基于人工免疫的异步电机故障检测与诊断技术研究

【作者】 廖章珍；

【导师】 陈强；

【作者基本信息】 江西理工大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 异步电机广泛应用于工业生产等部门,一些大型电机一旦发生故障,就可能引起生产线上其它部件甚至整个系统的停产或者损坏,有时甚至会造成严重的事故,这就给异步电机的故障诊断技术提出了非常高的要求。基于知识的智能化故障诊断方法是提高故障监测与诊断系统综合性能的有效途径。人工免疫网络和算法在故障检测与诊断方面有其独特的优点,本文尝试将其应用于异步电机故障监测与诊断问题。从电机正常和三种常见故障状态下的振动幅值信号和定子电流信号中采集特征参数,将其用于系统检测器训练。针对现有基于人工免疫的反面选择算法存在的以下不足:该算法产生的检测器是随机的,不可避免产生大量无效检测器,且检测器不能够很好的覆盖非己空间。本文引入适合异步电机故障检测与诊断的改进型反面选择算法,使产生的检测器不能检测自己空间,只捕获非己空间的特征;设定匹配阈值,依据部分匹配原则取消能检测共有特征空间且与两个以上故障模式相匹配的检测器,但保留与各故障模式独有空间相匹配且只能检测一种故障的检测器。按已设定的匹配阈值,将训练好的检测器集与自己集相比较来确定设备发生了何种故障;若没有检测器与自己串匹配,则尝试改变初始设定的匹配阈值重复上一步骤直到有检测器与自己串匹配。将改进型反面选择算法应用于异步电机故障检测与诊断问题,仿真结果表明其对自己串和非己串数据的异常检测准确率较高。借鉴免疫系统的克隆选择原理及已有的克隆选择算法,结合故障诊断的实际应用,研究了既具有故障诊断能力又具有对故障样本的连续学习功能的自适应故障诊断方法。尝试性引入了基于克隆选择原理的故障诊断模型,该模型可以实现对训练样本的学习、记忆、标识,同时还可以实现对未出现过的抗原样本进行在线诊断和动态再学习,通过笼型异步电动机故障诊断实例仿真研究验证了所引入方法和模型的有效性。更多还原

【Abstract】 Induction motor is widely used to industrial production departments. But once some large-scale motors go wrong, there may be cause to other parts in production-line even the overall system be stopped production or ruined, at times it will go to the length of making severe accidents. This makes the fault diagnosis technology of induction motor to put extraordinary high request. The method of intellectualized fault diagnasis based on knowledge is efficient path for advancing overall performance of fault monitoring and diagnostic system.Both artificial immunity networks and algorithms in aspect of fault detection and diagnosis have their unique advantage,and this paper attempts to apply them to the problem of induction motor fault monitoring and diagnosis. Collect characteristic parameters from motor vibration amplitude value signal and stator current signal in the state of normal and three kind of constant faults, and put them use to system detectors training.Aim at the available negative selection algorithm based on artificial immunity in being following short: The brought detectors of the algorithm is random, so inevitability produce a great deal invalidation detectors, even the detectors are out of condition covering non-self space well enough; The paper introduces the generation negative selection algorithm which suits fault detection and diagnosis of the induction motor: Make the produced detectors can not detect self space, only capture non-self spacial characteristics; Set match threshold, upon partial matching principle cancel the detectors which can detect the community feature space and match with being more than two faultpatterns. But reserve the detectors which match with each faultpattern possess singly space and only detect a sort of fault. Upon the set match threshold, compare the detectors multitude trained well with self multitude to determine the device has occur which fault; If there is no detecter matched with self train, then attempt to vary initial setting match threshold and repeat the back step till there will be detectors matched with self train. Apply generation negative selection algorithm to the problem of induction motor fault detection and diagnosis, the simulation result proves that the accuracy rate of anomaly detection towards self train and non-self train data is rather high.Borrowing ideas from clonal selective principle of immune system and the available clonal selective algorithms, combining the practical application of fault diagnosis, adapting fault diagnosis means is studied which possess fault diagnosis ability as well as serial learning function for fault samples. It has introduced tentatively fault diagnosis model based on clonal selective principle, which might realize study、memory、identification for train samples, at the same time can also realize online diagnosis and dynamic restudy to new antigen samples, and through the cage type induction motor experiments it has demonstrated the validity of methods and models this paper introduced.更多还原