Research On Key Techniques Of Maintenance-orientated Maintenance Guiding

Maintenance support for civil aircraft is an intensive and difficult labor. The eternal problemconcerned by Airline Company is how to reduce difficulty, improve efficiency, avoid error duringmaintenance and compress direct cost. Thus, research the theory and technology on the problem issignificant from theoretical and technological perspective. This thesis aims at construction ofprocess-oriented maintenance guiding system to make an intensive study on informative enforcementand information organization that are two inevitable contents for constructing the mentioned system.Functions of the system are to provide suggestive information including texts&graphics that arebased on maintenance task and process so that maintenance personnel can improve quickly theircognitive competence and understand easily maintenance task. Main contents and innovativehighlights of this research are now generized as follows:(1)According to the fact that rotation matrix is SO3manifold, this thesis proposes the novelalgorithm of estimation object pose based on optimization over manifold under the condition ofestablished2D-3D correspondences. The approach exploits Taylor Expansion and retraction mappingto eliminate the principal deviation imported by the substitution orthogonal projection model with thatof pin-hole projection and degrade the propagation of accumulated error when pose tracking inconsecutive video images. Experiments show that object pose obtained by optimization on manifold isbetter at precision and the number of iteration than orthogonal iteration,i.e., when noise varianceequals1, the relative deviation of new algorithm with synthetical datum is0.5%for rotation axis and0.25%for rotation angle respectively, which are only fifth and half of those for orthogonal iteration.In industrial environment, the object, generally short of point features on its surface, but with coplanarcircles, represents a category of industrial object. Thus, this thesis deliberates on how to constructobject coordinate frame and estimate its pose from coplanar circles using projective theory of absoluteconic. Experiment on real object illustrates that the relatively maximum errors obtained by coplanarcircles are respectively: is only2.7%for rotation axis,1.7%for rotation angle and1.7%fortranslation. Considering the difficult problem of retrieving3D coordinates for feature points,reconstruction of3D coordinates for stable matched SIFT points is showed as an applied example ofpose estimated from coplanar circles that is full of engineering significance.(2)Tracking dynamically object pose is the prerequisite for image registration. This thesisimproves object pose tracking with straight lines and proposes a tracking algorithm based on coplanar circles. Contour's location in previous frame is utilized to calculate normalize cross correlationefficiency of image gradient for search contour points implicitly. In order to promote searchingefficiency, Extended Kalman Filter (EKF) is used to predict object pose in current frame on one handand limit searching range and departure location. For object with straight line, Tukey Estimator andweight line fitting are exploited to construct objective function and obtain object pose via SVD.Experiments show that line-base tracking approach in this thesis can achieve the performance thatmaximum mean error for rotation angle is0.29o, absolute angle error is1.90o, and tracking speed isabout10frames per second. This explains that the comprehensive performance of ours is better thanthe LM-based one. For object with coplanar circles, Samposon distance becomes the standard ofremoving outliers and heteroscadastic regression is imported to improve accuracy of ellipse fitting forresolving object's pose. Experiments for coplanar circles' tracking illustrate that the maximumdeviation is1.4ofor angle,3.5mm for translation and tracking speed is10~12frames per second.Evidences' from experiments verify the effectiveness of using EKF that promote tracing rate morethan100percent.(3)Calibration of optical See-through Helmet Display (HMD) is a fundamental work inimplementation Augmented Reality. This thesis improves the calibration method and eliminates thedisadvantage proposed by Azuma. Basing on proceeding research perform, we firstly substituteelectromagnetic tracker with camera when calibrating to obtain HMD's extrinsic pose. Secondly,3Dreconstruction is used to determine the transform relationship between world coordinate system andobject ones which exempt the procedure from expensive calibration workbench; at last, intrinsic&extrinsic parameters are individually calibrated and the hypothesis that extrinsic parameters isconstant to reduce the burdens who takes parting in calibration. Evaluation of calibration gives us thatthe single maximum error is only2.15pixels that shows the promoted calibration method is effective,feasible and the hypothesis is right. Subsequently, this thesis studies how to calculate the posetransformation and scale factor during obtaining3D models for pose tracking.(4)In maintenance environment, enhanced instructions and human machine interaction (HMI)can exempt maintenance personnel from the defect of switching attention focus repeatedly. Because ofthe background, this thesis designs and implements a menu-typed HMI to modify instructions'registration. At first, the HMI translates maintenance task instructions into two layer menus anddisplays menu items on HMD as virtual image. Secondly, virtual images that act as focus and acceptclicking are fixed on object's surface with the help of object pose tracking. At last, finite state machineis utilized to drive relationship between gesture input and alteration of menu content. Experiments illustrate the success rates of the proposed HMI achieve76%, and input capacity achieves8~10frames per second.(5)Aiming at implement prototyped system of process-orientated, this thesis interprets how toobtain instructions and2D images for enhanced information from existing technical publications andpropose to construct necessary information for3D enhancement and object pose tracking; then thinkssome practical problems when actually implementing the system. Demo system is implemented viainstallation of oil warm-up torch on APU to exhibit intuitively the research objectives of this thesisand the effect of process-orientated maintenance would be taken in civil aircraft maintenance fields.