Application and Experimental Study of RBF Neural Network Algorithm in Flow-Induced Vibration of Pipelines
-
摘要: 针对传统流固耦合方法耗时较长,导致核电厂管道设计人员在设计阶段难以开展针对性减振计算的问题,本文采用基于数据驱动的径向基函数(RBF)神经网络算法进行管道流致振动分析。该算法利用数据库中大量节流管件的载荷数据进行训练,可以在短时间完成管道流致振动的定量计算。相对于传统流固耦合方法提高了管道流致振动的分析效率。为验证计算结果的有效性,对不同开度下的球阀和弯头管道进行了实验研究。实验研究发现,由于泵的激励等外界结构振动的存在,当流致振动在总振动中占主导时,本算法计算结果与实验结果比较接近;当外界结构振动在总振动中占主导时,本算法计算结果与实验结果在同一量级且变化规律一致。研究结果表明基于数据驱动的RBF神经网络方法分析管道流致振动是可靠且有效的。
-
关键词:
- 径向基函数(RBF)神经网络 /
- 数据驱动 /
- 流致振动 /
- 输流管道
Abstract: To address the issue of time-consuming traditional fluid-structure interaction methods, which make it difficult for nuclear power plant pipeline designers to perform targeted vibration reduction calculations during the design phase, this study adopts a data-driven radial basis function (RBF) neural network algorithm for pipeline flow-induced vibration analysis. This method can quantitatively calculate the pipeline flow-induced vibration in a short period of time by training on a large amount of load data of throttling fittings in the database. Compared to traditional fluid-structure interaction methods, it greatly improves the computational efficiency of pipeline flow-induced vibration. To validate the calculation results, experimental studies were conducted on on ball valves at different opening degrees and elbow pipes. Due to the presence of external structural vibrations such as pump excitation, experiments have found that when the flow-induced vibration dominates the total vibration, the simulation results are relatively close to the experimental results. When external structural vibration dominates the total vibration, the simulation and experimental results are of the same magnitude and have consistent variation patterns. The results demonstrate that the data-driven based RBF neural network method is reliable and effective for analyzing flow-induced vibrations in pipelines.-
Key words:
- RBF neural network /
- Data-driven /
- Flow-induced vibration /
- Flow pipeline
-
表 1 管道工艺系统参数
Table 1. Parameters of Pipeline Process System
工艺参数 仿真选取的参数值 管道公称
直径DN/mm15、20、25、32、40、50、70、80、
100、125、150压力/MPa 0.10、4.40、6.70、9.99、13.00、29.00、
16.59、19.89、23.18、26.48流速/(m·s−1) 1.00、8.38、15.75、23.13、30.5、37.88、
45.25、52.63、60.00表 2 节点15处流致振动速度的实验峰值和仿真峰值的对比
Table 2. Comparison of Experimental and Simulated Peak Values of Vibration Velocity at Node 15
球阀开度/% 速度实验峰值/(mm·s−1) 速度仿真峰值/(mm·s−1) 轴向仿真值与
实验值百分比/%X(轴向) Y Z X(轴向) Y Z 20 0.5119 0.2620 0.7752 0.509498 0.391249 0.028038 99.53 30 0.5828 0.3627 0.9592 0.503814 0.107594 0.000469 86.45 50 0.6023 0.3639 1.1119 0.41005 0.233034 0.002402 68.08 80 0.4675 0.3128 0.6418 0.185026 0.35819 0.000735 39.58 表 3 节点35处流致振动速度的实验峰值和仿真峰值的对比
Table 3. Comparison of Experimental and Simulated Peak Values of Vibration Velocity at Node 35
球阀开度/% 速度实验峰值/(mm·s−1) 速度仿真峰值/(mm·s−1) 轴向仿真值与
实验值百分比/%X(轴向) Y Z X(轴向) Y Z 20 1.589 0.3643 1.156 1.27463 0.68034 0.0487581 80.22 30 1.530 0.4711 1.230 1.26040 0.10078 0.0008766 82.38 50 1.889 0.4905 1.138 1.02583 0.20178 0.0041445 54.31 80 0.4982 0.4267 0.7043 0.26288 0.33157 0.0013803 52.77 表 4 节点45处流致振动速度的实验峰值和仿真峰值的对比
Table 4. Comparison of Experimental and Simulated Peak Values of Vibration Velocity at Node 45
弯头间距 速度实验峰值/(mm·s−1) 速度仿真峰值/(mm·s−1) 轴向仿真值与
实验值百分比/%X Y Z(轴向) X Y Z(轴向) 1.5D 2.285 2.497 6.034 3.412962 3.91727 5.939722 98.44 3.0D 4.336 3.021 14.65 1.472413 3.76232 14.566210 99.43 5.0D 2.285 2.497 7.496 1.276676 3.85715 5.024510 67.03 -
[1] 王丛林,柴晓明,杨博,等. 先进核能技术发展及展望[J]. 核动力工程,2023, 44(5): 1-5. [2] 郑宽,徐志成,鲁刚,等. 高比例新能源电力系统演化进程中核电与新能源协调发展策略[J]. 中国电力,2021, 54(7): 27-35. [3] 王国法,刘合,王丹丹,等. 新形势下我国能源高质量发展与能源安全[J]. 中国科学院院刊,2023, 38(1): 23-37. [4] 姜乃斌,冯志鹏,臧峰刚,等. 核工程中的流致振动理论与应用[M]. 上海: 上海交通大学出版社,2018: 91-93. [5] 刘诗文,赫荣辉,杨钊,等. 输流管网流致振动特性数值模拟研究[J]. 核动力工程,2022, 43(1): 187-191. [6] 王天富,唐科范,章期文,等. 核电站管路流致振动和噪声的数值模拟研究[J]. 水动力学研究与进展,2021, 36(1): 56-66. [7] HOPFIELD J J. Neurons with graded response have collective computational properties like those of two-state neurons[J]. Proceedings of the National Academy of Sciences of the United States of America, 1984, 81(10): 3088-3092. [8] RUMELHART D E, HINTON G E, WILLIAMS R J. Learning representations by back-propagating errors[J]. Nature, 1986, 323(6088): 533-536. doi: 10.1038/323533a0 [9] BROOMHEAD D S, LOWE D. Multivariable function interpolation and adaptive networks[J]. Complex Systems, 1988, 2: 321-355. [10] POGGIO T, GIROSI F. Networks for approximation and learning[J]. Proceedings of the IEEE, 1990, 78(9): 1481-1497. doi: 10.1109/5.58326 [11] 陈明. MATLAB神经网络原理与实例精解[M]. 北京: 清华大学出版社,2013: 156-204. [12] 杨庆华,占伟涛,吴海伟,等. 基于正交试验、BP神经网络和遗传算法的冷挤压模具优化设计方法[J]. 浙江工业大学学报,2015, 43(3): 251-256. doi: 10.3969/j.issn.1006-4303.2015.03.004 [13] 许开州,胡德金,魏臣隽. 基于正交试验和Vogl快速BP网络的球面磨削工艺优化方法[J]. 上海交通大学学报,2009, 43(12): 1956-1961. [14] 黄鹍,陈森发,亓霞,等. 基于正交试验法的神经网络优化设计[J]. 系统工程理论方法应用,2004, 13(3): 272-275. -