基于高分辨CT扫描和深度学习的典型烟草提取物精制滤膜的三维构效关系  

Three-dimensional structure-property relationships for filter membranes used to refine tobacco extracts based on high-resolution CT scanning and deep machine learning

作　　者：管明婧 张劲 周顺[1] 张晓宇[1] 王孝峰[1] 曹芸 田慧娟 丁乃红[1] 李延岩 陈韦剑 李鲁[1] 付硕 姜余婷 宋晓辉 GUAN Mingjing;ZHANG Jin;ZHOU Shun;ZHANG Xiaoyu;WANG Xiaofeng;CAO Yun;TIAN Huijuan;DING Naihong;LI Yanyan;CHEN Weijian;LI Lu;FU Shuo;JIANG Yuting;SONG Xiaohui(Key Laboratory of Combustion&Pyrolysis Study of CNTC,China Tobacco Anhui Industrial Co.,Ltd.,Hefei 231200,China;School of Materials Science and Engineering,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]安徽中烟工业有限责任公司烟草行业燃烧热解研究重点实验室,合肥市231200 [2]合肥工业大学材料科学与工程学院,合肥市230009

出　　处：《烟草科技》2025年第4期11-19,共9页Tobacco Science & Technology

基　　金：中国烟草总公司加热卷烟研制重大专项项目“颗粒型加热卷烟产品生产关键控制技术及效能提升研究”[110202201046(XX-05)];安徽中烟工业有限责任公司科研项目“加热卷烟用增香胶囊的开发及应用研究”(2022155);“加热不燃烧烟草制品用特种复合滤棒开发及应用研究”(2020130)。

摘　　要：为明确滤膜三维结构与烟草提取物靶向精制之间的构效关系,定量计算了典型滤膜在过滤前后结构的变化;结合对烟草提取物显效成分的分析,剖析了滤膜微观结构及其对烟草提取物分离精制的作用机制。通过DLS及GC-MS分析烟草提取物分离前后的粒径及成分变化,以及高分辨CT扫描技术对滤膜进行结构表征,结合U-Net神经网络算法进行深度学习,定量计算了滤膜参数。结果表明:①较PA膜而言,PTFE膜能够截留烟草提取物中更多的固相物及烟碱。②智能分割计算结果显示,过滤后PA膜的孔隙率降低4.4%、厚度增加0.6 mm、纤维平均直径减小154.73μm,可能是由于膜孔隙被堵塞使膜纤维受到挤压。③孔径0.22μm PTFE膜纤维的平均直径为1694.0μm,孔隙率为42.7%,比PA膜的纤维平均直径及孔隙率更小,致密的纤维使PTFE膜对固相物的截留效率更高。本研究中提供了一种滤膜三维结构的定量研究方法,将滤膜微观结构的表征结合深度学习计算,与膜分离前后样品成分的变化相对应,可为烟草提取物的靶向膜分离提供理论参考。To clarify the structure-property relationships between the three-dimensional structure of filter membranes to achieve the targeted refinement of tobacco extracts,the structural changes of typical filter membranes before and after tobacco filtration were quantitatively examined and calculated.By combining the analysis of the active components in the tobacco extracts,the microstructure of the filter membranes and their mechanisms of separation and refinement of tobacco extracts were dissected.Dynamic light scattering(DLS)and gas chromatography-mass spectrometry(GC-MS)analyses were used to characterize the changes in particle sizes and components of the tobacco extracts before and after membrane separation.High-resolution CT scanning technique was used for structural characterization of the filter membranes.In addition,the deep learning technique was combined with the U-Net neural network algorithm to quantitatively calculate the structural parameters of the filter membranes.The results showed that:1)Compared with PA membrane,PTFE membrane could retain more solid phase materials and nicotine from tobacco extracts.2)The calculations by intelligent segmentation showed that the porosity of PA membrane decreased by 4.4%,the thickness increased by 0.6 mm,and the average fiber diameter decreased by 154.73μm after filtration,which was caused by membrane fiber squeezing due to membrane pore blockage.3)The PTFE membrane with a pore size of 0.22μm had an average fiber diameter of 1694.0μm and a porosity of 42.7%,which were smaller than those of the PA membrane.The dense fiber alignment of the PTFE membrane resulted in greater retention of solid phase materials.A quantitative method to analyze the three-dimensional structure of the filter membranes was proposed by integrating the microstructural characterization of the filter membrane with deep learning and correlating it with the component changes in the extract sample before and after filtration,which provided a theoretical reference for targeted filter separation of tob

关 键 词：膜分离 烟草提取物 高分辨CT扫描 U-Net神经网络 深度学习 孔隙率 构效关系 

分 类 号：TS411[农业科学—烟草工业]