Industrial at-line analysis of coal properties using laser-induced breakdown spectroscopy combined with machine learning

• Published in: Fuel (Guildford) Vol. 306; p. 121667
• Main Authors: Song, Weiran, Hou, Zongyu, Gu, Weilun, Wang, Hui, Cui, Jiacheng, Zhou, Zhenhua, Yan, Gangyao, Ye, Qing, Li, Zhigang, Wang, Zhe
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 15-12-2021; Elsevier BV
• Subjects: Calorific value; Circuit design; Coal; Coal properties; Coal utilization; Coal-fired power plant; Environmental factors; Industrial pollution; Industry standards; Integrated circuits; Laser induced breakdown spectroscopy; Learning algorithms; Machine learning; Pollution control; Power plants; Quantitative analysis; Regression; Regression models; Signal to noise ratio; Spectroscopy; Spectrum analysis; Sulfur; Coal-fired power plant; Regression; Coal properties; Laser-induced breakdown spectroscopy; Machine learning; Quantitative analysis
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2021.121667

[Display omitted]
•A fully automatic LIBS system was developed for at-line, in-situ industrial coal analysis.•A series of techniques were integrated to improve measurement accuracy and repeatability.•Synergic regression was proposed to improve quantification performance and model interpretability.•The LIBS system was rigorously validated in an industrial setting.•The LIBS system met the industrial standards for coal analysis.
Coal analysis is of great importance to improve coal combustion/utilization efficiency and operation safety and to reduce pollution. In this work, we develop a LIBS system for at-line coal analysis, which can pre-treat coal blocks into pressed pellets, acquire sample spectra and quantify coal properties automatically and continuously. A series of techniques are integrated in the system, including laser energy monitor and control, plasma modulation and collinear spectra collection. Moreover, the system has an integrated circuit board that is designed to precisely control the time-sequence of the hardware components; the overall design of the system ensures environmental factors are stabilised. These design considerations improve raw signal repeatability and signal-to-noise ratio. Furthermore, to improve the quantification accuracy without the use of LIBS physics knowledge, a new machine learning method is proposed, namely synergic regression (SR), which embeds a linear model in nonlinear regression. It inherits the high accuracy of nonlinear methods whilst being able to explain how specific variables contribute to the prediction. The system was demonstrated and evaluated in a real power plant for 10 weeks. The average prediction errors of calorific value (MJ/kg), sulphur (%) and volatile (%) were 0.299, 0.077 and 0.590, respectively. The evaluation demonstrated that the developed LIBS system meets the industry standards for at-line and in-situ coal analysis. Therefore, the LIBS system has significant potential impact on practices in coal utilization and similar industrial processes.