【作者】 周磊；

【导师】 李光耀； 韩旭；

【作者基本信息】 湖南大学 ， 车辆工程， 2006， 硕士

【摘要】 本文应用神经网络方法对驾驶室座椅几何参数的反演做了深入的探索，应用基于改进的多层前向神经网络的结构设计方法，提出了一种用来设计优化驾驶室座椅几何参数的反求方法，并应用到实际的模型里。通过反演计算可以快速得到可行的驾驶室座椅几何设计方案。 1、对矩形封闭空间声辐射进行有限元数值模拟，对比解析解，经计算后结果与解析解比较吻合，证明使用有限元方法计算结构振动声辐射问题是可以保证精度的，并且可以应用在驾驶室内声场分析中。 2、根据某型轿车驾驶室的物理模型，使用ANSYS软件建立驾驶室的三维有限元模型，分析了其相应的声学模态，然后建立声-固耦合声学分析模型。并且进行了谐响应分析得到了相应的响应下驾驶员耳处的声压值分布。 3、考虑声-固耦合作用，使用SYSNOISE分析了整个车室的声场分布情况。 4、不同的座椅几何参数下的驾驶室模型的声场分布有一定的规律。随着其值的改变，在驾驶员耳处的声压值变化是有一定变化的，分析其可能的原因，驾驶室主要是由梁板组合而成，大部分的固有频率是由于壁板的振动引起，当每个振动源的相互影响改变的时候，会引起整个驾驶室声场分布比较大的变化。 5、提出了一种通过点下降来控制线下降的思路，并且比较成功的运用到整个分析过程里。但是研究过程中发现了许多的问题，比如如何选择需要下降的点，如何是使曲线达到优化了，如何增加样本来提高网络的精确度等等。 6、最后得到了在该有限元模型下的一种比较成功的反求模型，并得到了比较理想的优化结果。而对于不同的驾驶室模型，运用该方法，可以训练出其用于反求的神经网络结构，该方法可以很好的对整个驾驶室的座椅几何参数进行反演。更多还原

【Abstract】 This article has researched the inverse method that how to design andoptimize the geometric parameters of seat in cab based on the neuralnetwork. Using the BP algorithm in neural network to design the structure,we put forward an inverse method to design the geometric of seat. At lastwe can attain the scheme by using the inverse method.1 This article has simulated the rectangle close space by FEM, compare with the accurate answer, we can prove that FEM in the analysis of acoustic radiation which generated by vibration of structure is an exact method, and it could be used in the analysis of noise in cab.2 According to the real car, this article has used the software ANSYS to build the 3D FE model, and analyzed the modal of acoustic, then we have established coupling model between the acoustic and structure vibration. Also, we have obtained the distribution of noise near the ear of driver by using the humorous responses analysis.3 According to the coupling effects, we obtained the distribution of noise of whole cab by using the software SYSNOISE.4 We found that there is a rule in the noise distribution under the different geometric parameters. Along with this changes, the noise near the ear of driver will be different, it may caused by the structure characteristic of cab.5 This article brings forward a method that could reduce the noise curve near driver ear by reduces the noise of some specified point. But we found several problems, such as, how to chose the specified points, whether we obtained the optimized curve or not, and how to improve the definition of NN by increase the samples.6 At last, we obtained an inverse model that could be used to design the cab to reduce the noise in it. For different cab, using this method, we can train several inverse models to design and optimize the geometric parameters to make the reduction of noise in it.更多还原