【计算机科学】【2010.12】基于人工神经网络的印刷电路板电磁辐射建模方法

本文为美国托莱多大学（作者：DavidT. Kvale）的电子工程硕士论文，共130页。

本文提出了一种新的用于电子系统电磁干扰/兼容(EMI/EMC)分析的建模方法。通过改变屏蔽孔的数量、改变部分屏蔽PCB迹线的位置、改变PCB互连的位置以及在屏蔽外壳内移动EMI源来研究印刷电路板（PCB）的电磁辐射问题。EMC建模的问题在于，给定PCB配置求解麦克斯韦方程是非常复杂的，许多工程师的最佳思路是广泛遵循仅针对特定解决频率的特定几何结构适用的设计准则，而不是求解相应问题的麦克斯韦方程。在某些情况下，PCB设计和集成电路(IC)的复杂性是难以想象的，以至于对麦克斯韦方程组进行精确求解是不切实际的(例如，对机箱中广泛使用的服务器母板进行建模)。通常，如果被测设备(DUT)没有通过法规认证，则对PCB设计进行EMC修改，但这可能非常昂贵和耗时。这是PCB设计很少修改的许多原因之一，或者如果它们被修改，只会做出较小的变动。

在这篇论文中，将会展示人工神经网络（ANN）能够提供精确的、快速的、低运算量的电磁辐射估计。一个研究案例是采用这种计算工具找到PCB上互连线的一个最优位置。利用人工神经网络进行优化的意义在于，人工神经网络为设计和估算电磁辐射提供了一种快速准确的工具。然而，鉴于人工神经网络的高度可变特性，许多创建人工神经网络的方法都是针对特定EMC示例进行检查和评估。由于ANN模型不需要考虑PCB和电缆结构的详细几何配置，因此，与电磁电路工具相比，计算开销要求显著降低。正如本文所展示的，ANN模型的健壮性、效率、精确性和通用性在电子工业中特别有用，因为大多数制造商都喜欢在新产品中重复使用电路和PCB布局，只是对现有经过时间考验的设计稍加修改。

This thesis introduces a new modeling approach for efficient andaccurate Electromagnetic Interference/Compatibility (EMI/EMC) analysis ofelectronic systems. Printed Circuit Boards’ (PCB) radiated emissions wereinvestigated by varying the number of apertures on a shield, changing thelocations of partially shielded PCB traces, changing the locations of PCBinterconnects, and moving EMI sources within a shielding enclosure. The issuewith EMC modeling is that given the complexity of solving Maxwell’s equations fora given PCB configuration, the best course for many engineers is to broadlyfollow design guidelines that are only true for a specific geometry for aspecific solution frequency instead of solving Maxwell’s equations for a givenproblem. There are cases where the complexity of the PCB design and integratedcircuits (IC) is so extensive, that it is impractical to have an exact solutionof Maxwell’s equations (i.e., modeling a functioning populated servermotherboard within a cavity). Typically, EMC revisions are made to PCB designsif the Device Under Test (DUT) does not pass regulation certification, whichcan be very costly and time consuming. This is one of many reasons why PCBdesigns are infrequently changed, or if they are changed, only small variationsare made. In this thesis, it will be shown that Artificial Neural Networks(ANN) are capable of providing accurate, fast, and computationally lightestimates for radiated emissions. One case study employs this computationaltool to find an optimized location on a PCB for a trace interconnect. Thesignificance of utilizing ANNs for optimization is that ANNs provide a fast andaccurate tool for design as well as for estimating radiated emissions. However,given that ANNs are highly variable, many approaches to ANN creation areexamined and evaluated for specific EMC examples. Since ANN models do notrequire detailed geometrical configurations of the PCB and cable structuresunder consideration, computational overhead requirements are significantlyreduced as compared to electromagnetic and circuit tools. The robustness,efficiency, accuracy, and versatility of ANN models, as demonstrated in thisthesis, are particularly useful in the electronics industry since mostmanufacturers prefer reusing circuits and PCB layouts in new products withminor modifications to the existing time-tested designs.

1 电磁兼容建模简介

2 人工神经网络建模简介

3 穿孔腔体EMI辐射的EM-ANN建模

4 PCB腔体结构的EM-ANN建模

5 PCB内部互连的ANN建模

6 结论与未来工作展望

附录 AutoIt!与NeuroModeler

下载英文原文地址：

http://page2.dfpan.com/fs/8lcj0221e2916668608/

更多精彩文章请关注微信号：