民航发动机内窥式三维探查技术研究

Borescope inspection technology is an important part in the civil aeroengine health diagnosis, the accuracy of the diagnosis data is of great significance for the safety of civil aviation. It has drawn people's attention extensively that finding new methods for a more reliable detection instead of the traditional detection method which only depends on human vision and operating experiences. According to this situation, a new borescope inspection technology is proposed which consists of engine scene manifold detection and its 3D reconstruction. A high-performance system is studied and developed as well as to process borescope images precisely and in real time. Firstly, a multi-view borescope camera is designed to expand the inspection scene. Secondly, the whole framework of a novel 2D+3D algorithm which combines 2D scene mosaicing and 3D structure reconstruction from borescope images is put forward for parallel processing, in which the feature extraction algorithm is well optimized. Thirdly, a solution to achieve the 3D reconstruction through projective reconstruction, metric reconstruction and scaling transform is presented. In addition, an approach to achieve high-precision 2D scene mosaicing is implemented via estimating the intra-frame homography matrices through 3D data constrained optimization. Finally, a scheme of borescope scene real-time processing system is carried out, which is able to process the data through multi-task dispatching based on multi-core processor. The purpose of this research is to quantizing the broescopic inspection data of civil aeroengine, to improve the efficiency and reliability of the detection, and to provide technical supports to ensure the safety of civil aviation.

内窥镜检测技术是民航发动机健康诊断的重要内容，其检测数据的准确性对民航安全至关重要。如何改变以人工目视和操作经验为主要判别依据的传统检测方法，已经引起行业界和科学界广泛重视。基于这一背景，本项目提出一种可对民航发动机进行二维全景探察和三维场景重构的内窥镜测量新技术，并研制高性能内窥镜图像实时处理系统：首先设计多目内窥镜镜头，从物理上实现测量场景的直接扩展。然后，提出“二维＋三维”的内窥镜图像处理算法架构，实现二维数据和三维数据的并行处理，并优化其中的特征抽取算法。在此基础上，给出了通过射影重构、度量重构和尺度变换实现三维重构的具体方案，并提出以三维数据约束优化单应矩阵计算的技术思路，实现高精度的二维全景拼接。最后，研究内窥镜图像实时处理系统的实现方案，基于多核CPU处理器实现多任务分派的数据处理方式。本研究旨在量化民航发动机内窥镜检测数据，提高检测效率和可靠性，为保障民航安全提供技术支持。

内窥镜检查在民航飞机发动机检修中发挥着重要作用，其检测数据的准确性对民航安全意义重大。本项目从这一内容出发，研究一种高效的内窥镜数据三维重建方法，并研制了系统样机，所涉及的主要研究内容如下：.（1）内窥镜镜头研究。包括侧视内窥镜镜头原型设计与多照明光源的前视内窥镜镜头开发。采用3D打印设计了一套多目内窥镜侧视镜头原型装置，该装置包含2个CCD(Charge Coupled Device)相机，在2个CCD相机的连线上配置照明点光源。其次，在常规前视内窥镜镜头的基础上设计开发了具有4个照明光源的内窥镜前视镜头，该镜头前端长10mm，直径6mm，4个照明光源可以分别进行调整，实现对照明光强分布的控制。.（2）二维图像拼接算法研究。本部分工作用于发动机内部大范围二维图像获取。以SIFT(Scale-invariant feature transform)算法为基础，提出了一种新的特征点匹配算法。该方法首先在特征点特定邻域范围内计算每个像素的梯度模值与方向信息，极大提升了匹配效率。实验表明，相比于ANN(Approximate nearest neighbor)算法，本方法在能够获得90%以上匹配点数的前提下，减少大约60%的时间消耗。.（3）“投射光强—图像灰度”数学模型研究。本部分工作用于发动机内部三维图像重建。基于SFS技术，研究点光源下光源与待测表面、待测表面与成像透镜、成像透镜与CCD像元灰度之间的光线传播与灰度变化规律，获取从光源发出的光线经待测表面反射后进入相机并被感知输出的全过程，建立精确的“投射光强—图像灰度”数学模型，通过求解该模型即可完成表面信息的高精度获取。.（4）光源信息标定研究。光源信息是高精度表面重建的基础，包括光源位置与光源主光轴方向。研究了基于标准球的快速光源位置标定方法。在此基础上，获取点光源照明下的灰度图，由“投射光强—图像灰度”计算出其理论灰度，通过最优化方法确定出光源主光轴。.（5）实验分析。采用所开发的内窥镜镜头原型及相关技术，对航空发动机管路内表面进行了测量与分析。