基于双视角CT重建的固体发动机点火试验药柱缺陷实时检测技术

Real-time detection technology of grain defects in solid motor firing test based on dual-view CT reconstruction

为了在固体火箭发动机点火试验环境下,研究药柱裂纹等缺陷对燃面退移的影响,需要对药柱进行实时缺陷检测。但由于传统CT重建算法和现有的基于深度学习的方法无法实时、有效地重建内部缺陷来进行动态监测,提出了一种利用双视角投影数据进行CT重建来检测缺陷的方法。首先,该方法利用深度学习模型进行双视角下发动机内结构重建,完成二维到三维的跨维度重建;然后,利用无监督异常检测算法对双视角下投影图像进行缺陷检测及定位,获取缺陷区域坐标;最后,利用视差原理在三维结构体中重建缺陷,从而可以实时得到发动机内部三维结构,达到动态监测目的。利用仿真数据与模拟工件数据进行实验,实验结果表明,该方法仅利用双视角下的投影图像即可有效完成三维重建以进行缺陷检测。相比于现有的基于深度学习的超稀疏重建算法,无需缺陷区域先验信息。

In order to study the effect of defects such as propellant grain cracks on burning surface regression in the firing test environment of solid rocket motor,the propellant grain need to be real-time detected.However,since the traditional CT reconstruction algorithm and the existing deep learning-based method cannot reconstruct the internal defects in real time and effectively conduct dynamic monitoring,so a method on CT reconstruction to detect defects by means of dual-view projection data for was proposed.Firstly,the deep learning model was used to reconstruct the internal mechanism of motor from two perspectives,and the two-dimensional to the three-dimensional cross-dimensional reconstruction was completed.Then,the unsupervised anomaly detection algorithm was used to detect and locate the defects in the dual-view projection image,and the coordinates of the defect area were obtained.Finally,the parallax principle was used to reconstruct the defects in the 3D structure,so that the 3D internal structure of motor can be obtained in real time to achieve the purpose of dynamic monitoring.The simulation data and the simulated workpiece data were used for experiments.The experimental results show that the method can effectively complete the three-dimensional reconstruction for defect detection only by using the projection images from two perspectives.Compared with the existing ultra-sparse reconstruction algorithms based on deep learning,it does not need the prior information of defect regions.