基于紫外光谱的油气两相流含气率检测研究

doi:10.3964/j.issn.1000-0593(2025)02-0522-10

摘要
参考文献
相关文章 (15)

全文: PDF (3116 KB)
输出: BibTeX | EndNote (RIS)

摘要：发动机润滑油系统中油气两相流含气率的检测对工业过程的安全运行有重要意义，准确和快速地测量出含气率对监控发动机的运行状态尤为重要。鉴于传统含气率检测方法的局限性，提出一种紫外光谱结合建模算法预测含气率的方法。首先，在波长为185~430 nm的光谱范围内，在5个不同两相流温度、3个不同两相流流速下共15个工况下，分别采集31组含气率范围在0.9%~3%的吸收光谱数据，光谱波长变量为1 799个。对光谱数据采用光谱-理化值共生距离算法（SPXY）将数据集划分为21组校正集，10组测试集，对不同含气率工况进行偏最小二乘（PLS）建模，得到测试集决定系数（R²）范围为0.63~0.91。针对不同工况预测效果差别过大的问题，采用中心化（center）、标准化（autoscaling）、SG卷积平滑、多元散射校正（MSC）、标准正态变量变换（SNV）、去趋势算法（Detrend）、OPLS正交信号校正等数据预处理矫正方法优化模型，得到Detrend-center-PLS模型预测效果最好，其中R²由0.903 2提高到0.955 7。针对光谱波长变量过多的问题，选取竞争性自适应加权算法（CARS）、蒙特卡洛无信息变量消除算法（MCUVE）和遗传算法（GA）三种方法进行波段降维，对降维后的光谱数据分别进行多元线性回归（MLR）、偏最小二乘和最小二乘支持向量机（LS-SVM）建模，得到预测结果最优为Detrend-center-CARS-PLS模型，与全波段模型比较，R²从0.955 7提升到0.959 8。由于优化效果并不显著，对三种降维方法取交集，重新进行建模，R²从0.959 8提升到0.967 1，优化效果有所提升。考虑到温度和流速对预测模型的影响，建立多工况含气率预测模型，采用与单工况相同的建模流程，得到autoscaling为最优预处理方法，且autoscaling-PLS模型的R²为0.948 8，波段降维方法将1 799波长变量消减到400～500个变量，降维效果明显。针对不同的建模方法，得到autoscaling-MCUVE-LS-SVM模型为多工况最佳预测模型，R²达到0.992 6。最后，通过对比单工况预测模型和多工况预测模型，得知多工况预测模型含气率预测效果优于单工况模型，能改善模型预测效果。结果表明，采用光谱分析法结合建模算法预测油气两相流含气率是具备可行性的，为发动机的安全运行提供了有效的监控方法。

关键词：紫外光谱；含气率；两相流；发动机

Abstract：The accurate and rapid measurement of the gas void fraction in the oil-gas two-phase flow of the engine lubrication system holds significant importance for ensuring the safe operation of industrial processes. The precise determination of the void fraction is particularly crucial for monitoring the operational status of the engine. In light of the limitations associated with traditional gas void fraction detection methods, a method is proposed that combines ultraviolet spectroscopy with a modeling algorithm for predicting the void fraction. Initially, within the range of 185 to 430 nm, 31 sets of absorption spectra data were collected at five different two-phase flow temperatures and three different two-phase flow velocities, encompassing 15 operational conditions. The spectra were obtained for gas void fractions ranging from 0.9% to 3%. A total of 1799 spectral wavelength variables were considered for analysis. The spectral-physicochemical value coexistence distance algorithm (SPXY) was employed to partition the spectral data set into 21 calibration and 10 test sets. Partial least squares (PLS) modeling was conducted for various gas void fraction conditions, resulting in a test set determination coefficient R²) range of 0.63 to 0.91. To address the significant variations in prediction performance across different operating conditions, various data preprocessing methods, including centering, autoscaling, Savitzky-Golay convolution smoothing, multiplicative scatter correction (MSC), standard normal variate (SNV) transformation, detrending, and orthogonal partial least squares (OPLS) orthogonal signal correction, were applied to optimize the model. The Detrend-center-PLS model exhibited the best predictive performance, with the R² increasing from 0.903 2 to 0.955 7. To address the issue of excessive spectral wavelength variables, three methods, Competitive Adaptive Reweighted Sampling (CARS), Monte Carlo Uninformative Variable Elimination (MCUVE), and Genetic Algorithm (GA), were employed for band dimension reduction. The reduced spectral data were then modeled using Multiple Linear Regression (MLR), Partial Least Squares, and Least Squares Support Vector Machine (LS-SVM). The optimal predictive model was determined to be the Detrend-center-CARS-PLS model, which exhibited an improved R² from 0.955 7 to 0.959 8 compared to the full-wavelength model. Due to the limited improvement in optimization, the intersection of the three-dimensionality-reduction methods was taken, and the modeling was re-conducted. The R² increased from 0.959 8 to 0.967 1, indicating a noticeable enhancement in optimization. Considering the influence of temperature and flow rate on the predictive model, a multi-condition gas void fraction prediction model was established using the same mprocess as the single-condition model- Autoscaling was identified as the optimal preprocessing method, and the R² for the autoscaling-PLS model was 0.948 8. The wavelength dimensionality reduction method reduced the 1799 wavelength variables to 400~500 demonstrating a significant dimensionality reduction. For different modeling methods, the autoscaling-MCUVE-LS-SVM model was identified as the best multi-condition predictive model, achieving an R² of 0.992 6. Finally, by comparing the single-condition predictive model with the multi-condition predictive model, it was found that the gas void fraction prediction performance of the multi-condition model was superior, improving overall predictive accuracy. The results indicate that using spectral analysis combined with modeling algorithms for predicting gas void fraction in oil-gas two-phase flow is feasible and provides an effective monitoring method for the safe operation of engines.

Key words：Ultraviolet spectrum; Gas void fraction; Two-phase flow; Engine

收稿日期: 2024-01-14 修订日期: 2024-04-28

中图分类号:

TK314

基金资助: 国家重点研发计划项目（2021YFE0112800），陕西省企业院所联合重点专项（2023-LL-QY-29）资助

通讯作者: 周延 E-mail: yan.zhou@mail.xjtu.edu.cn

作者简介: 李敏，女，1995年生，西安交通大学能源与动力工程学院硕士研究生 e-mail: 3122303048@stu.xjtu.edu.cn

引用本文:

李敏，殷雄，刘雪婧，马世一，周延，种道彤，熊兵，李锟. 基于紫外光谱的油气两相流含气率检测研究[J]. 光谱学与光谱分析, 2025, 45(02): 522-531.
LI Min, YIN Xiong, LIU Xue-jing, MA Shi-yi, ZHOU Yan, CHONG Dao-tong, XIONG Bing, LI Kun. Study on Detection of Gas Void Fraction in Oil-Gas Two-Phase Flow Based on Ultraviolet Spectroscopy. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(02): 522-531.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2025)02-0522-10 或 https://www.gpxygpfx.com/CN/Y2025/V45/I02/522

[1] LI Shu, HU Jian-ping, TAN Yi, et al(李澍, 胡剑平, 谭逸, 等). Journal of Propulsion Technology(推进技术), 2024, 45(5): 130.
[2] LIU Qi-chao, ZHOU Yun-long, CHEN Cong(刘起超, 周云龙, 陈聪). CIESC Journal(化工学报), 2023, 74(6): 2391.
[3] Hernandez-Alvarado F, Kleinbart S, Kalaga D V, et al. International Journal of Multiphase Flow, 2018, 102: 119.
[4] Park H J, Akasaka S, Tasaka Y, et al. IEEE Sensors Journal, 2020, 21(1): 273.
[5] Jahangir S, Wagner E C, Mudde R F, et al. International Journal of Multiphase Flow, 2019, 120: 103085.
[6] CHENG Jie, GUO Ya-jun, WANG Teng, et al(程洁, 郭亚军, 王腾, 等). CIESC Journal(化工学报), 2019, 70(4): 1375.
[7] ZHANG Li, SUI Zhi-qiang, ZHOU Xiang, et al(张理, 随志强, 周翔, 等). Mechanical and Electrical Equipment(机电设备), 2021, 38(3): 90.
[8] Zych M, Hanus R, Wilk B, et al. Measurement, 2018, 129: 288.
[9] LI Tao, YUAN Chao, QI Feng-feng, et al(李涛, 袁超, 齐锋锋, 等). Instrument Technique and Sensor(仪表技术与传感器), 2023, (7): 105.
[10] FANG Li-de, KONG Heng-zheng, HAN Bang-bang, et al(方立德, 孔恒正, 韩棒棒, 等). Acta Metrologica Sinica(计量学报), 2022, 43(12): 1616.
[11] He D H, Chen S L, Bai B F. Experiment Thermal and Fluid Science, 2019, 102: 61.
[12] WANG Teng, BI Qin-cheng, GUI Miao, et al(王腾, 毕勤成, 桂淼, 等). Journal of Engineering Thermophysics(工程热物理学报), 2022, 43(6): 1565.
[13] LIU Shuai, CHEN Cong, LIU Li, et al(刘帅, 陈聪, 刘莉, 等). Nuclear Power Engineering(核动力工程), 2022, 43(5): 43.
[14] DAI Pei-yu, ZHANG Xin, MAO Xing, et al(戴佩玉, 张欣, 毛星, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2023, 39(16): 160.
[15] YI Rong-nan, WU Yan, WANG Jun-li, et al(易荣楠, 吴燕, 王俊力, 等). Chinese Journal of Analytical Chemistry(分析化学), 2023, 51(2): 147.
[16] LIU Qiang, LIU Shao-bo, LU Xue-song, et al(刘强, 柳少波, 鲁雪松, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2022, 42(9): 2679.
[17] Zhou Y, Cui H S. Measurement Science and Technology, 2023, 34(3): 035501.
[18] Wang C, Zhao N, Fang LD. Experimental Thermal and Fluid Science, 2016, 76: 98.
[19] FANG Li-de, LÜ Xiao-hui, TIAN Ji, et al(方立德，吕晓晖，田季, 等). China Measurement & Test(中国测试), 2018, 44(1): 21.
[20] XIE Chen, FANG Li-de, WANG Dong-xing(谢辰，方立德，王东星, 等). Metrology Science and Technology(计量科学与技术), 2018, 2: 3.