensate for systematic and human errors more than PCR. The predictions from all three techniques (NIR, MIR, and Raman spectroscopy) were similarly good and no specific technique was superior to the others.