硅橡胶泡沫（SiF）作为一种综合性能优良的高分子材料，被广泛应用于建筑、航空、汽车等多个领域，并已成为各行各业阻燃隔热发展的热点材料。其火灾性能的优化对于确保这些领域的安全至关重要。但传统的SiF阻燃研究方法耗时周期长、研发成本高，且难以满足快速发展的需求。现如今，数据驱动的机器学习（ML）和深度学习（DL）方法已被广泛应用于材料研发、性能预测等领域，但其在SiF材料阻燃性能预测方面的应用研究还较为缓慢。因此，结合ML和DL高速高精度的优势，探究数据驱动在硅橡胶泡沫复合材料（SiFs）阻燃性能中的应用，对加速新型阻燃SiFs的开发具有重要意义。本文提出了一种基于数据驱动技术的SiFs阻燃性能计算方法，旨在高效分析计算SiFs的关键性能指标。重点建立了计算SiFs阻燃性能的多标签深度神经网络模型。主要研究内容及结论如下：

（1）筛选整理了十六烷基三甲基溴化铵（CTAB）、四丁基溴化铵（TBAB）等4种改性剂改性蒙脱土，硼酸锌、水滑石、氢氧化铝等11种阻燃协效体系的硅橡胶泡沫复合材料阻燃数据集。进一步改变硅橡胶制备工艺中不同添加剂和阻燃剂的质量百分比（wt %）制备了不同配比的硅橡胶泡沫复合材料并进行材料性能表征，以此实验补充构建了本研究原始数据集。

（2）读取数据集中各添加剂的特征数据，分析计算了阻燃剂、改性剂与材料基体的相容性。通过数据集划分和预处理，发现LOI的性能增幅比与阻燃剂中ZB wt %之间的皮尔逊相关系数绝对值（Pab）最大为0.37；改性材料中，与CMMT的Pad更高；添加剂中，与催化剂和含氢硅油的质量之间的Pab分别为0.35和0.29。在PHRR和THR分析时，同样与催化剂和含氢硅油之间的Pad较低。因此在制备阻燃硅橡胶泡沫复合材料时，除阻燃剂的含量外，应重点考虑增加催化剂和含氢硅油的添加量。计算模型能够自动且高效地获得不同工艺参数对硅橡胶阻燃性能的影响。

（3）基于数据驱动方法，搭建了SiFs的发泡成型等级、力学性能和阻燃性能计算模型。采用六种ML算法，优化和筛选了计算各性能标签的最优模型。对比DNN与ML对阻燃标签计算的总体精度，结果显示，ML计算不同标签模型的决定系数（R²）为0.61至0.93，而DNN显示了更卓越的准确度（R²为0.754至0.955）。采用设置掩码和损失函数替代传统的删除或填充缺失值方法，并调整不同的模型架构进行DNN模型优化，实现了SiFs阻燃性能多标签的高效计算。此外，提出的基于原样的性能增幅比（ROI）计算方法，R²提高了3.18 %，进一步提升了阻燃性能的计算精度。通过搭建与传统材料表征方法不同的阻燃性能高效计算模型，为硅橡胶泡沫复合材料性能的早期分析、产品设计的优化、以及研发成本的有效降低提供一种新的路径和方法。最后设置反演实验，在实际应用场景中对模型进行进一步测试，确保模型在真实条件下的有效性和稳定性。

As a kind of polymer material with excellent comprehensive performance, silicone rubber foam (SiF) is widely used in many fields such as construction, aviation and automobile, and has become a hot material for the development of flame retardant and thermal insulation in all walks of life. The optimization of its fire performance is crucial to ensure the safety of these areas. However, the traditional SiF flame retardant research method takes a long time, has high research and development costs, and is difficult to meet the needs of rapid development. Nowadays, data-driven machine learning (ML) and deep learning (DL) methods have been widely used in material research and development, performance prediction and other fields, but their application in the prediction of flame retardant properties of SiF materials is still relatively slow. Therefore, combining the advantages of high speed and high precision of ML and DL, exploring the application of data-driven in the flame retardancy of silicone rubber foam composites (SiFs) is of great significance to accelerate the development of new flame retardant SiFs. In this paper, a method for calculating the flame retardancy of SiFs based on data-driven technology is proposed, which aims to efficiently analyze and calculate the key performance indicators of SiFs. A multi-label deep neural network model for calculating the flame retardancy of SiFs was established. The main research contents and conclusions are as follows :

(1) The flame retardant data sets of silicone rubber foam composites with 11 flame retardant synergistic systems such as cetyltrimethylammonium bromide (CTAB), tetrabutylammonium bromide (TBAB), zinc borate, hydrotalcite and aluminum hydroxide were screened and sorted out. The mass percentage (wt %) of different additives and flame retardants in the preparation process of silicone rubber was further changed to prepare silicone rubber foam composites with different ratios and characterize the material properties. The original data set of this study was supplemented by this experiment.

(2) The characteristic data of each additive in the data set were read, and the compatibility of flame retardant, modifier and material matrix was analyzed and calculated. Through data set partitioning and preprocessing, it was found that the maximum absolute value of Pearson correlation coefficient (Pab) between the performance increase ratio of LOI and ZB wt % in flame retardant was 0.37 ; among the modified materials, the Pad with CMMT is higher ; in the additives, the Pab between the mass of the catalyst and the hydrogen silicone oil is 0.35 and 0.29, respectively. In the PHRR and THR analysis, the Pad between the catalyst and the hydrogen silicone oil is also low. Therefore, in the preparation of flame retardant silicone rubber foam composites, in addition to the content of flame retardant, the addition of catalyst and hydrogen silicone oil should be considered. The calculation model can automatically and efficiently obtain the effects of different process parameters on the flame retardancy of silicone rubber.

(3) Based on the data-driven method, the calculation models of foaming grade, mechanical properties and flame retardancy of SiFs were established. Six ML algorithms are used to optimize and filter the optimal model for calculating each performance label. Comparing the overall accuracy of DNN and ML for flame retardant label calculation, the results show that the determination coefficient (R²) of ML for different label models is 0.61 to 0.93, while DNN shows more excellent accuracy (R² is 0.754 to 0.955). The traditional method of deleting or filling missing values is replaced by setting mask and loss function, and different model architectures are adjusted to optimize the DNN model, so as to realize the efficient calculation of multi-label flame retardancy of SiFs. In addition, the proposed performance increase ratio (ROI) calculation method based on the original sample increases R² by 3.18 %, which further improves the calculation accuracy of flame retardant performance. By building an efficient calculation model of flame retardant performance different from the traditional material characterization method, it provides a new path and method for the early analysis of the performance of silicone rubber foam composites, the optimization of product design, and the effective reduction of R & D costs. Finally, inversion experiments are set up to further test the model in practical application scenarios to ensure the validity and stability of the model under real conditions.

[1] 邓军, 李航, 康付如, 等. 硅橡胶复合材料阻燃研究进展[J]. 高分子通报, 2020, (03): 1-8.

[2] Han Y X, Yang L J, Yu Z, et al. Lightweight and flame retardant silicone rubber foam prepared by supercritical nitrogen: the influence of flame retardants combined with ceramicizable fillers[J]. Construction and Building Materials, 2023, 370(17): 130735.

[3] 平凡, 赵宝艳, 包锦标, 等. EPDM对VMQ硅橡胶泡沫发泡行为及力学性能的影响[J]. 材料导报, 2023, 37(06): 222-226.

[4] Montánsa F J, Chinestab F, Rafael Gómez B R, et al. Data-driven modeling and learning in science and engineering[J]. Comptes Rendus Mecanique, 2019, 347(11): 845-855.

[5] 李贺, 徐勇, 段文晖, 等.人工智能和数据驱动的计算模拟[J]. 中国科学:物理学 力学 天 文学, 2024, 54(04): 49-54.

[6] Wang Z, Sun Z H, Yin H, et al. Data-Driven Materials innovation and applications[J]. Advanced Materials, 2022 34(36): 2104113.

[7] He Q Z, Laurence D W, Lee C H, et al. Manifold learning based data-driven modeling for soft biological tissues[J]. Journal of Biomechanics, 2021 117: 110124.

[8] Li X, Li Z Q, Chen Y, et al. An enhanced data-driven constitutive model for predicting strain-rate and temperature dependent mechanical response of elastoplastic materials[J]. European Journal of Mechanics-A/Solids, 2023 100: 104996

[9] Sheng H, Chen F, Li J G, et al. The ceramifying process and mechanical properties of silicone rubber/ammonium polyphosphate/aluminium hydroxide/mica composites[J]. Polymer Degradation Stability, 2016, 126(8): 196-203.

[10] Deng J, Kang F R, Xiao Y, et al. Effects of platinum compounds/superfine aluminum hydroxide/ultrafine calcium carbonate on the flame retardation and smoke suppression of silicone foams [J]. Journal of Applied Polymer Science, 2020, 137(1): 47679.

[11] Kang F R, Wang C P, Deng J, et al. Effects of talc/hollow glass beads on the flame retardancy of silicone foams[J]. Materials Research Express, 2019, 6(9): 095318.

[12] Attia M K. Improving thermal stability and electrical insulation of gamma-irradiated silicone rubber nanocomposites[J]. Journal of Vinyl and Additive Technology, 2020, 26(3): 354-361.

[13] 刘博, 王春雨, 冀贤, 等. 改性蒙脱土对硅胶泡沫复合材料性能影响研究[J]. 消防科学与技术, 2023, 42(06): 808-811+817.

[14] Pang Q T, Kang F R, Deng J, et al. Flame retardancy effects between expandable graphite and halloysite nanotubes in silicone rubber foam[J]. RSC Advances, 2021, 11(23): 13821-13831.

[15] Hong L, Hu X X. Mechanical and flame retardant properties and microstructure of expandable graphite/silicone rubber composites[J]. Journal of Macromolecular Science, Part B, 2016, 55(2): 175-187.

[16] Januszewski R, Dutkiewicz M, Maciejewski H, et al. Synthesis and characterization of phosphorus-containing, silicone rubber based flame retardant coatings[J]. Reactive and Functional Polymers, 2018, 123: 1-9.

[17] Zhang Y J, Zeng X R, Li H Q, et al. Zirconium phosphate functionalized by hindered amine: A new strategy for effectively enhancing the flame retardancy of addition-cure liquid silicone rubber[J]. Materials Letters, 2016, 174(1): 230-233.

[18] Xu W Z, Zhang Y, Fan L J, et al. Improvement the flame retardancy, thermal conductivity and mechanical properties of silicone rubber via silane coupling agent modified CNTs‐β‐FeOOH[J]. Journal of Applied Polymer Science, 2023, 140(12): 53656.

[19] Liu J Q, Xin Z X. Effect of sepiolite on properties of silicone rubber/melamine/ starch/sepiolite flame retardant composites[J]. Journal of Applied Polymer Science, 2023, 140(9): 10.

[20] Yao L, Xu W Z, Ding D, et al. Core-shell carbon nanotubes/cobalt copper hydroxide hybrid/silicone rubber composite: flame retardancy and antistatic properties[J]. Iranian Polymer Journal, 2023, 32: 1-14.

[21] Chai W H, Su X Y, Xia Y R, et al. Construction of a novel halogen-free and phosphorus-free and synergistic flame retardant system with low addition in simultaneously improving the flame retardancy, water resistance and mechanical properties for silicone rubber[J]. Applied Clay Science, 2023, 238: 106926.

[22] 马砺, 杨昆, 蔡周全, 等. 含铂催化剂的硅胶泡沫阻燃性能实验研究[J]. 材料导报, 2018, 32(S2): 286-289.

[23] 邓军, 康付如, 吴长林, 等. 环保型硅胶泡沫阻燃抑烟及热分解特性[J]. 高分子材料科学与工程, 2018, 34(12): 90-94+100.

[24] 邓军, 周廷斌, 康付如, 等. 纳米NiO对硅橡胶泡沫阻燃抑烟及热稳定性的影响[J].中国安全生产科学技术, 2023, 19(08): 5-11.

[25] 邓军, 李航, 康付如. 硼酸锌和镁铝水滑石对硅橡胶泡沫性能的影响[J].高分子材料科学与工程, 2020, 36(12): 42-48.

[26] 商珂, 林贵德, 姜慧婧, 等. 助熔剂对可陶瓷化阻燃室温硫化硅橡胶泡沫性能的影响[J].复合材料学报, 2023, 40(07): 4060-4071.

[27] 姚远, 王乐, 崔正华, 等. 辐照交联无卤阻燃可陶瓷化硅橡胶复合材料的构建及性能研究[J]. 同位素, 2023, 36(05): 508-515.

[28] 刘伦洋, 丁芳, 李云琦. 高分子材料大数据研究: 共性基础、进展及挑战[J]. 高分子学报, 2022, 53(06): 564-580.

[29] 宫祥瑞, 蒋滢. 机器学习在高分子材料基因组研究中的进展与挑战[J]. 高分子学报, 2022, 53(11): 1287-1300.

[30] Sun Y T, Bai H Y, Li M Z, et al. Machine learning approach for prediction and understanding of glass-forming ability[J]. The Journal of Physical Chemistry, 2017, 8(14): 3434-3439.

[31] 李锐. 高丰度稀土快淬磁体的永磁性能、机理和机器学习研究[D]. 中国科学院大学:中国科学院物理研究所, 2019.

[32] 黄乐, 谭锋, 肖风亮. 基于BP神经网络算法的橡胶配方性能预测软件[J]. 橡胶工业, 2018, 65(08): 944-947.

[33] Lu L, Dao M, Kumar P, et al. Extraction of mechanical properties of materials through deep learning from instrumented indentation[J]. Proceedings of the National Academy of Sciences of the United States of America, 2020, 117(13): 7052-7062.

[34] 游洋, 杜婉, 李惟驹, 等. 基于机器学习方法的二维材料带隙预测[J]. 上海大学学报(自然科学版), 2020, 26(05): 824-833.

[35] Pathrudkar S, Yu H M, Ghosh S, et al. Machine learning based prediction of the electronic structure of quasi-one-dimensional materials under strain[J].Physical Review B, 2022, 105(19): 195141.

[36] Li W D, Suhayb M K, Thangavelu L, et al. Implementation of AdaBoost and genetic algorithm machine learning models in prediction of adsorption capacity of nanocomposite materials[J]. Journal of Molecular Liquids, 2022, 350: 118527.

[37] Gan Y, Wang G J, Zhou J, et al. Prediction of thermoelectric performance for layered IV-V-VI semiconductors by high-throughput ab initio calculations and machine learning[J]. NPJ Computational Materials, 2021, 7(1): 1609-1618.

[38] Roy D, Mandal S C, Pathak B. Machine learning-driven high-throughput screening of alloy-based catalysts for selective CO2 hydrogenation to methanol[J]. ACS Applied Materials & Interfaces, 2021, 13(47): 56151-56163.

[39] 何兴群, 付华栋, 张洪涛, 等. 机器学习辅助高性能银合金电接触材料的快速发现[J]. 金属学报, 2022, 58(6): 816-826.

[40] Liu Y M, Yan W S, Zhu H, et al. Study on bandgap predications of ABX3-type perovskites by machine learning[J]. Organic Electronics, 2022, 101: 106426.

[41] Zhang H, Li H Y, Xin H S, et al. Property prediction and structural feature extraction of polyimide materials based on machine learning.[J]. Journal of Chemical Information and Modeling, 2023, 63(17): 5473-5483.

[42] Liu Y, Wang S Y, Yang Z W, et al. Auto-MatRegressor: liberating machine learning alchemists[J]. Science Bulletin, 2023, 68(12): 1259-1270.

[43] Li X H, Guo M M, Zhang R R, et al. A data-driven prediction model for maximum pitting corrosion depth of subsea oil pipelines using SSA-LSTM approach[J]. Ocean Engineering, 2022, 261: 112062.

[44] 关少恒.基于数据驱动的筑坝颗粒材料多尺度计算方法研究[D]. 湖北: 武汉大学, 2023.

[45] Niaki M H, Ahangari M G, Pashaian M, et al. A material-independent deep learning model to predict the tensile strength of polymer concrete[J]. Composites Communications, 2022, 36: 101400.

[46] Liu W B, Wang Z D, Liu X H, et al. A survey of deep neural network architectures and their applications[J]. Neurocomputing, 2017, 234: 11-26.

[47] Amor N, Noman MT, Petru M. Prediction of methylene blue removal by nano TiO2 using deep neural network[J]. Polymers, 2021, 13(18): 3104.

[48] Yadav R, Panwar R. AI-based estimator for computational discovery and synthesis of customized microwave absorbing materials[J]. Journal of Applied Physics, 2023, 134(9): 094901.

[49] Asghari V, Leung Y F, Hsu S C. Deep neural network based framework for complex correlations in engineering metrics[J]. Advanced Engineering Informatics, 2020, 44: 101058.

[50] 彭仁苗, 徐鹏鹏, 赵一默, 等. 深度神经网络和高光谱显微图像的二维材料纳米片识别[J]. 光谱学与光谱分析, 2022, 42(06): 1965-1973.

[51] 彭凡, 邹司农, 任毅如. 基于深度学习的复合材料螺栓连接失效预测[J]. 机械强度,2023, 45(02): 447-453.

[52] 孙立望, 李洪, 汪鹏君, 等. 利用神经网络识别高分子链在表面的吸附相变[J]. 物理学报, 2019, 68(20): 60-66.

[53] Kamble A, He S H, Howse J R, et al. Exploiting the use of deep learning techniques to identify phase separation in self-assembled microstructures with localized graphene domains in epoxy blends[J]. Computational Materials Science, 2023, 229: 112374.

[54] Kuenneth C, Schertzer W, Ramprasad R. Copolymer informatics with multitask deep neural networks[J]. Macromolecules, 2021, 54(13): 5957-5961.

[55] So C, Kim Y S, Park J H, et al. Data-driven design of electrically conductive nanocomposite materials: a case study of acrylonitrile–butadiene–styrene / carbon nanotube binary composites[J]. Advanced Intelligent Systems, 2023, 5(2): 2370008.

[56] Jiang Z Y, Hu J J, Marrone B L, et al. A deep neural network for accurate and robust prediction of the glass transition temperature of polyhydroxyalkanoate homo- and copolymers[J]. Materials, 2020, 13(24): 5701.

[57] Ly H B, Nguyen T A, Mai H V T, et al. Development of deep neural network model to predict the compressive strength of rubber concrete[J]. Construction and Building Materials, 2021, 301(27): 124081.

[58] Zazoum B, Triki E, Bachri A. Modeling of mechanical properties of clay-reinforced polymer nanocomposites using deep neural network[J]. Materials, 2020, 13(19): 348-353.

[59] Niaki M H, Pashaian M, Ahangari M G. Experimental assessment of the efficiency of deep learning method in predicting the mechanical properties of polymer concretes and composites[J]. Journal of Building Engineering, 2023, 76: 107107.

[60] Niaki M H, Ahangari M G, Izadit M, et al. Evaluation of fracture toughness properties of polymer concrete composite using deep learning approach[J]. Fatigue Fracture of Engineering Materials Structures, 2023, 46(2): 603-615.

[61] Nguyen T, Bavarian M. A Machine Learning framework for predicting the glass Transition temperature of homopolymers[J]. Industrial & Engineering Chemistry Research, 2022, 62(34): 12690-12698.

[62] 王九龙, 盛俊杰, 张凯, 等.基于深度LSTM神经网络的泡沫橡胶材料剩余寿命预测方法[P]. 中国专利: CN202210573688.7, 2022-09-06.

[63] GB/T 528-2009, 硫化橡胶或热塑性橡胶拉伸应力应变性能的测定[S]. 北京: 中国标准出版社, 2009.

[64] GB/T 2406.2-2009, 塑料用氧指数法测定燃烧行为第2部分:室温试验[S]. 北京: 中国标准出版社, 2010.

[65] GB/T 16172-2007, 建筑材料热释放速率试验方法[S]. 北京: 中国标准出版社, 2008.

[66] GB/T 8627-2007, 建筑材料燃烧或分解的烟密度试验方法[S]. 北京: 中国标准出版社, 2008.

[67] 任海生, 王静波, 李象远. 燃烧动力学平台CDS网络版, 四川大学燃烧动力学中心, 2021.9, http://cds.scu.edu.cn/.

[68] 李航.改性硅系阻燃剂/硼酸锌/硅橡胶泡沫阻燃抑烟研究[D].西安:西安科技大学, 2022.

[69] 陈少康, 唐怡潇, 刘志超, 等. 基于正交实验的阻燃硅胶泡沫制备及优化研究[C]. 中国消防协会. 2018中国消防协会科学技术年会论文集, 2018: 162-164.

[70] 巩泽文, 贾建称, 许耀波, 等. 基于测井信息的煤层顶板水平井抽采煤层气技术[J]. 天然气工业, 2021, 41(02): 83-91.

[71] Breima L, Friedman J, Olshen R A, et al. Classification and regression trees[J]. Chapman and Hall, 1984.

[72] Breima L. Random forests[J]. Machine Learning, 2001, 45: 5-32.

[73] Han T, Jiang D X, Zhao Q, et al. Comparison of random forest, artificial neural networks and support vector machine for intelligent diagnosis of rotating machinery[J]. Transactions of the Institute of Measurement and Control, 2018, 40(8): 2681-2693.

[74] Natekin A, Knoll A. Gradient boosting machines, a tutorial[J]. Front Neurorobotice, 2013, 7(21): 1-21.

[75] 刘廷滨, 黄滔, 欧嘉祥, 等. 基于ANN和XGB算法的锈蚀钢筋混凝土高温粘结强度预测方法[J/OL]. 工程力学, 2024-04-19, 1-11.

[76] Friedman J H. Greedy function approximation: a gradient boosting machine[J].The Annals of Statistics, 2001, 29(5): 1189­1232.

[77] Smola A J, Schölkopf B. A tutorial on support vector regression[J]. Statistics and Computing. 2004, 14(3): 199-222.

[78] Cover T M, Hart P E. Nearest neighbor pattern classification[J]. IEEE Transactions on Information Theory, 1967,13(1): 21-27

[79] 王青林. 基于机器学习算法的铸造高温合金材料性能预测[D]. 武汉: 华中农业大学, 2022.

[80] 马天寿, 张东洋, 陈颖杰, 等. 基于神经网络模型的水平井破裂压力预测方法[J]. 中南大学学报(自然科学版), 2024, 55(01): 330-345.

[81] 安鹏, 曹丹平. 基于深度学习的测井岩性识别方法研究与应用[J]. 地球物理学进展, 2018, 33(03): 1029-1034.

[82] Uszkoreit J, Parmar N, Polosukhin L. et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30: 1-10.

[83] Wu B Y, Liu Z L, Wang S F, et al. Multi-label learning with missing labels[C]. International Conference on Pattern Recognition, 2014: 1964-1968.

[84] Bucak S S, Jin R, Jan A K, et al. Multi-label leanmg with incomplete class assign[J]. In IEEE Conference on Computer Vision and Pattern Recognition, 2011: 2801-2808.

[85] 郭宝林.标签有限或缺失情形下的多标签学习方法研究[D]. 长沙: 国防科技大学, 2018.

[86] Mazumder R, Hastie T, Tibshirani R.Spectral regularization algorithms for learning large incomplete matrices [J]. Journal of Machine Learning Research, 2010, 11: 2287-2322.

[87] 罗刚, 罗嗣慧, 肖立志, 等. 机器学习在核磁共振测井数据处理中的应用进展[J]. 测井技术, 2023, 47(06): 643-652.