基于数据驱动的电机轴承故障诊断方法研究

发布时间：2018-03-29 16:02

本文选题：实验数据　切入点：集合经验模态分解　出处：《沈阳理工大学》2017年硕士论文

【摘要】：随着现代工业规模的不断扩大和系统复杂度的日益提高,电机轴承被越来越多的应用在工业生产中,因此,对电机轴承进行有效精确的故障诊断便成为了一项十分有意义的科研课题。本文在基于数据驱动的基础上,提出基于集合经验模态分解(EEMD)-改进的局部均值分解(ILMD)-改进万有引力搜索算法(IGSA)-增量概率神经网络(IPNN)的电机轴承故障诊断集合方法,以此提高电机轴承故障诊断的精确性。本文全部的试验数据都来自美国凯斯西储大学轴承实验中心。而在工业生产中,由于电机轴承的工作环境往往十分嘈杂,再加上受到其他设备本身振动的干扰,使得其振动信号含有噪声,因此,要对采集的数据进行预处理,减低噪声。传统的降噪方法不能很好的对非平稳、非线性的数据进行降噪,本文采用对于非平稳非线性信号有很强分解能力的EEMD算法,通过计算相关系数并设定阈值,对数据进行降噪预处理。故障提取方面,针对LMD存在的端点效应问题,提出改进LMD方法对数据进行分解以改善端点效应影响,并计算乘积函数分量的样本熵和能量作为特征参数,组成故障特征向量,作为故障诊断神经网络的输入。故障诊断方法采用基于统计原理的前馈型神经网络IPNN,IPNN不需要设置初始权值,训练简洁,分类能力强。由于故障诊断中,网络模型的参数会对诊断性能有着重大影响,故本文采用基于时变权重和边界变异的改进GSA优化算法对网络模型的阈值进行优化,以改善标准GSA算法收敛速度较慢且容易陷入到局部最优状态等缺点,提高分类结果的精确性。通过理论研究和实验结果可以表明,本文提出的基于数据驱动的EEMD-ILMD-IGSA-IPNN电机轴承故障诊断集合方法诊断性能良好,能够有效的对电机轴承故障进行诊断且准确率较高。
[Abstract]:With the continuous expansion of modern industrial scale and the increasing complexity of the system, motor bearings are more and more used in industrial production. Effective and accurate fault diagnosis for motor bearings has become a very meaningful research topic. A set method of motor bearing fault diagnosis is proposed based on set empirical mode decomposition (EMD) and improved local mean decomposition (ILMD)-improved universal gravity search algorithm (IGSA-incremental probabilistic neural network / IPNN). In order to improve the accuracy of fault diagnosis of motor bearings, all the test data in this paper come from the bearing Experimental Center of case Western Reserve University. In industrial production, the working environment of motor bearings is often very noisy. In addition, the vibration signals of other equipments contain noise because they are disturbed by the vibration itself. Therefore, the collected data should be preprocessed to reduce the noise. In this paper, EEMD algorithm, which has a strong ability to decompose nonstationary nonlinear signals, is used to pre-process the data by calculating the correlation coefficient and setting a threshold. Aiming at the problem of endpoint effect in LMD, an improved LMD method is proposed to decompose the data to improve the effect of endpoint effect. The sample entropy and energy of the product function component are calculated as feature parameters to form the fault eigenvector. As the input of neural network for fault diagnosis, the method of fault diagnosis adopts the feedforward neural network IPNNNIPNN based on statistical principle, which does not need to set initial weights, is simple in training, and has strong classification ability. The parameters of the network model will have a significant impact on the diagnostic performance. Therefore, an improved GSA optimization algorithm based on time-varying weight and boundary mutation is used to optimize the threshold of the network model. In order to improve the accuracy of the classification results, the convergence speed of the standard GSA algorithm is slow and it is easy to fall into the local optimal state. The data-driven EEMD-ILMD-IGSA-IPNN motor bearing fault diagnosis set method presented in this paper has good diagnostic performance and can effectively diagnose motor bearing fault with high accuracy.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM307

【参考文献】