Research On Noiseproof Processing Methods For Cloud Data And Its Application In Turbine Blade Reconstruction

With high speed development of computer technology and robotized measuretechnique, reverse engineering has been a method which utilizes computer techniqueto simulate product and develop new product according to practicality. It plays a moreand more important role in manufacturing industry. The data we deal with in reverseengineering are noisy unorganized 3D data cloud, which bring great difficulty toreconstruction process. In this dissertation, artificial neural network (ANN) andwavelet methods are introduced to study some key technologies of reverseengineering for scattered noisy data cloud, which include segmentation, triangulationand fairing curve reconstruction. Based on the methods presented in the research,some turbine blade reconstruction algorithms are studied also in this dissertation. Thefollowing are listed works done on this topic:The application area and pivotal technology of reverse engineering areintroduced firstly, some problems occurred in current methods are analyzed. Then, themain research aim of this dissertation was put forward and the research contents weresummarized.Three kinds of segmentation methods in reverse engineering which areregion-based methods, Edge-based methods and clustering methods are compared. Animproved fuzzy Self-Organizing Feature Map Network (fuzzy SOFM) is presented atfirst. The algorithms which use this network to implementation segmentation andcharacteristic curve extraction are also given. Compared with other methods, this kindof improved network increases the convergence speed notably and can deal with noisydata directly. Furthermore, the result of segmentation is irrelevant to input order. Sincethe points on characteristic curve can also be extracted quickly according to thedegree of membership in this method, it provides a more complete segmentation resultthan region-based methods and edge-based methods. The validity of this method isproved by the real scanned point-cloud.Triangulation of scattered data is a very important problem in reverseengineering. Based on a kind of dynamic neural network—Growing Cell Structure (GCS), a new method of triangulation is presented. This method uses all noisy datapoints directly to implement triangulation and dispenses with smoothing and datareduction process, which may lead to loss of geometric character. Moreover, thismethod is of lower complexity than other triangulation and the amount of vertexes inmesh can also be calculated, so we can control the size of mesh in advance. At last, apractical example is given to prove its effectiveness.Some algorithms involved in the section-feature based reverse engineering arepresented in this dissertation. These algorithms include data sorting and contourseparating method for complicated section curve, comer detecting method, and fairingplanar curve reconstruction method. Some examples are given to verify theireffectiveness.At first, an algorithm for data sorting and contour separating based on polarcoordinates is presented. This algorithm can also be used realize contour tracking forimage. Secondly, a kind of corner detecting method is improved to be more veracious.In reverse engineering, problems of how to reconstruct a fairing contour curveare often met. A new fairing method for planar curves based on waveletdecomposition and Geometric Hermite Interpolation (GHI) is given in thisdissertation. We decompose curvature signal with wavelet at first and then extract itslow frequency coefficients as new smooth curvature signal, which is used to constructa G~2 continuous piecewise rational cubic Bezier curve with GHI method subsequently.Compared with other fairing methods, this method can dispose of data with noisedirectly and realize local fairing and integer fairing easily. The effect and efficiency ofthis method are also proved with some examples.Turbine Blade is seen as one key part of the aero-engine. Some importantalgorithms involved in blade surface reconstruction are studied. A method forcomputing skeleton curve of blade cross-section by means of offset curves is given.The problem of computing skeleton curve is converted into how to find theself-intersections of offset curves. A multiple step algorithm is presented to determineskeleton curve and some relevant important parameters. It is shown that this algorithmis of good precision, robustness and lower complexity than other methods. Theproblem of unstableness of computation is also avoided in this method.It has been proved that the performance of turbine blade can be improved if itsleading/trailing edge is an ellipse arc. A fitting algorithm based on the principle of least square and polar line geometric properties of ellipse is described. This newalgorithm is of good robustness and high accuracy for fitting an ellipse, especially asegment of ellipse. It can be used to reconstruct the leading/trailing edge of blade inreverse engineering and fit the ellipse arc with high precision. Compared with othermethod's extruding outliers manually, the most important advantage of this method isoutliers can be extruded automatically during the fitting process, which ensures thehigh precision of fitting.Based on above research, we model the leading/trailing edge with cubic rationalBezier curve. Three kinds of modeling methods for blade surface are compared in thisdissertation. An approximate arc length parameterization method which plays animportant role in blade modeling is also presented.In the summary of this dissertation, a number of advanced topics for featuretechnique in reverse engineering are addressed for the future research.