【作者】 包振明；

【导师】 赵德有； 马骏； 林哲；

【作者基本信息】 大连理工大学 ， 船舶与海洋结构物设计制造， 2013， 博士

【摘要】 海洋平台结构庞大,受风、浪、流和冰等环境因素长期作用,同时还受到地震、台风、海啸和船舶碰撞等意外作用的威胁。在载荷作用下,导管架海洋平台会出现裂纹。由于平台部分结构位于海面以下,裂纹不易被发现,难以直接进行人工检测。当重要部件发生裂纹并在极端海况下产生扩展时,会导致整个结构失效,危及工作人员的生命安全,产生重大的经济损失和海洋环境污染。所以,及时并尽早地发现结构的裂纹具有重要意义。本文以某一含四分之一跨处单裂纹的简支梁为例,计算了该梁的非贯穿单边裂纹损伤侧上、下表面、贯穿裂纹的上表面和内部非贯穿裂纹表面裂纹和内部裂纹等情况的不同损伤程度的位移模态和应变模态。根据已有的基于应变模态差分原理的损伤位置直接指标法ISMSD,利用等间距差分格式计算该简支梁非贯穿单边裂纹应变模态差分曲线,经Matlab编程计算将曲线进行光滑,计算得到直接指标值。由直接指标值的最大值找到对应的两有效极值点,这两个有效极值点间即是损伤位置。实例计算简支梁非贯穿单边裂纹损伤应变模态差分曲线,这些应变模态差分曲线在损伤处发生剧烈变化。差分曲线非峰值点损伤在损伤处不出现极值,因而损伤处的差分值不为零。损伤量不同,差分曲线损伤处突变程度略有不同,其规律相似。运用带有Grubbs的支持向量机法和带有Grubbs的BP神经网络法对该非贯穿裂纹简支梁进行损伤程度智能识别,识别并评估了四分之一跨处单裂纹的损伤程度,并从性能及准确度方面对两种方法进行了比较。若选取应变模态差作为网络输入指标,本文采用的两种方法都可以得到比较高的识别精度,而且有良好适应性。支持向量机方法相对误差更小。采用有限元软件ANSYS计算了某导管架海洋平台模型的一水平管件在完整状态、含单裂纹、含双裂纹三种情形时不同位置和不同损伤程度的频率和应变模态。验证了损伤会引起结构的频率降低和应变模态突变,频率降低的幅度随损伤程度的增加而增大。此外还发现,微小裂纹损伤时引起的频率变化很小：水平管端点损伤和中点损伤的频率下降幅度基本一致；双裂纹情形时的频率下降幅度均高于单裂纹情形时；损伤处应变模态曲线发生了显著改变,随损伤量的增加,应变模态曲线突变增大。采用SCE-UA算法和粗粒度并行遗传算法对平台模型的10处单裂纹进行了损伤程度的逐一智能识别。将应变模态差作为SCE-UA算法和遗传算法的输入数据,这两种方法均能取得较高的识别精度,具有良好的适应性。其中SCE-UA算法损伤识别结果误差更小,更精确。振动诊断中的应变模态法具有相对简单,成本较低,具有实时性、在线性、提取信号方便性和遥测性、可控性等诸多优点。本文的研究为工程实际应用提供了一定的参考价值,在结构损伤诊断识别中具有推广价值。更多还原

【Abstract】 Offshore platforms are giant structures, subject to various environmental loads, such as wind, wave, currents, and ice. Other environmental loads may also come from earthquakes, typhoons, tsunamis, ship collisions and other accidents. Under these dynamic loads, cracks may occur and further develop at the structural members of an offshore platform. It is impossible to inspect the platform structural components beneath the sea level. Therefore, cracks can be a great threat to the platform. Damages at important structural components may lead to catastrophic results. Therefore, it is necessary to discover and identify the cracks in time to ensure the structural safety.This dissertation adopts the intelligent diagnosis method based on the strain mode method to identify the crack at quarter span of a simply supported beam and a single crack at a jacket offshore platform pipe. The crack can be non-impenetrable into the beam section, while staying at the surface and in the beam with a certain depth. The author computes the displacement modes and the strain modes of all above cases. Based on the strain mode difference principle, direct index Ismsd——uses the equidistant difference scheme, without the modal characteristics of the original structure. FEM numerical simulations for non-impenetrate crack of different levels are also carried out. It is found that the strain mode differential curve of the damaged beam sharply changes at the damage location. The criteria of damage location detection are obtained by strain mode difference curves through a cubic spline interpolation. Through the two extremum points corresponding to ISMSD, the damage location can be found.To intelligently identify the damage level for a single crack at the non-impenetrable beam, the support vector machine (SVM) with the Grubbs and the BP neural network with Grubbs network are used. The differences between the strain mode shapes can be due to the network input. The performance and the accuracy of the two methods have been compared. The SVM approach is of higher accuracy.FEM numerical simulation by ANSYS is used to obtain the frequencies and strain mode shapes of the horizontal pipe at a offshore jacket platform model. A single crack and double cracks of the pipe with different positions and different degrees are considered. Damage in the pipe would cause the natural frequencies shift to lower values. The frequencies decrease with the increase of damage level. In addition, the frequency drop is very obvious. The frequency variation level of the double-crack case is higher. The strain mode differential curve of the damaged pipe sharply changes at the damage location.The ten single-cracks of the pipe in the platform model can be intelligently identified by SCE-UA and the coarse-grained parallel genetic algorithm. The network input are the differences between the strain mode shapes. The results of these two algorithms for the structural degree damage diagnosis show that both of the two methods have high identification accuracy and good adaptability. The error of SCE-UA algorithm is smaller.The strain mode method is a goog method for damage diagnosis, which is simple, cheap and has the virtue of real time, convenience, remote control. The study of this dissertation has favorable application foreground and spread value for engineering.更多还原