【作者】 赵永杰；

【导师】 万德安；

【作者基本信息】 同济大学 ， 机械制造及其自动化， 2006， 硕士

【摘要】 汽车变速箱是汽车的关键部件,它的性能好坏直接影响到汽车性能。如果汽车变速箱质量不过关,小则噪音大,运行不平稳:大则出现故障,引发交通事故,甚至致人死亡。因此迫切需要开发一套变速箱出厂质量检验及故障诊断系统。这套系统不仅要对出厂的变速箱质量进行检验,更要对变速箱进行故障诊断,辨别故障类型并找出缘由,从而指导维修并为变速箱的改进提供依据。 本课题是上海汽车协会及上海汽车齿轮总厂共同资助的项目《SAGW汽车变速箱的故障诊断仪研制》。它以变速箱故障原理分析、数字信号处理为基础,运用人工智能方法,根据故障特征参量对典型故障模式进行识别。 本文对人工智能方法进行了深入的研究,研究了粗糙集理论及基于粗糙集理论的决策系统约减方法,构建了最优决策系统,简化了诊断结构,提高了诊断效率。在最优决策系统基础上,研究了神经网络和模糊系统的故障诊断方法。研究过程中发现模糊系统缺乏自学习能力,隶属度函数和模糊规则的选取带有一定的主观性且依赖于专家;神经网络所获得的输入/输出关系无法用容易被人接受的方式表示出来,且无法应用专家的知识积累,难以保证诊断精度。针对上述缺点,通过对神经网络和模糊系统的结合方式的研究,提出了一种基于模糊神经网络的变速箱故障诊断方法,并将其应用于变速箱的故障诊断。实验结果表明,与常用的神经网络和摸糊系统诊断方法相比,该方法能够弥补模糊和神经网络分别单独应用时所存在的不足,具有更高的诊断准确率。在旋转机械故障诊断领域具有较好的应用前景。 实际运用方面,开发了汽车变速箱故障诊断系统。该系统综合运用了分布式数据库及局域网等知识,实现了数据存储共享及远程访问。在变速箱生产组装流水线上进行的大量诊断测试实验表明,该系统方便、性能稳定、精度高,具有较高的推广应用性。更多还原

【Abstract】 Auto gearbox is the main part of vehicle. It influences the performance significantly. It may be noisy or cause traffic accident for serious if it’s not qualified. So it is important and urgent to develop a kind of fault diagnosis system to verify auto gearboxes, not only to find out whether it has faults, but also find the faults’ source.This subject is based on the project "SAGW Auto Gearbox Fault Diagnosis System Development". It is supported by ShangHai Auto Association and ShangHai Gearbox Company. It integrates auto gearbox fault theory analysis area, digital signal processing area and artificial intelligence area etc.In this paper, artificial intelligence is analyzed in depth. Researched on rough set theory, considering the ability of rough set theory to analysis data, diagnosis decision system is reduced to find the key conditions for diagnosis, so that the cost can be reduced and the efficiency can be raised. Researched on the fault diagnosis method of neural network and fuzzy system. Fuzzy system lacks self-study ability and its membership functions and fuzzy rule are chosen by experts subjectively;Input/output relationship obtained by neural network can’t be expressed in acceptable way, Neural network is hard to learn from experts. So a fault diagnosis method base on fuzzy neural network is put forward and applied to the fault diagnosis of auto gearbox. The experimental result indicates that this method, compared with the common one, can make up the shortcoming of the single-handed application of fuzzy classification or neural network. Moreover, it owns the better validity and popularity. It has a good application prospects in rotating machinery fault diagnosis.An auto gearbox fault diagnosis system is developed, it integrates distributed database knowledge, intranet knowledge etc. The performance of this system on manufacturing line indicates that this system has high precision and stability.更多还原