www.elsevier.com/locate/advengsoft

Advances in Engineering Software 37 (2006) 592–600

Adaptive restoration of complex geometry parts throughreverse engineering application

Jian Gao a,*, Xin Chen a, Detao Zheng a, Oguzhan Yilmaz b, Nabil Gindy b

a School of Mechanical and Electronic Engineering, Guangdong University of Technology, 729 East Dongfeng Road, Guangzhou 510090, PR Chinab School of Mechanical, Materials, and Manufacturing Engineering, The University of Nottingham, University Park, Nottingham NG7 2RD, UK

Received 7 October 2004; accepted 18 January 2006Available online 13 June 2006

Abstract

After a certain number of hours of running, no two mechanical components are completely the same due to normal wear or foreignobject damage. A nominal CAD model from a component designer is different from its corresponding worn one and therefore cannot bedirectly used for tool path generation for build up and machining repair processes. This is the main reason that most repair process usedfor complex geometry parts, such as gas turbine blades, is currently carried out manually and is called the ‘‘Black Art’’.

This paper proposes a defects-free model-based repair strategy to generate correct tool paths for build up process and machining pro-cess adaptive to each worn component through the reverse engineering application. Based on 3D scanning data, a polygonal modellingapproach is introduced in this paper to rapidly restore worn parts for direct use of welding, machining and inspection processes. Withthis nominal model, this paper presents the procedure to accurately define and extract repair error, repair volume and repair patch geom-etry for the tool path generation, which is adaptive to each individual part. The tool paths are transferred to a CNC machine for therepairing trials. Further research work is performed on repair geometry extraction algorithm and repair module development withinthe reverse engineering environment.� 2006 Elsevier Ltd. All rights reserved.

Keywords: Repair; Defects; 3D digitising system; Polygonal modelling; Adaptive tool paths; Reverse engineering

1. Introduction

Due to the long hours of run in operation and harsh envi-ronments, complex components such as tools, dies and gasturbine components may suffer from various defects, suchas distortion, wear, impact dents or under nominal dimen-sional limitations. As a result no part is completely the sameas any other one. A nominal CAD model from a componentdesigner is different from its corresponding ‘‘in service’’ oneand therefore cannot be directly used for tool path genera-tion for the build up process and machining process [1–4].Fig. 1 shows the cross-section comparison between a bladeCAD model and a used scanning model of this type. It is

0965-9978/$ - see front matter � 2006 Elsevier Ltd. All rights reserved.doi:10.1016/j.advengsoft.2006.01.007

* Corresponding author. Tel.: +86 20 3762 6604; fax: +86 20 3762 6109.E-mail addresses: gaojian@gdut.edu.cn, Jian.Gao@nottingham.ac.uk

(J. Gao).

clear that the used blade cross-section is distorted from theoriginal position, it will cause incorrect repairing if theCAD model is still used for tool path generation for additiveand subtractive processes. Therefore, to generate correcttool paths for the repair process for each worn componenta defects-free model needs to be created based on the compo-nent scan data. This model needs to have the same attitude orstance as the worn part and can work as a nominal modelused for adaptive welding, machining and inspection.

For most conventional CAM systems a surface modelcan be input and used to generate a NC tool path for free-form geometries. Much research has been focused on find-ing a better approach to generate tool paths effectivelyand accurately for freeform surface machining [5–8]. Atpresent, most reverse engineering (RE) software serves thefunction of creating NURBS (Non-uniform [knot sequence]rational B-splines) curves and surfaces. However, surface

Fig. 1. The cross-section comparison of a CAD model and a scannedmodel of its used blade.

J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600 593

generation in a RE or a CAD system requires a major com-mitment in time, both in training and in operation for mostreal jobs [9]. When modelling complex shapes, the biggestproblem for RE users is determining a reasonable surfacepatch boundary layout where the (u, v) surface parameteri-sations of each patch are in reasonable alignment with theunderlying surface structure. Design options are improvedbut complexity is increased by the option of trimming offany portion of any surface using a trimming curve [9].Mainly due to the complexity and time consumed in creat-ing a high-quality NURBS surface model from scan data,the potential for a 3D digitising system is largely to date cur-rently untapped in most manufacturing organisations eventhough high-performance scanners have been around since1980.

Data Acquisition hardware(CMM, LaserScanner,

3D vision)

3D scan

Correct scanprocess errors

Repair of wodefect

Auto-align the dfree model with

worn part mo

repair profor we

cl

stage 2

stage 3

stage 1

Reverse Engineering environment

build-up proc

Fig. 2. Flowchart of the adaptive rest

This paper proposes a polygonal modelling approach torapidly restore worn parts for direct use of the machiningand inspection process. Compared with the complex proce-dures and difficulty in patch dividing in surface modelling,the polygon modelling can be both quicker and involve lesscomplexity. A rapid repair process can be achieved througha commercial CAM system such as Tebis and Delcam Pow-ermill for five-axis tool paths generation for machining.

With a defects-free polygonal model adaptive to its com-plex component geometry, it is possible to capture the max-imum/minimum errors of worn components and to realiseadaptive welding and machining.

In this paper, Section 2 presents the proposed repair sys-tem structure for complex geometry parts. Section 3 intro-duces the modelling approaches used in processing 3Dscanning data and creating defects-free polygonal modelsthrough a RE tool. Based on the reference model created,the method of repair volume determination is discussedin Section 4, and the procedure for some repair patchextraction are presented in Section 5. Conclusions are givenin Section 6.

2. Repair system structure

The structure of the proposal complex component repairsystem is illustrated in Fig. 2. In the acquisition phase, con-sidering the factors of part size, accuracy, scanning speed,data processing speed, and restored part data quality forinspection, a GOM ATOS II non-contact digitising systemis selected to scan the worn parts for repair and the

data of wornparts

rn parts

efect- the

del

Determine maximumrepair error, profiles

and patches

file extractionlding/laser

adding

repair patchextraction for

machining

Create defects-freemodel for repair

ess machining process

oration for complex components.

594 J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600

repaired parts for inspection. The ATOS topometricmetrologies are based on the principle of optical triangula-tion by means of structured illumination. The measurementsystem consists of projecting different fringe patterns ontothe object’s surface using a white light projection unitand capturing these patterns by two integrated camerason either side of the sensor head. The ATOS software cal-culates precise 3D coordinates of up to 1.3 million pointsper view. All measurements are automatically transformedinto a common coordinate system. The complete 3D dataset can then be exported using standard file formats foreasy post-processing.

Based on the scan data, a modelling approach is devel-oped to create a defects-free polygonal model, which iscrucial for the definition of repair patch and tool path gen-eration. The model is also important for the geometricdimension and tolerance (GD&T) inspection to be per-formed in the RE environment. The repair proceduremay be summarised as follows.

Stage 1: Automatically correct scan data error caused bydigitising process, edit each defect, re-sample thedefect patch and create a polygonal model with-out defects. This is the key stage for the wholerepair process.

Stage 2: Automatically align the edited model with theworn part scan data and perform comparison.Then extract the maximum error of the worn partdefect and identify repair patch.

Stage 3: According to the repair patch geometry, extractrepair profiles for the part build-up process andrepair patch/geometry for the machining process.

3. Polygonal modelling from reverse engineering

Reverse engineering (RE) is mainly used to reconstructgeometric shapes from physical parts via data acquisitionusing a CMM, a laser scanner, or other forms of data dig-itisation. RE is defined as ‘‘systematic evaluation of a prod-uct with the purpose of replication. This involves eitherdirect copies or adding improvements to existing design’’[10]. Normally, RE is accomplished in the following threesteps: part digitising, feature extraction and CAD surfacemodelling. Once parts are digitised through various contactor non-contact digitisers, segmentation techniques are usedto identify boundaries and surface segments from the cloudof points acquired. Based on the output of the segmenta-tion process, surface fitting by NURBS or Bezier surfacesis performed. Currently, RE is mainly used to re-engineera component to get its CAD model for Rapid Prototype(RP) or Rapid Manufacturing (RM). Since a surface modelas a fundamental geometry representation has beenemployed by many CAD/CAM systems, the majority ofRE applications focus on surface modelling to suit thesesystems [11,12]. However, data segmentation in most REsoftware is done by manual operation, and so is the surface

modelling. The operator defines the type of surface to befitted to each segment of the data. It not only takes a longtime to perform this procedure, but also undoubtedlyaffects the modelling repeatability and accuracy [13].

As stated previously, a CAD model cannot be directlyused in the part repair process, a repair model needs tobe created to adapt for part-to-part variations. Unlike mostRE application, the RE tool in this repair system is used torestore the worn part model and to extract repair geome-tries used for tool paths generation for a laser claddingand machining process. This section presents the restora-tion procedure of worn part defects and introduces theapproach to create a defect-free polygonal model for fur-ther repair process.

3.1. Repair of holes and gaps

In the application of complex components refurbish-ment, it is necessary to scan a component in various direc-tions and then align these scan data sets into one completemodel. However, errors in the scan data may occur in thedigitising process due to problems of unexpected light dis-turbance, sharp edge and invisible occluded surfaceregions. Also, the alignment process may cause abnormaltriangles due to data overlap and holes or gaps between dif-ferent data patches. Therefore, it is important to pre-pro-cess the scan data and derive a corrected scan model. Inpractice, this editing process can be realised by means ofreverse engineering software, such as Polyworks, anderrors from the original scan data can normally be elimi-nated [14]. The process procedure for correcting the origi-nal multi-view scan data is outlined by the following steps.

Step 1: Alignment of multi-view scan data sets.Step 2: Delete abnormal topological triangles over the

whole model.Step 3: Fill holes and gaps.Step 4: Optimise and subdivide the model, save as a STL

(stereolithography) file.

However, when holes or gaps on the model are too bigor in a difficult position, the automatic operation mayresult in an abnormal polygon patch over the area. Torestore the holes or gaps properly and accurately, a Beziersurface can be used to construct the repair area. By meansof fitting and re-sampling the surface patch to the model, aproper polygonal model can be created for further use.Fig. 3 shows an example of the restoration of a large holeon a blade edge through a surface fitting and re-samplingprocess. Fig. 4 shows the restoration result of the auto-holefilling and the Bezier filling approach.

3.2. Repair of component wear or impact defects

After long-hours serves in a harsh environment bladesmay suffer from defects, such as distortion, cracks,nicks and dents. Some blades may be under dimensional

Fig. 3. Restoration of blade edge (unscanned hole) by fitting a Bezier surface on the top of it and re-sample to polygon mesh.

Fig. 4. Restoration comparison. (a) Automatic hole filling and (b) create a surface and resample.

Fig. 5. A worn blade with the defects of dent a, b, and nick c.

J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600 595

limitations due to wear. Fig. 5 shows an example of a wornblade with two dents (a and b) on the blade edge and onenick defect c on the top corner. A nominal CAD model

from a blade designer is no longer valid for use in tool pathgeneration. To generate tool paths adapted for each part forlaser cladding and machining process, a defects-free polyg-onal model needs to be created based on the actual scandata of the worn part. Compared with the complex proce-dures and difficulty in patch dividing in surface modelling,polygonal modelling can be much quicker and less complex,and most important, the model can be directly used forother repair processes, such as repairable decision, repairerror/volume determination and repair patch geometryextraction.

The procedure of defect repair is carried out by the fol-lowing steps (based on the Polyworks software) [14]:

Step 1: Subdivide the mesh triangles to get a higher resolu-tion model by the function Subdivide.

Step 2: Delete the triangles in the defect area.Step 3: Layout a Bezier surface over the deleted patch or

the created hole and fit the surface to the model.Step 4: Fill the patch by triangulating the model using the

fitted surface.

Fig. 6. Repair of defects on blade edge with different approaches. (1) Repair – defect a, (2) Repair – defect b and (3) Repair – defect c.

596 J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600

Step 5: Optimise the mesh triangles orientation using thedefault parameters.

With above repair steps, defects on blade airfoil can eas-ily be restored, because the Bezier surface can be createdand fitted on the rest of the triangular mesh when the repairpatch mesh is deleted on the model. When the defects occuron blade’s leading edge or trailing edge, the Bezier surfaceneeds to be created carefully over both sides of the edge.Fig. 6(1) and (2) shows the restored defect a and b on eachedge of the blade.

For a defect on the blade corner, such as the defect c,there will be a problem in creating a Bezier surface overthe defect patch because the surrounding repair patch isinsufficient to create a surface covering the patch. In thiscase, a Bezier surface below the defect patch needs to becreated and fitted to the model first, then the stretch oper-ation is used to extend the surface created over the defectarea along the blade’s sweeping trend. Once a suitable sur-face is created over the repair area, the next step is to re-sample the surface to triangular mesh and to optimise thewhole model mesh. The defect c was repaired by meansof the above procedure and is shown in Fig. 6(3).

Fig. 7. Comparison of the worn blade with the defects-free polygonalmodel created previously.

4. Repair volume determination

4.1. By model comparison

Since the reference model created is based on the wornblade model, it has the same coordinate system andsame stance as the worn blade and the two models canthus be aligned automatically without interactive mani-

pulate. Through model comparison, the maximum/mini-mum errors of the worn part and repair patch geometriescan be obtained and extracted by means of a properalgorithm.

Fig. 7 shows an error colour map produced by the com-parison of the worn blade shown in Fig. 5 and the defects-free model created previously in RE. The worn areas of thepart are shown in a colour map and the maximum/mini-mum errors and the error coordinate position can beobtained directly. When a part has several defects to berepaired, a local area comparison can be performed toget the local maximum/minimum error and coordinatefor each repair defect. Fig. 8 shows the error map for the

Table 1Comparison report (unit: mm)

Report type Cross-sections

Cmp Dist 4.000 Err Dir Shortest distance

Name Index #Points Mean StdDev Maximum error Minimum error Profile of a line %Out HiTol %Out LoTol

Cross-section 1 1 717 0.000688 0.073654 0.451805 �0.309660 0.761465 6.694561 8.786611Cross-section 2 2 1038 0.006012 0.047987 0.248329 �0.201839 0.450169 6.069364 9.537572Cross-section 3 3 1119 �0.000171 0.005347 0.028050 �0.036543 0.064593 0.000000 0.000000

Fig. 8. Error map and boundary curves for the defect b with the minimum error of �0.145 mm and the coordinate of (�20.789, 1.673, �2.055).

J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600 597

defect b of the blade model, where the minimum error andcoordinate position are obtained. This maximum/mini-mum repair error determined the additive material volumeof a build up process.

4.2. By cross-section comparison

To get a general error comparison result for each defecton a worn part, a cross-section comparison can be used. Inthis mode three cross-sections were created over the bladedefects a, b and c described previously and a correspondingcomparison report was produced directly which containsdefect error information. As listed in Table 1, it is possibleto view repair defects in detail.

5. Extraction of repair geometries

Once a worn part is restored geometrically, the repairvolume/error for each defect can be derived as discussedin previous sections, then the repair geometry needs to beextracted for material deposition process, such as a lasercladding process and for material subtraction, such as amilling, grinding/polishing process. In order to generatesuitable tool paths and realise adaptive repair, the repairgeometry needs to be defined and extracted individuallyto suit the part-to-part variation. It is obvious that complexparts with different repair defects will require different

geometries extracted for the additive process and machin-ing tool path generation. Several defects for repair are dis-cussed in the section.

5.1. Blade tip repair

For blade tip repair, tool paths for a laser claddingprocess can be generated by the blade tip profile extractedfrom the restored model. According to blade size, hollow/solid state blade and additive process parameter setting,such as feed rate, flow, speed, etc., the profile requiredfor a build-up process can be the middle curve of the tipcontour or the tip contour for one path or multi-pathsgeneration. Through the profile extraction programmedin the RE environment, blade tip profile and mid-curvefor the build-up process can be generated automaticallybased on the polygonal model and can be output as IGESformat file. Fig. 9(a) shows the repair profile for thin bladetip repair. Since the repaired model is created based on theworn part geometry the profiles extracted are adapted tothe repair part, adaptive tool paths can therefore be gener-ated for the build up process.

Based on the defects-free polygonal model it is possibleto generate tool paths directly for 5-axis machining processthrough a CAM software, such as Tebis and PowerMill.For some surface-model based CAM software, such asCATIA, Pro/Engineer and MasterCAM, it is also possibleto create a surface patch for the tip area machining.

Fig. 9. Repair profile and patch created for build-up and machining process: (a) additive profiles, (b) machining patches (polygon/surface) and (c) toolpath for tip machining process.

598 J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600

Fig. 9(b) shows the tip machining patches and Fig. 9(c)shows the tool path generated for the blade tip machiningprocess.

5.2. Blade edge repair

For blade edge repair, the repair patch along the edgeshown in Fig. 10(a) is a suitable feature for the build-upand machining process. Once the worn part is built up, amachining process is used to machine it back to size.Depending on the weld bead volume on the restored part,corresponding machining processes, such as milling, grind-ing or/and polishing processes, are selected. It may be nec-essary to digitise the welded part again to determine themaximum material to be removed and to optimise toolpaths. Through the comparison between the welded partand the reference model, maximum removal error can beobtained. However, if the build up process can stablydeposit a certain amount of material on the repair area,then the machining process can always offset a certain dis-tance as the thickness of the weld bead and generate toolpath for machining. Fig. 10(b) shows the tool paths gener-ated for the edge patch repair.

Fig. 10. Repair patch defined for edge restoration (a) edge rep

5.3. Surface repair

For freeform surface repair, such as gas turbine bladeairfoil and forging die surface, the repair patch for buildup process and machining process needs to be definedand extracted for tool path generation. Since the referencemodel inherited the actual blade geometry and attitude, thetool paths created on this model are adaptive to the wornarea and the adaptive cladding and adaptive machiningcan thus be achieved.

In order to realise a near-net-shape welding process forsurface defect repair, it is necessary to get the repair areageometry (boundary) and precisely define tool paths basedon each layer’s boundary. Through a developed algorithm,a multi-layer boundary corresponding to the repair patchgeometry is generated automatically based on the resultof error comparison. The algorithm of additive curvesextraction for the repair model is implemented in the Poly-works environment and described as follows:

Step 1: Generate comparison points {CP} from the com-parison error map and convert them into pointprimitives.

air patch created (b) tool paths generated for edge repair.

Fig. 11. Repair geometry for blade airfoil defect.

Fig. 12. Tool path for surface machining is generated based on theboundary of repair volume and shape to reduce machining time.

J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600 599

Step 2: Create a base plane Pb by fitting a plane to thecomparison points {CP}.

Step 3: Extract the maximum error for the repair defectdmax, set Dd = dmax /n; for the repair geometry, aseries of planes parallel to the plane Pb are createdat dmax, dmax � Dd, dmax � 2Dd, . . . ,dmax � nDd,the corresponding errors are denoted as e0,e1, . . . ,en and the created planes are represented as Pi

(i = 0,1,2, . . . ,n).Step 4: Let i = 1.Step 5: Select points from {CP} within the area {Pi�1Pi},

these selected points correspond to errors within{ei�1,ei}.

Step 6: Project the selected point onto the plane Pi and fit aclosed curve to these points.

Step 7: If i < n, then i = i + 1, go to step 4; otherwise, go tostep 8.

Step 8: Optimise and output created curves in *.igs format,and program end.

By means of the algorithm, the additive curve featurecorresponding to each layer can be created for an additiveprocess. Fig. 8(b) shows an example of the additive 3-layercurves created. These curves are useful for a welding pro-cess to precisely define tool paths, especially when therepair area is irregular. A one-path or multi-path will bedetermined in a CAM system by the factors of feed rate,flow, speed parameters.

For a machining process, a repair area boundary and asurface patch are normally used to generate tool paths. Sim-ilar to the build up process, the maximum error between thenominal part and the welded part is useful to the optimisa-tion of the machining tool paths. When the repair geometryis irregular, the repair area boundary to each level can becreated to realise an efficient machining strategy. The algo-rithm can be developed for the repair system with a similarprocedure as the additive boundary generation.

Fig. 11 shows an example of a repair surface with a weldbead. Through error comparison, a maximum error (weldbead height) can be obtained and be used to determine

the additive material volume. Several curve boundariesare generated for the tool path creation. Fig. 12 showsthe tool paths generated on the boundary of the repair vol-ume and shape.

6. Conclusions

This paper has presented a reverse engineering-basedadaptive restoration approach to recreate worn parts andextract repair geometry for the build up process andmachining process. By means of polygonal modellingapproach, defects on worn parts can be repaired andrestored to nominal geometries. The restored part geome-try possesses each 10 part’s stance and can be used to gen-erate tool paths adaptive to part-to-part variation. Withthe restored model, repair maximum error/volume can bederived through the comparison with its worn part data.The approach presented therefore provides an advanced

600 J. Gao et al. / Advances in Engineering Software 37 (2006) 592–600

solution to solve the part-to-part variation repair problemand to realise rapid repair for complex geometry parts.Compared with the complex procedures and difficulties insurface modelling, polygonal modelling can be bothquicker and less complex. Further work is carried out onimplementing of the adaptive machining strategy forthin-blades and repair system information integration.

Acknowledgement

This research work is supported by the Natural SciencesFoundation of Guangdong Province, PR China (GrantNo. 04009491 and Team-Project No. 05200197) and theUK Engineering and Physical Sciences Research Council(EPSRC) under Nottingham’s IMRC (Grant No. GR/R67019/01).

References

[1] Gaumann M et al. Single-crystal laser deposition of superalloys:processing – microstructure maps. Acta Mater 2001;49:1051–62.

[2] Isotek Aerospace welding applications, Available from http://www.isotek.co.uk.

[3] Stimper B. Using laser powder cladding to build up worn compressorblade tips, Available from http://www.sg.daimlerchrvsler.com/Pro-iects/wi/channel/files/11766.pdf.

[4] Gao J, Folkes J, et al. Investigation of a 3D non-contact measure-ment based blade repair integration system. Aircr Eng AerospTechnol 2005;77(1):34–41.

[5] Park H. An approximate lofting approach for B-Spline surface fittingto functional surfaces. Int J Adv Manuf Technol 2001;18:474–82.

[6] Gregory J. N-sided surface patches. In: Gregory JA, editor. TheMathematics of Surface.

[7] Farin G. Curves and surfaces for computer aided geometric design: apractical guide. second ed. New York: Academic Press; 1990.

[8] Weir D, Milroy C. Reverse engineering physical models employingwraparound B-Spline surfaces and quadratics. Proc Instn Mech Eng,Part B: J Eng Manuf 1996;210:147–57.

[9] Besl P. Hybrid modeling for manufacturing using NURBS, polygons,and 3D scanner data in IEEE International symposium on circuitsand system, ISCAS, 1998.

[10] Bardell R, Balendran V, Sivayoganathan K. Accuracy analysis of 3Ddata collection and free-form modelling methods. J Mater ProcessTechnol 2003;133:26–33.

[11] Gao J, Gindy N, Chen X. Automated GD&T Inspection Systembased on Non-contact 3D Digitisation. Int J Prod Res 2006;44(1):117–34.

[12] Chen LC, Lin GCI. Reverse engineering in the design of turbineblades – a case study in apply the MAMDP. Robot Comput IntManuf 2000;16:161–7.

[13] Motavalli S. Review of reverse engineering approaches. Comput IndEng 1998;35(1–2):25–8.

[14] Polyworks-IMInspect comparison and verification software, refer-ence guide version 8.0 for windows. July 2003: InnovMetric SoftwareInc. Canada.