新型集成神经网络在故障诊断中的应用分解.docxVIP

本科毕业设计（论文）题目：新型集成神经网络在故障诊断中的应用学 院 自动化科学与工程学院 专 业 自动化 学生姓名 沈麟 学生学号 201130612280指导教师 罗家祥提交日期 2015 年 月 日摘要为了给工业生产中大型复杂关键设备的状态检测与故障诊断提供更为可靠的方法，近几十年来，国内外学者一直致力于将先进且自动化程度高的人工智能技术运用于其中。在研究大量相关文献的基础上，本文主要提出了一种新型的集成神经网络，并最终将其应用对钢板表明损伤的诊断当中。首先，本文从故障诊断问题出发，介绍了其定义、流程以及常用方法。然后，文章系统的给出了与人工神经网络相关的理论知识。其中包括神经网络的基本概念，BP网络算法原理，神经网络的泛化能力问题，主元分析法对网络结构的优化，粒子群算法对网络初始权值的优化，Bagging集成网络，Adaboost集成网络等。紧接着，文章详细的介绍了新型集成神经网络的设计思想与实现方法，这是本文的重点。最后，为验证这种网络的性能，本文进行了大量实验，以将其与各种传统的单个或集成的神经网络进行对比。实验结果表明，对于单个网络而言，结构优化的方法能使网络性能有所改善，而权值优化却对网络并无改善作用；对集成网络而言，从总体上看，它们的泛化能力要优于单个网络，同时，本文所提的新型集成网络表现又要优于其它传统的集成网络。关键词：集成神经网络；故障诊断；泛化能力AbstractScientists have been inventing advanced and automatic artificial intelligent techniques that could be implemented for more reliable methods in condition detection and fault diagnosis in large-scale, crucial, and complex mechanical systems in the area of industry for the past decades. After many researches and studies on the subject, this paper proposed a new type of ensemble artificial neural network that could be applied to the diagnosis of flawed appearances of the steel. First of all, this paper starts off with problems of fault diagnosis and then introduces its definition, procedures and methods. Secondly, this article mentions the relevant theories of artificial neural networks precisely which includes its basic concepts, the algorithm principle of BP network, generalization ability of machine learning, improved network structure with PCA, optimized initial weights with PSO, ensemble network of Bagging and Adaboost. After that, the article describes the design and implementation of a new type of ensemble neural network in details, which is the main portion of the paper. Lastly, in order to test the proposed network, many experiments have been done to compare it with those traditional single or ensemble neural networks. The result demonstrates that optimized network structure is more successful for single networks than optimized initial weigh