基于高阶谱和支持向量机的电梯故障诊断

发布时间：2018-01-13 00:24

本文关键词：基于高阶谱和支持向量机的电梯故障诊断　出处：《华侨大学》2013年硕士论文　论文类型：学位论文

【摘要】：目前随着国家的发展，电梯运用得越来越多，然而，电梯事故率也增大，因此，人们对电梯运行的安全性和及故障的预测提出了更高的要求。考虑到电梯系统的复杂性及其故障的隐蔽性，单靠人力来排除故障是远远不够的。本文的贡献在于采用时间序列的高阶累计量和时间序列的高阶累积量系数的支持向量机结合的方法，对电梯的振动加速度信号进行分析，达到故障识别的目的。电梯运行过程中的振动信号存在明显的非线性、非高斯性的特征，高阶累积量是处理这类信号的有力工具，高阶累积量理论上可以完全抑制高斯噪声（高斯白噪声和色噪声）和对称分布噪声，将其应用于电梯作业的研究中，分析其高阶谱特性，试图找出电梯不同工况下高阶谱图之间的区别；同时，利用高阶累积量建立的自回归（Autoregressive，AR）模型系数作为支持向量机的输入，以电梯的正常和故障状态作为输出，通过支持向量机的自学习能力，达到机器自动识别故障的功能。两种方法结合应用于电梯的信号分析中，可得到以下结论：（1）分析所采集的电梯振动的加速度信号，发现不论电梯处于何种工况，高阶谱图中谱峰分布各不相同，这是由于电梯的信号具有非线性和非高斯性，它们的频率成分以及各频率分量之间发生不同程度的二次相位耦合的结果。（2）对高阶谱进行切片，发现电梯在不同工况下的切片图中，出现的谱峰个数以及谱峰所在的频率位置均不同。因此，从谱图中反映的信息，能有效区分出电梯是处于正常还是故障状态。（3）将高阶累积量所建立的AR模型系数与支持向量机结合应用于电梯的信号分析中，引入核主元分析，以降低支持向量机的输入维数，从而增强支持向量机模型的推广能力。（4）支持向量机对数据建模过程中，，惩罚因子及核函数参数的值会影响模型预测的精度，为了选取更好的参数组合，采用交叉验证法优化参数，获得了较高的故障识别率。
[Abstract]:At present, with the development of the country, with more and more lift, however, elevator accident rate is also increased, therefore, people put forward higher demand forecast for the operation of the elevator safety and failure. Considering the complexity of the elevator system and its fault concealment, rely on people to troubleshooting is not enough. The contribution of this paper is using the time series of high order cumulant and the time sequence of higher order cumulants combined with support vector machine weight coefficient method, the vibration acceleration signal of the elevator, to achieve fault recognition.
There are obvious nonlinear vibration signals of elevator in operation process and characteristics of non Gauss, high order cumulant is a powerful tool to deal with this kind of signals, high order cumulant theory can completely suppress Gauss noise (Gauss white noise and color noise) and symmetrical distributed noise, and applied it to study the elevator operation the high order spectrum analysis, and tries to find out the difference between the spectral diagram of high order lift under different working conditions; at the same time, since the regression using high order cumulants (Autoregressive, AR) to establish the model coefficients as the input of support vector machine, as output to lift the normal and fault condition, through self-learning ability of support vector machine to machine recognition, automatic fault function. Two methods used in signal analysis of the elevator, the following conclusions can be obtained:
(1) the acceleration signal of elevator vibration analysis of the acquired that the elevator in what condition, high order spectrum peak distribution is different, this is because the elevator signal is non-linear and non Gauss, two phase coupling results in different degrees between the respective frequency components and each frequency component.
(2) slice the higher order spectrum. It is found that the number of peaks and the frequency locations of the peaks are different under different conditions. Therefore, information reflected from the spectrum can effectively distinguish whether the elevator is in normal or malfunction state.
(3) combining the AR model coefficients and support vector machines constructed by high-order cumulants to the signal analysis of elevator, the kernel principal component analysis is introduced to reduce the input dimension of support vector machine, so as to enhance the generalization ability of SVM model.
(4) in the process of data modeling, the value of penalty factor and kernel function parameter will affect the accuracy of model prediction. In order to select better parameter combination, cross validation method is used to optimize parameters, and a higher fault recognition rate is obtained.

【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TU857;TP18

【相似文献】