梳状滤波阵图及其应用

提出了一般梳状滤波器的设计方法,证明了传统的时域平均只是一般梳状滤波的一个特例。在此基础上提出了一种新的信号分析方法——梳状滤波阵图法:首先粗略估计周期信号基频的取值范围,然后以一定频率间隔离散化该频率区间,以得到的一系列离散化的频率值作为信号基频的估计,逐次对采样信号做梳状滤波处理,将得到的一组时域滤波波形以二维灰度图或三维阵图的形式显示,最后从中挑选出特征明显的若干行做进一步的分析。该方法可以在无同步信号的情况下有效地提取出混杂于强背景噪声中的周期信号,将其应用于滚动轴承故障信号精细特征分析取得了满意的效果。     

基于时域准同步的谐波和间谐波检测算法

傅里叶分析法因其简单方便成为常用的谐波、间谐波检测方法,但该方法在非同步采样时存在频谱泄露问题,由此产生的谐波和间谐波的相互干扰会严重降低测量准确性.通过采用时域准同步方法减少频谱泄漏,并采用梳状滤波器分离谐波和间谐波,可以有效提高谐波和间谐波的检测精度.在电能质量测量实时性要求高的场合,为了避免梳状滤波器导致的长时延,可以通过引入时域平均法在时延较小情况下获得相当的间谐波测量精度.给出该方法在电力谐波、间谐波测量方面应用的算法流程,并分别在MATLAB和谐波分析仪上实现谐波、间谐波分析算例.     

往复式压缩机气缸压力模拟曲线提取

提出了一种基于振动信号分析的往复式压缩机故障诊断方法.应用频域能量准则判定汽缸压力变化引起的振动响应信号所处频带,构造梳状滤波器提取出响应信号,对得到的相应信号进行包络分析,得到的包络曲线体现了气缸压力的变化趋势,可应用该曲线绘制模拟示功图对往复式压缩机故障进行诊断.对工程信号的分析表明,该方法不仅可以区分出压缩机正常与故障状态,还可以区分出不同的故障类型,具有较好的工程应用价值.     

数字下变频的FPGA实现

研究了高倍抽取的数字下变频设计,重点分析了基于级联积分梳状滤波器和级联半带滤波器的多级抽样频率算法.并提出了用最新的Systemgenerator软件实现FPGA的设计、仿真方案,缩短了开发周期,简化了设计流程,增加了系统的集成度和稳定性,降低了开发成本.对于混频器、级联积分梳状滤波器和数字下变频器都给出了仿真波形.     

电荷藕合器件在视频信号处理中的应用

介绍一种用CCD321型电荷藕合器件设计的梳状滤波器,这种滤波器能较好分离出色度视频信号的两个分量.     

结合ACF和TFBSS的心冲击伪迹抑制方法

同步脑电-功能磁共振成像(EEG-f MRI)可同时对大脑活动进行高时间和高空间分辨率观测,但磁共振扫描导致采集得到的脑电信号中包含心冲击伪迹,严重制约了同步EEG-f MRI应用的发展。为此,提出一种结合自适应梳状滤波器(ACF)与时频盲源分离(TFBSS)的心冲击伪迹抑制方法,首先估计脑电信号中心冲击伪迹的J峰位置,并根据该位置自适应调整梳状滤波器参数,对脑电(EEG)信号进行自适应梳状滤波,初步抑制其中的心冲击伪迹。然后利用TFBSS对心冲击伪迹作进一步抑制,最终获得较为清晰的EEG信号。采用临床EEG信号进行心冲击伪迹抑制实验,选择视觉效果、归一化功率谱比值和峰峰值比值评价性能。实验结果表明,所提出的方法能成功抑制心冲击伪迹;且相对于传统的心冲击伪迹抑制方法更有优势。     

一种新型光梳状滤波器自动装配系统控制分析

光梳状滤波器是一种有效提升光通信技术主干网络传输容量的关键光通信器件,过去一直以手工方式生产组装,对实现自动化生产无疑具有非常重要的意义。介绍了一种基于数字PID的新型光梳状滤波器自动装配系统的开发。     

基于连续小波灰度图的变速箱故障诊断

为了诊断汽车变速箱的周期性冲击故障，利用连续小波变换灰度图分别对正常和故障汽车变速箱振动信号进行了分析。结果发现，连续小波灰度图不仅能识别变速箱的正常与故障，准确提取出周期性冲击故障信息，而且能够非常直观形象地表达出信号的细微结果，并进一步显示出故障变速箱中同时存在的两种相同频率的故障信息，从多层次、多方位观察到了分析信号的细微变化。     

基于图结构的轴承故障诊断北大核心

滚动轴承故障振动信号具有非线性、非平稳的特征,在轴承早期破损阶段,即使轴承表面出现了损伤,故障产生的振动信号仍然表现得非常微弱,再加上大量噪声的影响,仅从时域和频域很难发现故障特征,给故障检测造成了较大的难度。针对轴承振动信号的特点,将短时Fourier变换与图模型相结合,提出了一种基于图模型的时频分析方法,利用短时Fourier变换得到信号的时频图,选取每一时刻频谱图中各主频的幅值构建图模型,通过图模型的相似性对比检测轴承故障并通过主频幅值的变化量确定故障频率。     

基于双时域微弱故障特征增强的轴承早期故障智能识别*

针对轴承早期微弱故障难以准确识别的问题,提出一种基于双时域微弱故障特征增强的轴承早期故障智能识别方法。利用广义S变换和Fourier逆变换推导出一种双时域变换,将轴承振动信号变换为双时域二维时间序列。根据双时域变换的能量分布特点,提取二维时间序列的主对角元素以构建故障特征增强的时域振动信号。仿真信号和轴承故障信号分析验证了双时域微弱故障特征增强的可行性和有效性。采用脉冲耦合神经网络和支持向量机对增强后的轴承信号进行时频特征参数提取和智能识别,平均识别精度达到了95.4%。试验结果表明所提方法能有效提高轴承早期故障的智能识别精度。     

Time domain averaging based on fractional delay filter

For rotary machinery, periodic components in signals are always extracted to investigate the condition of each rotating part. Time domain averaging technique is a traditional method used to extract those periodic components. Originally, a phase reference signal is required to ensure all the averaged segments are with the same initial phase. In some cases, however, there is no phase reference; we have to establish some efficient algorithms to synchronize the segments before averaging. There are some algorithms available explaining how to perform time domain averaging without using phase reference signal. However, those algorithms cannot eliminate the phase error completely. Under this background, a new time domain averaging algorithm that has no phase error theoretically is proposed. The performance is improved by incorporating the fractional delay filter. The efficiency of the proposed algorithm is validated by some simulations.     

Flexible time domain averaging technique

Time domain averaging(TDA) is essentially a comb filter,it cannot extract the specified harmonics which may be caused by some faults,such as gear eccentric.Meanwhile,TDA always suffers from period cutting error(PCE) to different extent.Several improved TDA methods have been proposed,however they cannot completely eliminate the waveform reconstruction error caused by PCE.In order to overcome the shortcomings of conventional methods,a flexible time domain averaging(FTDA) technique is established,which adapts to the analyzed signal through adjusting each harmonic of the comb filter.In this technique,the explicit form of FTDA is first constructed by frequency domain sampling.Subsequently,chirp Z-transform(CZT) is employed in the algorithm of FTDA,which can improve the calculating efficiency significantly.Since the signal is reconstructed in the continuous time domain,there is no PCE in the FTDA.To validate the effectiveness of FTDA in the signal de-noising,interpolation and harmonic reconstruction,a simulated multi-components periodic signal that corrupted by noise is processed by FTDA.The simulation results show that the FTDA is capable of recovering the periodic components from the background noise effectively.Moreover,it can improve the signal-to-noise ratio by 7.9 dB compared with conventional ones.Experiments are also carried out on gearbox test rigs with chipped tooth and eccentricity gear,respectively.It is shown that the FTDA can identify the direction and severity of the eccentricity gear,and further enhances the amplitudes of impulses by 35%.The proposed technique not only solves the problem of PCE,but also provides a useful tool for the fault symptom extraction of rotating machinery.     

Dynamic properties of measuring circuits applied to frequency output transducers

Frequency output measuring transducers transform the measured quantity x into the frequency f or the period T of the output signal. Frequency is usually measured by counting the number of pulses produced by the output voltage during a constant counting time t_c, long enough to minimize a random error. The period is measured by counting the pulses of a reference oscillator in the varying time equal to N periods of the transducer's output signal. Considering the dynamic properties of both measuring chains one may conclude that averaging of the measured occurs, and in order to calculate the dynamic error, an averaging transducer ought to be taken as a reference. The paper indicates that under dynamic conditions the frequency measuring chain may be treated as a really averaging one, the period measuring chain only as an approximately averaging one, and in both cases averaging is not sufficient to eliminate the aliasing error of the sampling procedure. A method for accelerating the measurement and then avoiding the aliasing error is the “one-period method” considered in the paper. The averaging error and delay error of this method are considered in detail, under the assumption that a parameter-controlled oscillator is used as an element of the measuring chain.     

电域自适应中红外激光甲烷传感系统

为了抑制甲烷传感器中统计特性无法预知的电学噪声,本文结合递归最小二乘自适应去噪算法和直接吸收光谱技术,使用中心波长为3 291nm的带间级联激光器和多反射吸收气室,研制了一种电域自适应中红外甲烷传感器系统。在传统探测器输出信号(称为信号通道)的基础上,增加了激光器电流驱动器的反馈信号作为噪声通道来感知电学噪声。利用MATLAB软件对最小二乘法在直接吸收光谱技术中的滤波效果进行了仿真。通过在激光器驱动信号中施加不同的噪声,实验验证了最小二乘法的去噪效果。针对该传感器的Allan标准差结果表明,当不使用自适应最小二乘法时,系统在积分时间为6s的检测下限为78.8nL/L;使用RLS自适应算法时,系统的检测下限为43.9nL/L。相比基于传统传感结构和滤波方法的中红外直接吸收光谱传感器,本文所报道的中红外甲烷传感器由于采用了电域自适应滤波方法,因而呈现出更好的抗干扰性和稳定性。     

时域同步平均原理与应用

在回转和往复机械的状态监测和故障诊断中,时域同步平均可有效衰减与回转频率无关的干扰,提取与工作状态直接相关的周期信号,是一种非常有用的信号分析和预处理技术。本文对时域同步平均的梳形滤波器特性,实际操作中采用频率跟踪的必要性,实现频率跟踪的不同技术方案等几方面进行了论述和分析。实验分析结果表明了这一技术的有效性和可行性     

振动故障信号的软件积分研究与应用

分析了常用软件积分方法及其误差产生的原因，采用基于时域插值和FFT变换的波形修正算法进行改进。在时域插值后积分，利用最小二乘法拟合并去除积分后信号波形的趋势曲线；在频域使用FFT变换滤除积分后信号的趋势成分，修正积分后的曲线。汽轮发电机组振动分析表明，基于FFT变换的波形修正算法得到了较好的效果，为判断机组运行状态、进行故障诊断提供了依据。     

光镊系统随机漂移建模和误差补偿

针对光镊系统本身噪声对测量精度的影响,提出了一种光镊系统随机漂移误差的有效补偿方法。首先,介绍了时间序列分析法和卡尔曼滤波技术,基于时间序列分析法建立了光镊的随机漂移误差模型;然后,用基于时间序列模型的卡尔曼滤波方法来减小该漂移误差。采用提出的方法对光镊设备实测数据的误差进行了补偿,结果表明：数据的误差方差由补偿前的188.90 nm²减小为8.41 nm²。计算补偿前后的艾伦方差可知,系统在平均时间为1 s时可使最小位移误差从 0.7 nm降低到0.1 nm。得到的结果显示：提出的滤波方法有效地抑制了光镊系统的漂移误差,将其用于双光镊对准可提高捕获光和探测光的对准精度,进而提高光镊系统的性能指标。     

Practical method for noise removal in scanning electron microscopy

A new smoothing filter has been developed for noise removal of scanning electron microscopy (SEM) images. We call this the complex hysteresis smoothing (CHS) filter. It is much easier to use for SEM operators than any other conventional smoothing filter, and it rarely produces processing artifacts because it does not utilize a definite mask (which usually has processing parameters of size, shape, weight, and the number of iterations) like a common averaging filter or a complicated filter shape in the Fourier domain. Its criterion for distinguishing noise depends simply on the amplitude of the SEM signal. When applied to several images with different characteristics, it is shown that the present method has a high performance with some original advantages.