大型常压储罐底板的声发射在线检测
李一博1,孙立瑛2,靳世久1,邢菲菲1,杜 刚1
(1.天津大学精密测试技术与仪器国家重点实验室,天津300072
2.天津城市建设学院能源与机械工程系,天津300384)
摘要:基于大型常压立式金属储罐底板在线声发射检测及定位的原理,针对声发射检测过程中因声源性质不明确导致的罐底完整性评价结果不准确的问题,采用小波分析方法对罐底声发射信号进行了分解.通过提取声发射信号在不同小波分解频带上的特征频谱系数,与声发射波形参数共同作为BP神经网络学习样本集的特征向量,对神经网络的模式识别性能进行了优化.采用该神经网络对罐底裂纹、腐蚀、泄漏、机械噪声和电磁噪声等不同性质的声发射源进行判别时,其正确识别率均在90%以上,使基于声发射在线检测技术的储罐底板结构完整性评价技术更趋于完善和实用化.
关键词:声发射;罐底;在线检测;定位;神经网络;模式识别
中图分类号:TE972.1;TP274.5 文献标志码:A 文章编号:0493-2137(2008)01-0011.06
On-Line Acoustic Emission Inspection Method for Large Normal Pressure Storage Tank Bottom
(1. State Key Laboratory of Precision Measurement Technology and Instrument, Tanjin University, Tianin 300072, China;
2. Energy Technology and Mechanical Engineering Department, Tanjin Insiute ơf Urban Construction, Tanjin 300384, China)
Abstract:
The acoustic emission (AE) inspection and location principle of large vertical normal pressure storage tank bottom on-line inspection was studied. Aiming at solving the problem of inaccurate structure integrityevaluation resulted from ambiguous AE sources, wavelet analysis method was used to decompose acoustic emission signals derived from storage tank bottom. The characteristic frequency factors in diferent wavelet decom-position frequency bands were extracted as the BP neural network's characteristic input vector together withoriginal acoustic emission waveform parameters. Thus, performance of the BP network is optimized and its rec-ognition capability for AE sources is improved. The correct recognition rate of AE sources, such as crack, cor-rosion, leakage, mechanical noise and EMI noise are all increased to above 90% . The research makes thetank bottom structure integrity evaluation technology based on AE on-line inspection result to be more sophisti-cated and practical.
Keywords: acoustic emission; tank bottom; on-line inspection; location; neural network; pattern recognition
随着石油工业的迅速发展和能源需求的不断增加,原油和成品油的储备受到了世界各国的普遍关注。大型储罐是石化行业中油品储存的重要设施,一旦发生事故,极易引起火灾和爆炸,并且造成严重的环境污储罐底板是储罐最难检测的部位,长期以来,罐底的缺陷一直采用定期开罐的离线方法进行检测,即按照储罐的安全检测周期,停工后将储罐内的油品全部清空,再采用漏磁、涡流、超声等常规的无损检测方法对整个储罐底板进行检测.一般说来,对大型原油储罐进行一次内部(离线)检测需要几十至数酉万元,倒空、清洗、检测等时间可能达到几十天甚至数月[1]。而且,采用传统方法定期开罐普查时,一般仅有9%的罐确实发生了严重的腐蚀或泄漏,需要及时检测和修复[2]。可见,企业若不考虑储罐的状况,而仅仅按照计划进行检修,就无法科学地判断哪些储罐才是腐蚀损伤严重恧急需捡测的储罐.与传统的储罐底板腐蚀检测方法相比,声发射是一种在线、高效和经济的检测方法,也是嚣前醑际上唯一认可的大型鬻压金属储罐底板在线检测方法[3],它克服了传统技术需要停工置换、清理罐底、逐点扫描检查造成的费时、费力,总体检测费用高的缺点,因而具有非常广阔的应用前景.笔者通过提取声发射信号在不同小波分解频带上的特征谱系数,与声发射波形参数共同作为BP神经网络学习样本集的特征商量,优化了粹经髓络模式识别的性能,解决了声发射检测过程中因现场噪声干扰和声发射源性质不明确等导致罐底结构完整陛分级不准确问题。