Optimization of LBM multiphase flow models based on machine learning and particle swarm algorithm

doi:10.11949/0438-1157.20240813

Abstract

Abstract:

When using the lattice Boltzmann method (LBM) to simulate the slug flow with low capillary number, the bubble development process is complex and the model control parameters are difficult to select. When the selected parameters are inappropriate, erroneous non-physical phenomena will occur, thereby reducing the calculation accuracy. The LBM multiphase flow process model is established through machine learning, and the particle swarm algorithm is used to optimize the hyperparameters of the machine learning model, and further optimize the control parameters in the LBM modeling process. In this paper, a coupled multiphase flow numerical simulation model of LBM-machine learning-particle swarm algorithm is established. Based on this model, the influence of the flow parameters of the elastic flow in the T-shaped microchannel on the stability of the bubble evolution process is investigated. The simulation results show that the proposed LBM multiphase flow model can predict the bubble elongation rate under complex conditions, based on which the optimal gas-liquid two-phase inlet flow rate relationship is found through elongation rate analysis, which effectively solves the problem of elastic flow instability under low capillary number, and significantly improves the simulation calculation accuracy and computational efficiency.

Key words: lattice Boltzmann method, microchannel slug flow, machine learning, particle swarm optimization algorithm, model optimization

摘要：

在利用格子Boltzmann方法（LBM）模拟低毛细数的弹状流流动时，由于气泡发展过程复杂，模型控制参数选择难度大，当所选参数不当时，会产生错误的非物理现象，从而降低计算精度。通过机器学习建立LBM多相流过程模型，采用粒子群算法优化机器学习模型的超参数，进一步优化LBM建模过程中的控制参数，建立了LBM-机器学习-粒子群算法耦合多相流数值模拟模型。基于该模型研究了T型微通道内弹状流流动参数对气泡演化过程稳定性的影响。模拟结果表明，所建LBM多相流模型能预测复杂条件下气泡伸长率，在此基础上通过伸长率分析找到了最优气液两相进口流速关系，有效解决了低毛细数下弹状流流动不稳定性问题，显著提高了模拟计算精度与计算效率。

关键词: 格子Boltzmann法, 微通道弹状流, 机器学习, 粒子群算法, 模型优化

CLC Number:

TQ 052

Yaqi HOU, Wei ZHANG, Hong ZHANG, Feiyu GAO, Jiahua HU. Optimization of LBM multiphase flow models based on machine learning and particle swarm algorithm[J]. CIESC Journal, 2025, 76(3): 1120-1132.

侯亚祺, 张玮, 张鸿, 高飞雨, 胡嘉华. 基于机器学习与粒子群算法的LBM多相流模型优化[J]. 化工学报, 2025, 76(3): 1120-1132.

Figures/Tables 24

References 32

1	尧超群, 乐军, 赵玉潮, 等. 微通道内气-液弹状流动及传质特性研究进展[J]. 化工学报, 2015, 66(8): 2759-2766.
	Yao C Q, Le J, Zhao Y C, et al. Review on flow and mass transfer characteristics of gas-liquid slug flowin microchannels[J]. CIESC Journal, 2015, 66(8): 2759-2766.
2	Zaloha P, Kristal J, Jiricny V, et al. Characteristics of liquid slugs in gas-liquid Taylor flow in microchannels[J]. Chemical Engineering Science, 2012, 68(1): 640-649.
3	Peng D Y, Xin F, Zhang L X, et al. Experiments and modeling on bubble uniformity of Taylor flow in T-junction microchannel[J]. Chemical Engineering Science, 2015, 128: 11-20.
4	Gupta R, Fletcher D F, Haynes B S. On the CFD modelling of Taylor flow in microchannels[J]. Chemical Engineering Science, 2009, 64(12): 2941-2950.
5	Jeong H H, Yadavali S, Issadore D, et al. Liter-scale production of uniform gas bubbles via parallelization of flow-focusing generators[J]. Lab on a Chip, 2017, 17(15): 2667-2673.
6	Dangla R, Kayi S C, Baroud C N. Droplet microfluidics driven by gradients of confinement[J]. Proceedings of the National Academy of Sciences of the United States of America, 2013, 110(3): 853-858.
7	Chen L, Kang Q J, Mu Y T, et al. A critical review of the pseudopotential multiphase lattice Boltzmann model: methods and applications[J]. International Journal of Heat and Mass Transfer, 2014, 76: 210-236.
8	Molaeimanesh G R, Saeidi Googarchin H, Qasemian Moqaddam A. Lattice Boltzmann simulation of proton exchange membrane fuel cells-a review on opportunities and challenges[J]. International Journal of Hydrogen Energy, 2016, 41(47): 22221-22245.
9	Qian D Y, Lawal A. Numerical study on gas and liquid slugs for Taylor flow in a T-junction microchannel[J]. Chemical Engineering Science, 2006, 61(23): 7609-7625.
10	Jang H S, Park D S. Microfabrication of microchannels for fuel cell plates[J]. Sensors, 2010, 10(1): 167-175.
11	Yuan M, Feng H B, Zhang W, et al. Gas-liquid sulfonation in T-shaped microchannel for DBSA synthesis and process optimization with response surface method[J]. Chemical Engineering and Processing-Process Intensification, 2022, 174: 108890.
12	Takebayashi Y, Sue K, Yoda S, et al. Direct carbonylation of nitrobenzene to phenylisocyanate using gas-iquid slug flow in microchannel[J]. Chemical Engineering Journal, 2012, 180: 250-254.
13	Pahinkar D G, Garimella S, Robbins T R. Feasibility of using adsorbent-coated microchannels for pressure swing adsorption: parametric studies on depressurization[J]. Industrial & Engineering Chemistry Research, 2015, 54(41): 10103-10114.
14	Shafizadeh A, Shahbeig H, Nadian M H, et al. Machine learning predicts and optimizes hydrothermal liquefaction of biomass[J]. Chemical Engineering Journal, 2022, 445: 136579.
15	Bandi P, Manelil NP, Maiya MP, et al. CFD driven prediction of mean radiant temperature inside an automobile cabin using machine learning[J]. Thermal Science and Engineering Progress, 2023, 37: 101619.
16	Qian Y H, D'Humières D, Lallemand P. Lattice BGK models for Navier-Stokes equation[J]. EPL (Europhysics Letters), 1992, 17: 479.
17	Guo Z L, Zheng C G, Shi B C. Discrete lattice effects on the forcing term in the lattice Boltzmann method[J]. Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics, 2002, 65(4 Pt 2B): 046308.
18	Yuan P, Schaefer L. Equations of state in a lattice Boltzmann model[J]. Physics of Fluids, 2006, 18(4): 042101.
19	Li Q, Luo K H, Li X J. Forcing scheme in pseudopotential lattice Boltzmann model for multiphase flows[J]. Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics, 2012, 86(1 Pt 2): 016709.
20	Li Q, Luo K H, Li X J. Lattice Boltzmann modeling of multiphase flows at large density ratio with an improved pseudopotential model[J]. Physical Review E, 2013, 87(5): 053301.
21	Krüger T, Kusumaatmaja H, Kuzmin A, et al. The Lattice Boltzmann Method: Principles and Practice[M]. Switzerland: Springer, 2017.
22	Ziegler D P. Boundary conditions for lattice Boltzmann simulations[J]. Journal of Statistical Physics, 1993, 71(5): 1171-1177.
23	Li Q, Luo K H, Kang Q J, et al. Contact angles in the pseudopotential lattice Boltzmann modeling of wetting[J]. Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics, 2014, 90(5): 053301.
24	Garstecki P, Fuerstman M J, Stone H A, et al. Formation of droplets and bubbles in a microfluidic T-junction-scaling and mechanism of break-up[J]. Lab on a Chip, 2006, 6(3): 437-446.
25	Warnier M J F, de Croon M H J M, Rebrov E V, et al. Pressure drop of gas-liquid Taylor flow in round micro-capillaries for low to intermediate Reynolds numbers[J]. Microfluidics and Nanofluidics, 2010, 8(1): 33-45.
26	Ganapathy H, Al-Hajri E, Ohadi M M. Phase field modeling of Taylor flow in mini/microchannels (part Ⅰ): Bubble formation mechanisms and phase field parameters[J]. Chemical Engineering Science, 2013, 94: 138-149.
27	Dai L, Cai W, Xin F. Numerical study on bubble formation of a gas-liquid flow in a T-junction microchannel[J]. Chemical Engineering & Technology, 2009, 32(12): 1984-1991.
28	Awad M, Khanna R. Support Vector Regression[M]//Efficient Learning Machines. Berkeley, CA: Apress, 2015: 67-80.
29	Masampally V S, Pareek A, Runkana V. Cascade Gaussian process regression framework for biomass prediction in a fed-batch reactor[C]//2018 IEEE Symposium Series on Computational Intelligence (SSCI). Bangalore, India: IEEE, 2018: 128-135.
30	刘毅, 王海清, 李平. 局部最小二乘支持向量机回归在线建模方法及其在间歇过程的应用[J]. 化工学报, 2007, 58(11): 2846-2851.
	Liu Y, Wang H Q, Li P. Local least squares support vector regression with application toonline modeling for batch processes[J]. Journal of Chemical Industry and Engineering (China), 2007, 58(11): 2846-2851.
31	Yi G, Shi J Q, Choi T. Penalized Gaussian process regression and classification for high-dimensional nonlinear data[J]. Biometrics, 2011, 67(4): 1285-1294.
32	刘东飞, 张帆, 刘铮, 等. 机器学习势及其在分子模拟中的应用综述[J]. 化工学报, 2024, 75(4): 1241-1255.
	Liu D F, Zhang F, Liu Z, et al. A review of machine learning potentials and their applications tomolecular simulation[J]. CIESC Journal, 2024, 75(4): 1241-1255.

特性	BP神经网络	SVR	GPR	随机森林
非线性建模能力	强	强	强	强
学习与泛化能力	强	强	强	强
适用性	分类、回归、聚类等	回归、高维数据处理	回归、不确定性量化	回归、高维数据处理
训练时间	长	较长	较长	较短
局部最优	易陷入	可能	较少	较少
参数敏感性	高	高	中等	低
适用数据规模	大规模	中等规模	小规模	大规模
不确定性估计	不提供	不提供	提供	不提供
计算复杂度	中等	高	高	高

特性	BP神经网络	SVR	GPR	随机森林
非线性建模能力	强	强	强	强
学习与泛化能力	强	强	强	强
适用性	分类、回归、聚类等	回归、高维数据处理	回归、不确定性量化	回归、高维数据处理
训练时间	长	较长	较长	较短
局部最优	易陷入	可能	较少	较少
参数敏感性	高	高	中等	低
适用数据规模	大规模	中等规模	小规模	大规模
不确定性估计	不提供	不提供	提供	不提供
计算复杂度	中等	高	高	高

序号	连续相速度	分散相速度	gama值	连续相黏度	分散相黏度	伸长率/%
1	0.0015	0.0015	0.12	2.20	0.6	62.6
2	0.0020	0.0020	0.12	2.20	0.6	74.5
3	0.0025	0.0025	0.12	2.20	0.6	70.2
4	0.0030	0.0030	0.12	2.20	0.6	67.8
5	0.0035	0.0035	0.12	2.20	0.6	56.4
6	0.0040	0.0040	0.12	2.20	0.6	45.5
7	0.0045	0.0045	0.12	2.20	0.6	38.2
			...
9	0.0550	0.0550	0.12	2.20	0.6	-17.9
10	0.0060	0.0060	0.12	2.20	0.6	-10.3
133	0.0035	0.0035	0.12	1.25	0.8	155.7
134	0.0050	0.0050	0.16	2.20	0.6	140.3
135	0.0050	0.0050	0.17	2.20	0.6	243.1
136	0.0040	0.0040	0.14	2.20	0.6	116.0
137	0.0050	0.0050	0.15	2.20	0.6	95.5

序号	连续相速度	分散相速度	gama值	连续相黏度	分散相黏度	伸长率/%
1	0.0015	0.0015	0.12	2.20	0.6	62.6
2	0.0020	0.0020	0.12	2.20	0.6	74.5
3	0.0025	0.0025	0.12	2.20	0.6	70.2
4	0.0030	0.0030	0.12	2.20	0.6	67.8
5	0.0035	0.0035	0.12	2.20	0.6	56.4
6	0.0040	0.0040	0.12	2.20	0.6	45.5
7	0.0045	0.0045	0.12	2.20	0.6	38.2
			...
9	0.0550	0.0550	0.12	2.20	0.6	-17.9
10	0.0060	0.0060	0.12	2.20	0.6	-10.3
133	0.0035	0.0035	0.12	1.25	0.8	155.7
134	0.0050	0.0050	0.16	2.20	0.6	140.3
135	0.0050	0.0050	0.17	2.20	0.6	243.1
136	0.0040	0.0040	0.14	2.20	0.6	116.0
137	0.0050	0.0050	0.15	2.20	0.6	95.5

项目		连续相流速	分散相流速	连续相黏度	gama值	伸长率
连续相流速	相关系数	1.000	0.454	0.124	0.157	-0.288
连续相流速	显著性	—	0	0.148	0.067	0.001
分散相流速	相关系数	0.454	1.000	-0.077	0.092	-0.509
分散相流速	显著性	0	—	0.368	0.287	0
连续相黏度	相关系数	-0.124	-0.077	1.000	0.033	0.189
连续相黏度	显著性	0.148	0.368	—	0.703	0.027
gama值	相关系数	0.157	0.092	0.033	1.000	0.373
gama值	显著性	0.067	0.287	0.703	—	0
伸长率	相关系数	-0.288	-0.509	0.189	0.373	1.000
伸长率	显著性	0.001	0.002	0.027	0	—
分散相黏度	相关系数	0.066	0.084	-0.34	-0.059	-0.015
分散相黏度	显著性	0.443	0.331	0	0.492	0.860