【作者】 成建国；

【导师】 毛汉领；

【作者基本信息】 广西大学 ， 机械电子工程， 2008， 硕士

【摘要】 金属结构件在交变载荷作用下,出现疲劳裂纹是常见的故障之一,尤其是对于混流式水轮机转轮等结构复杂的大型构件更是如此。由于交变载荷的作用,裂纹不断扩展,常导致结构件的断裂,因而对疲劳裂纹的监测是非常必要的。声发射检测技术是一门新兴的多学科交叉的无损检测技术,已被广泛应用于设备的无损检测、在线监测中。本文在广泛查阅大量国内外科技文献资料的基础上,将声发射技术应用于金属疲劳裂纹的在线检测。由于声发射信号的瞬态性和随机性,它由一系列频率和模式丰富的信号组成,以及现场工作环境恶劣,声发射源种类较多。很难通过简单的频率或者幅值等滤波方法来获取比较纯净金属疲劳裂纹声发射信号。结合现场混流式水轮机转轮工作实际情况,初步确定现场较多的声发射信号为金属疲劳裂纹声发射信号、空化信号以及摩擦声发射信号,并增加标准断铅信号。综述分析声发射的分析方法及13个表征参数;通过采用神经网络模式识别的方法,设计出特征提取器,筛选出最能表达声发射本质的五个特征参数;同时采用可分离性判据的方法得到对分类效果最显著的特征参数,验证特征提取器的正确性;通过实验验证,利用所提取的五个特征参数可以识别出金属裂纹声发射信号。此外,在大型复杂构件声发射检测中,经常采用多传感器检测系统。可以采用数据层融合的独立分量分析方法和决策层融合的D-S证据理论,来融合多传感器信息,减少识别的不确定性,提高系统的识别、容错和抗干扰能力。在前面神经网络模式识别的基础上,借助数据融合技术和报警理论实现了对金属裂纹的发生进行报警。本课题来源于国家自然科学基金项目(50465002)——《混流式水轮机转轮叶片裂纹监测的理论和方法研究》。更多还原

【Abstract】 Under the alternate loading, accruing flaw in metal structure is one of faults that happen in metal structure frequently, especially in large-scale complex structural parts such as Francis turbine etc. As the effect of alternate loading, the flaw in the metal structure could develop continuously, and at the beginning, the fatigue flaw is very thin, therefore in this case, detecting flaw seems to be very necessary. Acoustic emission (AE) testing is applied widely as a rising multidisciplinary nondestructive testing technology in equipment of nondestructive examination and on-line monitoring. In this article, based on a great deal of resources from home and foreign country, AE technique will be applied to metal fatigue crack’s on-line monitoring.Since AE signal’s transient and randomness, it is composed of a series of rich frequency and pattern, and poor working environment and many acoustic emission source categories at the scene. It is very difficult to gain comparatively pure metal fatigue AE signal by simple methods such as frequency or amplitude filtering.After analyzing the actual work environment of the runner of Francis turbine, many AE signals can be determined initially at the scene, such as metal fatigue crack AE signals, cavitations, fricative AE signals, and increase the standard lead off signals.Summary of analysis method of AE and 13 characteristic parameters; Feature extraction was designed by use of neural networks and pattern recognition method, and five characteristic parameters that can express AE most were filtered; At the same time, the most notable feature parameters on classification were obtained by use of the separability criterion, to verify the accuracy of feature extraction; According to the experimental results, metal crack AE signals can be identified with the five characteristic parameters.In addition, in the large and complex component of the AE testing, multi-sensor detection system was used frequently. It can be fused in data layer by the independent component analysis, and decision-making layer by D-S evidence theory, to integrate multi-sensor signals, reduce uncertainty of recognition, improve the capabilities of system’s identification, fault-tolerance and anti-jamming.Based on the analysis of neural networks and pattern recognition referred, when metal crack happened, alarm can be carried out in virtue of the data fusion technology and the warning theory.更多还原