中国激光, 2006, 33 (7): 953, 网络出版: 2006-08-08
基于人工神经网络激光烧蚀制备PDPhSM基纳米复合薄膜聚合效率的预测
Prediction of Polymerization Efficiency for PDPhSM Matrix Nanocomposite Thin Film Prepared by Laser Ablation Based on Artificial Neural Networks
薄膜 PDPhSM基纳米复合薄膜 激光烧蚀 聚合效率 人工神经网络 thin films PDPhSM matrix nanocomposite thin film laser ablation polymerization efficiency artificial neural network
摘要
为有效缩短脉冲激光烧蚀制备有机硅聚合物聚二苯基硅亚甲基硅烷(PDPhSM)基纳米复合薄膜工艺中繁琐的实验过程,分别采用多层前馈(BP)神经网络和径向基函数(RBF)神经网络对PDPhSM基纳米复合薄膜的制备工艺与聚合效率之间的关系进行建模,并将其运用到聚合效率的预测中去,讨论了激光能量密度、环境压强、靶衬距离、沉积时间和聚合效率之间的关系。克服了以往单因素实验法不能正确反映制备工艺和聚合效率之间复杂的非线性关系的弱点。预测和验证结果均表明实验值和网络预测值之间相对误差都在10%以内,但径向基函数神经网络较多层前馈神经网络能够更精确、更可靠地逼近它们之间的非线性关系。该方法为有效、快捷、经济地开发研制PDPhSM基纳米复合薄膜提供了新的思路和有效手段。
Abstract
In order to shorten the fussy experimental process in synthesizing polydiphenysilylenemethylene (PDPhSM) technology, a back propagation (BP) neural network model and a radial basis function (RBF) neural network model are developed to approach the complex nonlinear relationship between technology parameters and polymerization efficiency for synthesizing PDPhSM matrix nanocomposite thin film respectively. By using the constructed neural network model, the relationship between the technology parameters (laser fluence, ambient pressure, distance between target and substrate, deposition time) and polymerization efficiency is discussed, and the weakness that the nonlinear relationship could not be approached more accurately, effectively by using of single-factor-experiment method is overcomed. Predicted and test results showed that all the relative errors between the desired values and predicted outputs of the network are less than 10%, but the predicted data of RBF model are well acceptable when comparing them to the real test values, hence providing a effective, economical way for synthesizing PDPhSM matrix nanocomposite thin film.
唐普洪, 宋仁国, 柴国钟, 张奇志. 基于人工神经网络激光烧蚀制备PDPhSM基纳米复合薄膜聚合效率的预测[J]. 中国激光, 2006, 33(7): 953. 唐普洪, 宋仁国, 柴国钟, 张奇志. Prediction of Polymerization Efficiency for PDPhSM Matrix Nanocomposite Thin Film Prepared by Laser Ablation Based on Artificial Neural Networks[J]. Chinese Journal of Lasers, 2006, 33(7): 953.