卷积混合旋转机械故障信号的盲分离

发布时间：2018-11-29 08:21

【摘要】：随着科学技术与现代化工业的不断发展，各种工业设备日趋集成化、高速化和智能化，，设备振动监测与故障诊断越来越重要，对振动信号的采集、分析与处理是设备故障诊断的基础，传统的振动信号分析方法虽然比较成熟，但是他们都有其自身的局限性。盲源分离（Blind Source Separation，BSS）是现代振动信号处理领域中的一个新的研究热点，由于其可以在源信号和传输通道等先验知识均未知的情况下，仅依靠观测信号就能恢复出源信号，目前已经被用于语音信号处理、阵列信号处理、数据挖掘、图像识别、生物医学信号处理等很多领域。本文就盲源分离的混合模型、理论算法及其在旋转机械故障诊断中的应用等方面开展研究并取得了一些有意义的结论。从瞬时线性混合的盲源分离模型出发，介绍了几种基于信息论的盲源分离独立性判据，选取三种瞬时线性混合盲源分离算法（FastICA算法、EASI算法、SOBI算法）进行了仿真试验，仿真试验证明， FastICA算法的分离效果优于EASI算法和SOBI算法。考虑到实际应用中传感器接收的信号往往是振动源信号与传递通道冲击响应的卷积，本文重点研究了卷积混合信号的分离问题，对时域RLS盲解卷算法和频域复数FastICA盲解卷算法进行的仿真试验表明，时域盲解卷积算法较频域盲解卷积算法复杂，求解速度相差数十倍。在模拟故障试验台上采集了滚动轴承外圈故障和齿轮断齿故障振动信号，对实测信号进行了时域RLS盲解卷与频域复数FastICA盲解卷，更进一步，对解卷结果进行小波分解，得到了比较理想的分析结果。盲解卷与小波分解相结合的信号处理方法较单纯的盲解卷方法及瞬时混合盲分离方法能获取更清晰和更丰富的故障特征信息。
[Abstract]:With the continuous development of science and technology and modern industry, all kinds of industrial equipments are becoming more and more integrated, high-speed and intelligent, the vibration monitoring and fault diagnosis of equipment is more and more important, and the acquisition of vibration signals is becoming more and more important. Analysis and processing are the basis of equipment fault diagnosis. Although the traditional vibration signal analysis methods are mature, they all have their own limitations. Blind source separation (Blind Source Separation,BSS) is a new research hotspot in the field of modern vibration signal processing, because it can recover the source signal only by observation signal when the prior knowledge such as source signal and transmission channel are unknown. At present, it has been used in many fields, such as speech signal processing, array signal processing, data mining, image recognition, biomedical signal processing and so on. In this paper, the mixed model of blind source separation, theoretical algorithm and its application in fault diagnosis of rotating machinery are studied and some meaningful conclusions are obtained. Based on the instantaneous linear mixing blind source separation model, several independence criteria of blind source separation based on information theory are introduced. Three instantaneous linear mixed blind source separation algorithms (FastICA algorithm, EASI algorithm, SOBI algorithm) are selected for simulation. Simulation results show that the separation effect of FastICA algorithm is better than that of EASI algorithm and SOBI algorithm. Considering that the signal received by the sensor is usually the convolution between the vibration source signal and the shock response of the transmission channel in practical applications, the separation of convolution mixed signals is studied in this paper. The simulation results of the time domain RLS blind deconvolution algorithm and the frequency domain complex FastICA blind deconvolution algorithm show that the time domain blind deconvolution algorithm is more complex than the frequency domain blind deconvolution algorithm, and the speed difference is tens of times. The vibration signals of rolling bearing outer ring fault and gear broken tooth fault are collected on the simulated fault test bed. The measured signals are analyzed by RLS blind deconvolution in time domain and FastICA blind deconvolution in frequency domain. Furthermore, the deconvolution results are decomposed by wavelet transform. An ideal analysis result is obtained. The signal processing method combining blind deconvolution with wavelet decomposition can obtain clearer and richer fault feature information than the simple blind deconvolution method and instantaneous mixed blind separation method.
【学位授予单位】：华东交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【参考文献】