某型四余度舵机的神经网络故障诊断模型

某型四余度电液舵机是飞机平尾操纵系统的关键部件，是机、电、液高度耦合的复杂系统。本文根据实际需求，将系统分为元件、部件及系统三种不同层次的故障诊断问题，运用神经网络故障诊断理论，分别建立其BP神经网络诊断模型，并介绍了神经网络故障诊断系统的实现。     

RBF级联型神经网络在挤出吹塑制品质量控制中的应用

通过对挤出吹塑工艺过程建立了成形工艺参数与制品质量关系的RBF(径向基函数)级联型神经网络,经训练,该网络可预测各成形工艺参数对制品质量的影响,从而可实现对制品质量的有效控制,并实验验证了其正确性和可行性。     

利用神经网络技术对板料成形摩擦系数进行预测

针对板料拉深成形过程中摩擦系数较难确定的客观情况,结合人工神经网络所表现出来的良好特性,提出了一种将神经网络技术、数值模拟技术相结合,通过建立摩擦系数与板料成形厚度的非线性映射关系来反求摩擦系数的方法。实践证明,该方法是可行的。     

基于小波神经网络信息融合的板材拉深故障诊断研究

总结了板材拉深试冲时的综合故障及原因,应用小波技术加混合式数据融合方法,将数据级、特征级和决策级融合通过神经网络方法综合在一起,对板材拉深试冲时的缺陷进行智能诊断,描述了小波神经网络的建模过程,探讨了通过多源互补信息减少故障诊断系统不确定性的优化方法,并给出了诊断结果。     

基于人工神经网络的拉形回弹预测技术研究

以蒙皮拉形为对象,探索了人工神经网络技术在拉形回弹预测中的应用。研究了利用神经元网络预测拉形回弹的关键技术,将人工智能技术和拉形数字仿真技术有机结合,建立了拉形回弹预测的人工神经网络模型,并验证了预测模型的有效性,为推动知识工程在飞机板金精密成形中的应用积累了经验。     

The integrated intelligent system for welding seam error and penetration depth identification

A integrated intelligent system for seam tracking and penetration control is given. The system received information of welding seam error and penetration depth from only one sensor, then, it realized seam tracking and penetration control simultaneously. This paper introduces constitution of the system, methods of information recognition, design of the neuralfuzzy controller and results practically.     

基于人工神经网络的铜合金形变热处理工艺和性能

利用神经网络对Cu-Cr-Zr合金变形量、时效温度和时间与硬度和电导率样本集进行训练和学习，采用改进的BP网络算法-Levenberg—Marquardt算法，建立了形变热处理工艺BP神经网络模型，得出了具有较高综合性能的最佳工艺参数：在80％变形量，450-480℃，2～5h形变热处理条件下，硬度和电导率分别可达HV150～157和74％～77％(IACS)。     

Surface wavelength content based clustering using neural networks for manufacturing process mapping

Manufacturing processes produce a unique texture on the surface that serves as a fingerprint of the process. It is possible to provide feedback to the process by studying the surfaces carefully. Analytical techniques such as Fourier analysis and digital filters are commonly used to characterize surface profiles. Parameters extracted from filtered profiles are monitored to detect variations in the process. This requires the development of an inference engine to map metrology parameters to manufacturing process parameters. This paper presents an artificial neural network (ANN) based inference engine for providing process feedback with surface finish input. Parameters such as R_a and W_a as well as advanced wavelet based features are extracted from surface finish data collected from a crankshaft manufacturing line and fed as input to the neural network. This input is then clustered using a competitive neural network trained in unsupervised mode. The resulting clusters are analyzed and discussed. The network is then tested with new data to evaluate the quality of the clusters previously generated and to demonstrate the applicability of this technique for detecting process variations.     

A354合金热处理制度的人工神经网络模型

采用人工神经网络方法,研究了固溶温度、固溶时间、时效温度和时效时间对A354合金热处理性能的影响,建立了A354合金热处理制度的人工神经网络模型.     

基于遗传算法的Al-5%Cu合金电脉冲孕育处理参数优化

以实验为基础,利用神经网络和遗传算法优化Al-5%Cu合金的电脉冲孕育处理工艺参数。神经网络的输入参数为脉冲电压、脉冲时间和电脉冲孕育处理时熔体温度,输出参数是合金凝固组织的晶粒度。在神经网络训练的基础上,采用遗传算法优化神经网络的输入参数。结果表明,神经网络和遗传算法的组合建模获得了较好的优化结果。