Improving the Laser-Induced Breakdown Spectroscopy (LIBS) data for analyzing major elements and calorific value of mixed coal using data pre-processing

국문 초록 (Abstract)

레이저 유도 붕괴 분광법(LIBS)은 시료에 별도의 화학처리 없이 시료 표면에
레이저를 조사하여 시료 표면에서 발생된 플라즈마로부터 방출되는 원자들의 분
광선을 측정함으로써 물질의 화학적 조성을 감지, 확인할 수 있는 정성 및 정량
분석 기술이다. 실험시료는 분말형태의 석탄 표준시료를 혼합한 혼합석탄으로 펠
렛 형태로 만들었다. 그리고 그것을 레이저 유도 붕괴 분광법을 이용해 시료를
측정하며 실험을 진행하였다. 혼합석탄에서 측정한 스펙트럼 데이터와 공업분석
한 혼합석탄의 원소 별 농도간의 관계를 부분 최소 자승 회귀 모형(Partial Least
Square Regression model)으로 나타내었다. 마찬가지로, 측정한 데이터를 두 가지 방
식의 데이터 전처리(Savitzky-Golay (SG) smoothing, SG derivative)를 활용하여 공업
분석한 혼합석탄의 원소 별 농도간의 관계도 PLSR model로 나타냈다. 세 가지 방
식의 PLSR model의 신뢰성을 비교하기 위한 매개변수로 결정계수(𝑅2), 평균제곱
근오차(Root Mean Square Error), 상대오차, 제곱근오차평균(RMSE average)을 이용하
였다. 그 결과, SG derivative processing 데이터의 PLSR model에서 가장 신뢰도가 높
고 각 원소들의 RMSE (avg.)가 가장 낮았다. 가장 신뢰성이 높게 나타난 SG
derivative pre-processing 데이터의 PLSR model의 예측능력을 알아보기 위해 잔차 예
측 편차(Residual Predictive Deviation) 매개변수를 이용하여 각 원소들의 PLSR
model의 신뢰할 수 있는 모형인지 알아보았다. 마지막으로, PLSR model을 통해 알
아본 주요 원소들의 예측 농도를 알 수 있으며, 각 원소들의 농도를 고위발열량
식인 Dulong’s equation에 대입하여 실제 발열량과 예측 발열량을 상대오차로 비교
하였고, SG derivative pre-processing 데이터에서 가장 낮은 오차가 나타나는 것을 확
인하였다.

다국어 초록 (Multilingual Abstract)

Laser induced breakdown spectroscopy (LIBS) does not require chemical treatment of the
sample. The sample is irradiated with a laser, and plasma is generated on the sample surface.
After that, the ray of light emitted from the generated plasma is measured. It is an analytical
technique that can detect, quantify, and confirm the chemical composition of a substance.
Experimental samples were made by mixing standard samples of powdered coal in the form
of mixed coal pellets. These pellets were measured using LIBS. The measured sample data
and comparable data obtained by industrial analysis of mixed coal were expressed in terms
of partial least square regression (PLSR) model. Likewise, the measured data were
subjected to two pre-processing methods (Savitzky-Golay (SG) smoothing, SG derivative)
to develop two additional PLSR models of the relationship between LIBS readings and the
composition of mixed coal obtained by industrial analysis. The concentration of major
elements can be predicted through the above PLSR models. To compare the reliability of
the three PLSR models, I used the determination coefficient (𝑅2), root mean square error
(RMSE), relative error, and RMSE average as parameters. The SG derivative preprocessing
PLSR model showed the highest reliability and the average RMSE of each
element was the lowest. To determine the predictability of the most reliable SG derivative
PLSR model, I used the residual predictive deviation (RPD) parameter to determine
whether the PLSR model of each element was reliable. Finally, I can find the predicted
concentration of the major element using the PLSR model. Then, the concentration of each
element is put into Dulong’s equation, which is a higher heating value equation, and the
difference between the measured and predicted calorific values are compared by relative
error. Therefore, it was confirmed that the lowest error appears in the SG derivative preprocessing
PLSR model.

목차 (Table of Contents)

• Chapter 1 Introduction 1
• 1.1 Research background 1
• 1.2 Research objectives 3
• Chapter 2 Materials and methods 4
• 2.1 Laser induced breakdown spectroscopy (LIBS) overview 4
• 2.2 Sample preparation 6
• 2.3 Data processing and analysis 11
• Chapter 3 Results and discussion 12
• 3.1 Mixed coal major element analysis 12
• 3.1.1 Partial least square regression (PLSR) of major elements of mixed coal 12
• 3.1.2 Root mean square error (RMSE) average of major elements of mixed coal 30
• 3.1.3 Residual predictive deviation (RPD) of major elements of mixed coal 33
• 3.2 Calorific value of mixed coal analysis 35
• 3.2.1 Relative error of calorific value for the PLSR of the original data 35
• 3.2.2 Relative error of calorific value for the PLSR of the SG derivative pre-processing data 36
• Chapter 4 Conclusions 39
• References 41
• 국문요지 44
• 감사의 글 45