Edge scanning and swept surface approximation in reverse engineering

Schreve, Kristiaan (2001-12)

Thesis (PhD)--University of Stellenbosch, 2001.

Thesis

ENGLISH ABSTRACT: Broadly speaking Reverse Engineering is the process of digitising a physical object and creating a computer model of the object. If sharp edges formed by two surfaces can be extracted from a point cloud (which is the set of measured points) it can speed up the segmentation of the point cloud and the edges may also be used to construct swept surfaces (or various other types of surface that best captures the design intent). A strategy is presented to "scan" edges. The strategy simulates a CMM (Coordinate Measurement Machine) as it would scan a sequence of short lines straddling the edge. Rather than measuring on a physical object, the algorithm developed in this dissertation "scans" on the points in the point cloud. Each line is divided in two parts, or line sections, belonging to the surfaces fanning the edge. The points of the line sections are then approximated with polynomials. Each edge point is the intersection of two such polynomials. In many engineering components sharp edges are replaced with fillet radii or the edges become worn or damaged. This algorithm is capable of reconstructing the original sharp edge without prior segmentation. A simple analytical model was developed to determine the theoretically achievable accuracy. This Analytical accuracy was compared with the accuracy of edges extracted from point clouds. A series of experiments were done on point clouds. The input parameters of the experiments were chosen using the technique of Design of Experiments. Using the experimental results the parameters that most significantly influences the accuracy of the algorithm was determined. From the Analytical and experimental analysis guidelines were developed which will help a designer to specify sensible input parameters for the algorithm. With these guidelines it is possible to find an edge with an accuracy comparably with an edge found with the traditional method of finding the edges with NURBS surface intersections. Finally the algorithm was combined with a swept surface fitting algorithm. The scanned edges are used as rails and profile curves for the swept surfaces. The algorithms were demonstrated by reverse engineering part of another core box for an inlet manifold. If the edge detection parameters are specified according to the guidelines developed here, this algorithm can successfully detect edges. The maximum gap size in the point cloud is an important limiting factor, but its effect has also been quantified.

AFRIKAANSE OPSOMMING: In Truwaartse Ingenieurswese word 'n fisiese voorwerp opgemeet en 'n rekenaar model word daarvan geskep. Die segmentering van die puntewolk (dit is die versameling gemete punte) sal aansienlik vergemaklik word indien dit moontlik is om skerp rante in die puntewolk te identifiseer. Die rante sal dan gebruik kan word om veegvlakke (swept surfaces), of enige ander tipe oppervalk wat die ontwerp die beste beskryf, te konstrueer. Hierdie proefskrif beskryf 'n strategie wat die rante kan opmeet. Dit simuleer die manier waarvolgens 'n Koërdinaatmeetmasjien 'n reeks lyne, wat oor die rant lê, sou meet. In plaas van op 'n fisiese voorwerp op te meet, "meet" die algoritme op 'n puntwolk. Elke lyn word dan in twee dele verdeel (elke deel word 'n meetlynseksie genoem). Elke meetlynseksie behoort aan een van die twee oppervlaktes wat die rant vorm. Die rant punte word bereken as die interseksie van twee polinome wat deur die punte van die meetlynseksie gepas is. Dit is dikwels die geval met meganiese onderdele dat skerp rante vervang word met 'n vulstraal of dit kan ook gebeur dat die rant verweer het of beskadig is. Die algoritme, wat hier beskryf word, kan selfs die oorspronklike skerp rant in sulke gevalle herkonstrueer. 'n Eenvoudige analitiese model is ontwikkelom die teoretiese akkuraatheid van die algoritme te bepaal. Die teoretiese akkuraatheid is vergelyk met die akkuraatheid van rante wat uit puntewolke bepaal is. 'n Reeks eksperimente is op puntwolke gedoen. Die parameters vir die eksperimente is gekies deur van Eksperimentele Ontwerp gebruik te maak. Met behulp van hierdie tegniek kon bepaal word watter meetparameters die grootste invloed op die akkuraatheid van die gemete punte het. Die teoretiese en eksperimentele resultate is gebruik om riglyne daar te stel waarmee die intreeparameters van die algoritme gekies kan word. Met hierdie riglyne is dit moontlik om 'n rant te vind met 'n akkuraatheid vergelykbaar met die tradisionele metode om die rante te vind met behulp van NURBS oppervlakte interseksies. Laastens is die algoritme gekombineer met 'n algoritme wat veegvlakke deur punte kan pas. Die gemete rante word gebruik as spore en profiele vir die veegvlakke. Die tegnieke is gebruik om 'n CAD model van 'n sandkernvorm (vir die giet van 'n inlaatspruitstuk) te maak. Deur die riglyne te gebruik om die intreeparameters vir die algoritme te spesifiseer, kan rante suksesvol uit puntewolke bepaal word. Die maksimum afstand tussen naburige punte in die puntewolk beperk die gebruik van die algoritme, maar die effek hiervan is ook vasgevat in die riglyne wat ontwikkel is vir die algoritme.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/52250
This item appears in the following collections: