基于神经网络模型的振动筛损伤检测研究

发布时间：2018-01-05 02:18

本文关键词：基于神经网络模型的振动筛损伤检测研究　出处：《太原理工大学》2012年硕士论文　论文类型：学位论文

【摘要】：随着机械制造技术的提高和现代工业发展的需求,机械设备的结构愈来愈复杂,生产的高效率依赖机械设备提供的强大生产能力,设备的任何故障都会给生产带来巨大的损失。因此,在生产中对设备进行故障诊断是非常有意义的。故障诊断技术能够在监测设备运行状态的基础上,分析和诊断出机械设备的故障状态以及故障发展的程度。目前以时域和频域分析为主的方法研究具有强非线性特性系统的故障诊断不是很好。本文利用系统辨识建立振动筛缩小模型系统的模型,通过分析辨识模型的特性,对直线振动筛进行故障诊断和裂纹发展趋势研究。为了从采集到的信号中提取出特征信息,首先对实验测得的振动信号做预处理分析,将数据进行归一化,然后用小波消噪方法对信号消噪、采用最小二乘法消除信号的趋势项,最后去除信号的直流分量。研究振动筛缩小模型下横梁存在裂纹时系统的辨识方法。分别运用线性模型、非线性模型、神经网络模型对振动筛系统进行建模,通过分析模型和实际系统的拟合度,得出神经网络模型的精度较高。进一步通过判断不同神经网络辨识模型的拟合度、检测残差等,选择神经网络NNARX模型。最后确定了该模型的各个参数(包括网络层数、隐层神经元个数、激活函数等)以及各参数对辨识精度的影响。在振动筛缩小模型下横梁有无裂纹时,分别辨识出系统的神经网络模型,通过分析振动筛在不同状态下实测振动信号的幅值谱、辨识模型的虚拟响应谱、模型的权值,得出分析辨识模型的特性可以作为判断振动筛是否有裂纹的依据。最后,将分析模型特性的方法应用到实际振动筛裂纹发展趋势的研究上。辨识出实际振动筛存在裂纹时的模型,以天数增加的方式获取振动筛的振动信号,研究辨识模型在不同时刻时权值的变化,经统计分析得出,随着时间的增加,模型的权值呈逐渐减小且集中的趋势。实验表明,通过分析辨识模型的权值来研究振动筛裂纹的发展趋势是可行的,也是有意义的。
[Abstract]:With the increase of mechanical manufacturing technology and the needs of modern industrial development, more and more complicated structure of mechanical equipment, strong production capacity of the production of high-efficiency rely on mechanical devices, any equipment failure will bring huge loss to the production. Therefore, the fault diagnosis of the equipment in production is very meaningful for fault diagnosis. Based on the technology can monitor equipment running condition, analyze and diagnose the fault state of mechanical equipment fault and the degree of development.
Research on fault diagnosis method of the time domain and frequency domain analysis which has strong nonlinear characteristics of the system is not very good. In this paper, using system identification to establish vibration sieve scale model system model, by analyzing the characteristics of the two models, the research of fault diagnosis and crack development trend of linear vibrating screen.
In order to extract the feature information from the collected signal, first preprocessing analysis of vibration signal measured, the data were normalized, and then use the wavelet denoising method for signal denoising, eliminating trend signal by using the least square method, finally removing DC component signal.
Study on vibration sieve narrow crack identification method of system model under the beam. By using linear model, nonlinear model, neural network model of vibrating screen system, through the analysis of model and actual system fitting degree, that neural network model is of higher precision. Further by judging the different neural network model fitting. Detection and selection of NNARX neural network model. Finally, the parameters of the model are determined (including network layers, number of neurons in hidden layer activation function, etc.) and the influence of parameters on the identification accuracy.
In the narrow beam model crack free vibration sieve, respectively, to identify the neural network model of the system, through the analysis of vibration sieve in the condition of different amplitude of vibration signal spectrum, virtual spectrum response identification model, model weights, the characteristic analysis of identification model could be used to judge whether there are cracks on the vibration sieve.
Finally, the research methods of the characteristic analysis model is applied to the actual vibration sieve crack on the trend of development. To identify the actual vibration sieve crack model, vibration signal acquisition of vibrating screen to increase in the number of days the way, change of identification model in different time weights, according to statistical analysis, with the increase of time. The model weight decreases gradually and the concentration trend. Experimental results show that the development trend by analyzing the identification model of the weights of the vibration sieve crack is feasible, but also meaningful.

【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH237.6;TH165.3

【相似文献】