【作者】 王飞；

【导师】 张树生；

【作者基本信息】 西北工业大学 ， 航空宇航制造工程， 2005， 硕士

【摘要】 固体火箭发动机体积巨大,结构复杂,如何对其进行有效的检测,一直以来都是一个难点。传统的检测手段主要有两种,一种是超声脉冲回波技术,主要用于检测固体火箭发动机的脱粘,采用手检由人工对回波波形判读,确定是否脱粘。该方法存在误检率高、无法定量分析、检测周期长等缺点;另一种是X射线投影技术(又称数字影像技术),主要用于检测发动机第二、三界面的脱粘及药柱内的裂纹、夹渣等缺陷。该方法获得的是间接的一维重叠信息,通常需要专业技术人员根据经验进行人工判读,通过综合分析对缺陷的大概方位和大小进行粗略判别,存在效率低,主观性、随意性大等缺点。 基于ICT的无损检测是目前公认的最佳的无损检测手段,但是,其在固体火箭发动机检测领域却尚未大范围使用,主要原因就是固体火箭发动机体积巨大,采用高能ICT机检测消耗时间长,检测效率很低,从而导致检测成本高,企业无法承受。针对这一问题,本文提出了一个可行的基于ICT的固体火箭发动机实时检测和成像方案,并针对方案中的两个关键技术——发动机ICT切片少量投影重建和局部重建,进行了深入的研究。主要的研究内容和取得的成果如下: 1.提出了一个可行的固体火箭发动机实时检测和成像方案。在该方案中,我们将传统的X射线投影技术和ICT技术相结合,先使用X射线投影技术进行初检,发现存在缺陷可疑区域,然后利用ICT作精检。精检分两步进行:第一步作稀疏扫描并实施少量投影下重建,快速获取切片的低分辨率图像;第二步,根据低分辨率图像所确定的缺陷区域实施局部重建,获取高分辨局部图像,从而对缺陷做出准确判定。 2.提出了固体火箭发动机少量投影下的多准则神经网络重建算法和遗传重建算法。从投影重建切片图像,可以看作是解一个线性方程组的问题,由于投影数目少,该方程组无唯一解。本文的思路是首先加入三个先验约束条件:极大熵、峰值最小和重建图像再投影与实测投影差值最小,将图像重建问题转换为有约束的多目标优化问题,然后,分别利用HNN神经网络和遗传算法进行求解,所得解即对应待重建切片图像。 3.将小波运用到固体火箭发动机切片的局部图像重建。由于傅立叶变换的非局部性,使得FBP等传统变换算法无法求解局部图像重建问题。而与此相反,小波则具有良好的局部性质,再结合多分辨率分析理论和快速小波分解及重构算法,使得小波能够快速的实现发动机的切片图像局部重建,并且所需投影数据范围只稍大于待重建局部区域。更多还原

【Abstract】 Due to the huge volume and complex structure, the NDT (non-destructive testing) of the solid rocket engine is a difficult problem all along. There are two main traditional methods: one is the ultrasonic method, the other is DR (Digital Radiation) method. The former is most used to check the debonding of the solid rocket engine by receiving the reflected ultrasonic wave and then judging by human beings according the shape of the wave. This method has a number of weaknesses such as random deviation, no quantitative analysis, and a longer testing cycle. The latter is most used to check the crack and slag inclusion inside the solid rocket engine. The acquired information by this method is the one dimension projection data, and the experience of operator himself is needed to determine the size and orientation of the faults. So the second method is also not enough efficient as its dependence of manual subjectivity.ICT (Industrial Computerized Tomography) is one of the best nondestructive detection methods currently, but it hasn’t been widely used in the field of nondestructive detection of solid rocket engine. The main reason is that it costs too much time for each detection, so the detection efficiency is lower and the detection cost is too high to endure for manufacturer. Aiming at this problem, we propose an innovative detection schema to reduce the detection time, in which there are two key technologies, one is the image reconstruction based on a few of projections, and the other is the local image reconstruction. The main research contents of this paper are as follows:1. Propose a new detection and image reconstruction schema. The detection schema consists of three main steps. 1) The DR method is used to check roughly whether an abnormity exists. 2) The image reconstruction based on a few of projections is implemented to determine the range of the faults. 3) Through the local high-resolution slice image acquired by the local tomography we can make the ultimate conclusion on the properties of the faults.2. Propose a multicriterion HNN neural network image reconstruction algorithm and a Genetic image reconstruction algorithm based on a few of projections of the solid rocket engine. Normally, image reconstruction from incomplete projections is an ill-conditioned inverse problem which can not be solved by traditional FBP or ART algorithm. However, by adding in advance some constraints, it is converted into a multiple objective optimal problem, and then with a HNN neural networkand a modified genetic algorithm the global optimal solution is gained separately. Typical experiments proved that the reconstruction results of the HNN algorithm is the best in all algorithms considered. 3. Apply the wavelet to local image reconstruction of the solid rocket engine. For the non-local property of Fourier transform, it isn’t fit for local image reconstruction (also called ROI, Region of Interesting). On the contrary, the wavelet has a good local property. Adding the multiresolution analysis theory and the fast wavelet transform algorithm, a high-resolution local slice image can be reconstructed by it with the projections passing through a region only slightly larger than ROI.更多还原