Study on Extraction Method of Failure Signal and Automatic Generation Method of Feature Parameters

近年来，由于柔性和智能制造系统的发展，世界范围内的工业正在发生深刻的变化。这种无人工厂的趋势将在21世纪继续发展。根据这些发展，工厂维修的作用也将继续演变为高生产力和高质量的“保证者”之一。在工厂机械的状态监测领域，测量振动或声音信号，以检测故障和识别故障种类。当在机器故障的早期阶段或在远离故障部件的位置测量用于诊断的信号时，由于故障信号受噪声的强烈污染，故障信号的提取和故障的早期检测是困难的。为了提高故障检测的灵敏度，从声音信号中尽可能清晰地消除噪声是很重要的。为了消除噪音，许多人提出了更多的方法。例如，带通滤波器、自适应滤波器、维纳滤波器、卡尔曼滤波器等。但在机械故障诊断领域，这些方法并不总是适用于故障信号的提取。在用计算机对工厂机械进行状态监测时，必须有良好的特征参数，才能准确地区分模式。目前还没有一种可接受的提取优秀特征参数的方法。为了克服这些困难，本研究提出了以下新方法。(1)故障信号的提取方法1)利用遗传算法和统计信息从机器异常状态下测得的c信号中提取故障信号的方法。(2)通过顺序统计试验从机器异常状态下测得的频谱中提取故障频域的方法。(2)特征参数自动生成方法1)利用遗传算法对特征参数进行时域自重组2)利用遗传算法对特征参数进行频域自重组。3)基于小波分析和遗传算法的特征参数自动生成方法，用于非定常工况下机器的诊断。(3)智能诊断方法提出了“部分线性化神经网络（pnn）”和粗糙集知识获取方法，用于齿轮设备的故障诊断和模糊诊断的神经网络处理。通过对滚动轴承、齿轮设备等实际故障诊断的应用，验证了所提方法的有效性。少

Recently, industry world wide has been experiencing profound changes as the result of the development of flexible and intelligent manufacturing system. This tendency towards unmanned plants will to continue to develop in the 21st century. In line with these developments, the role of plant maintenance will also continue to evolve to one of a "guarantor" or high productivity and quality.In the field of condition monitoring for plant machinery, vibration or sound signal for measured for detection of failures and discrimination of kinds of failure. When the signals for the diagnosis are measured at an early stage of a machine failure or at a distant location from the failure parts, the extraction of failure signal and the early detection of failure are difficult, because the failure signal is strongly contaminated by noise. It is important to cancel the noise from the sound signal as cleanly as possible in order to increase the sensitivity of failure detection. For noise canceling, many me … More thods have been proposed. For example, band pass filter, adaptive filter, Wiener filter, and Kalman filter etc.. But in the field of machinery diagnosis, these methods can not always be applied to failure signal extraction.Furthermore. When using a computer for condition monitoring for plant machinery, excellent feature parameters are necessary, by which patterns can be precisely distinguished. Currently there is not an acceptable method for extracting the excellent feature parameter.For overcoming these difficulties, this study proposes new method as follows.(1) extraction methods of failure signal1) Extraction method of the failure signal from thc signal measured in the abnormal state of a machine using genetic algorithms (GA) and statistical information.2) Extraction method of failure frequency areas from spectrum measured in the abnormal state of a machine by sequential statistical tests.(2) Automatic Generation Method of Feature Parameters1) Self-reorganization of feature parameters in time domain by genetic algorithms2) Self-reorganization of feature parameters in frequency domain by genetic algorithms.3) Automatic generation method of feature parameters by Wavelet analysis and genetic algorithms for diagnosis of machine in unsteady operating conditions(3) Intelligent diagnosis methodThe "Partially-linearized Neural Network (P.N.N.)" and the knowledge acquisition method by rough sets have been proposed, in order to diagnosing failures of a gear equipment and processing ambiguous diagnosis by neural network.The efficiencies of all the methods proposed in this study have been verified by applying them to practical failure diagnosis, such as, rolling bearing, gear equipment etc.. Less