Int J Adv Manuf Technol (2006) 31: 374387 DOI 10.

1007/s00170-005-0201-z

ORIGINA L ARTI CLE

M. Mahesh . Y. S. Wong . J. Y. H. Fuh . H. T. Loh

A Six-sigma approach for benchmarking of RP&M processes

Received: 21 July 2004 / Accepted: 14 June 2005 / Published online: 3 January 2006 # Springer-Verlag London Limited 2006

Abstract This paper presents a methodology of using sixsigma quality tools for benchmarking of rapid prototyping & manufacturing (RP&M) processes. It involves the fabrication of a geometric benchmark part and a methodology to control and identify the best performance of the process to reduce the variablity in the fabricated parts. The approach is demonstrated with a case study based on the direct laser sintering (DLS) process for prototyping using plastic powder. In the case study an identified set of six-sigma/ statistical process control tools is employed to determine and best tune factors affecting the desired outcomes of the built parts. Keywords Benchmarking . Decision support . Process characterization . Process optimization . Six-sigma Abbreviations GBP: Geometric benchmark part . MBP: Mechanical benchmark part . BL: Blend . CF: Chamfer . CB: Cube . CN: Cones . CH: Cylindrical holes . FB: Flat beam . FL: Fillet . FF: Free-form features . HC: Hollow cylinders . HS: Hollow squares . SB: Square base . SC: Solid cylinders . SH: Small holes . SL: Slots . SP: Spheres . TC: Thin cylinders . TS: Thin slots . TW: Thin walls . RP&M quality characteristics GA: Geometric accuracy (deviation measured in mm) . SR: Surface roughness, Ra in m . Control factors:LP: Laser Power in W . LT: Layer thickness in mm . PBT: Part bed temperature in C . SS: Scan speed in mm/s . Compensations:L: linear dimension (length, breadth or height in mm) . La: actual design dimension of STL file in mm . Lm: measured dimension on fabricated GBP in mm . k: scaling factor . b: laser beam offset factor in mm . s: shrinkage in mm
M. Mahesh . Y. S. Wong (*) . J. Y. H. Fuh . H. T. Loh Department of Mechanical Engineering, National University of Singapore, Singapore, Singapore e-mail: mpewys@nus.edu.sg Tel.: +65-68742221 Fax: +65-67791459

1 Introduction
Benchmarking aims to identify the best achievable practices and processes. For rapid prototyping and manufacturing (RP&M) several benchmarking studies have been reported. Kruth [1], Gargiulo [2], Lart [3], Juster and Childs [4, 5], Ippolito et al. [6, 7], Reeves & Cobb [8], and Shellabear [9] have proposed geometric benchmark or test parts for comparative evaluation of RP&M processes and systems. However, these studies have not considered the use of the benchmark part for development of a procedure for process evaluation/ optimization. Jayaram et al. [10] emphasized the importance of benchmarking standards to assist in the selection of suitable systems based on the applications. Wong et al. [11, 12] presented details on evolving generic benchmarks parts used in the evaluation of RP&M systems. Additionally several process planning techniques [1317] have also been reported for RP&M. This paper presents an approach to identify the best achievable RP&M process to serve as a benchmark process using a case study involving the DLS RP&M process. Camp [18] discusses investigations that are to be conducted to benchmark process against industry practice. The proposed approach is based on the use of sixsigma tools and a geometric benchmark reported [11, 12]. The six-sigma method defines and uses tools for process investigations and improvements [19]. 1.1 Geometric benchmark A geometric benchmark part (GBP), shown in Fig. 1, has been designed for the process evaluation of RP&M processes with the aim to identify a process benchmark based on the geometric accuracy (GA) and surface roughness (SR) as quality characteristics of the geometric benchmark part built by the process. The GBP is used to test the capability of a RP&M process/ system in fabricating prototypes [11, 12].

375 Fig. 1 GBP used in the parametric optimization

Fig. 2 Proposed methodology of RP&M process benchmarking

376

2 Methodology
The objective is to minimize process inconsistencies and defects of fabricated and to identify a best process/ procedure to achieve desirable geometric accuracy and surface roughness. The approach is presented with a case study based on the direct laser sintering (DLS) process (See Appendix for brief description of the process). The steps involved are presented in Fig. 2. 2.1 Process analysis (step 1) The first step in the proposed approach is process analysis. The purpose of the process analysis is to understand the entire process, including the critical relations between the quality requirements and the performance metrics of both input and output conditions [2022]. For this case study the aforementioned geometric benchmark part is used to serve as a de facto standard part for fabrication across the different RP&M processes [23, 24]. The initial understanding and selection of factors in the order of their importance aims to reduce inefficiencies in the process. A six-sigma tool - the Ishikawa fish bone diagram [22] is used to visualize potential causes for variability in the DLS process as shown in Fig. 3. This diagram presents a pictorial display of all possible causes of problems, or factors needed for a successful process benchmark. It facilitates the understanding of the factors from the part design to the final measurements that affect the process by indicating the relationships between the factors. 2.2 Screening experiments (step 2) For the screening experiments a number of preliminary tests were conducted. From the first few experiments and the outcomes, further tests were conducted on a trial-anderror method using various parameter settings with the aim to improve the GBP fabricated, and identify the key factors.
Fig. 3 Ishikawa fish bone diagram

Table 1 Characteristics of the nylon powder used in the experiment Type Size of the powder particles Recommended preheating temperature Recommended layer thickness Nylon polymer 100m 60C 0.2mm

The material used in the fabrication of the RP&M prototype is one of the key factors as the rapid prototyping process is material-dependent. The material used in the case study has the characteristics shown in Table 1. The following factors were identified as important. Selection of a suitable base plate, Preheating for fabrication of the prototypes, Proper sequential movement of the scrapper, Laser power distribution for properly sintered layers, Layer thickness control to achieve good surface roughness.

2.2.1 Base plate In the fabrication of parts using the DLS process, a good base plate is important for sintering both metal and plastic powders. The necessity of a base plate for the DLS system is due to the scrapper powder laying mechanism used. The base plate is mounted on the building platform using suitable fixtures. Various types of base plates used for the DLS process are listed in Table 2. Fig. 4 shows the choices of the base plate available. 2.2.2 Preheating Initially the powder was directly sintered without preheating. The result is shown in Fig. 5. After only a few layers, there was bulging, giving rise to cracking. Fig. 5a and b show respectively the side and top views, of the bulging on a built GBP during the screening experiments. Pre-heating for the plastic material before the actual sintering process

377 Table 2 Relative rating of the base plates used Base plates Resistance to temperature Bonding of base layers to the plate Relative rating Steel Very good Very poor 4 Aluminium Good Poor 3 Polymer Poor Good 2 Wood Good Very good 1

reduces cracking and is therefore considered essential, although the actual degree of pre-heating (level of part-bed temperature) could again possibly influence the final prototype. 2.2.3 Sequential movement of the scrapper Even after proper base selection and preheating of the plastic powder, defects could still be observed when building the benchmark parts, such as that in Fig. 6 showing a few shifted or gradually broken layers.
Fig. 4 Choice of base plates for the fabrication on the DLS system

A careful investigation of the DLS powder laying system revealed that rubbing between the scrapper (Fig. 7) and the deposited fresh layer of powder caused the shifting and breaking up of the layer from the previously sintered layer. A solution to this problem was to change the sequence of operation between the scraper, the part-bed, and the powder bed mechanisms for the DLS system. The original sequence of operations is shown in Fig. 8. The part bed (3) descends by a layer thickness. The powder cylinder (2) ascends so that the scrapper (1) would wipe the powder to lay it smoothly on the part bed surface. This is followed by the return of the scraper to the initial position.

Fig. 5 Bulging ultimately causing the base to break

378 Fig. 6 Pictures of the distorted GBPs as a result of the rubbing induced by the scraper

The return of the scrapper to the initial position over the laid powder surface sometimes causes sintered layers of the part to break away. To prevent this, the sequence has been reshuffled as follows: After the scraper lays the powder on the part bed surface it stays there till the scanning/sintering is over; after which the part bed goes down, followed by the scrapper returning to the initial position. Then the powder bed goes up and the scraper lays a new layer of powder on the part surface. The process is repeated till the entire part is fabricated. This sequence for the DLS system is much better than the previous, since it avoids rubbing and thereby shifting or breaking of the layers.

2.2.4 Temperature distribution in the DLS process using plastic powder Generally for laser-sintered RP&M processes, proper temperature distribution of the part bed is crucial. Poor adhesion between layers is primarily attributed to uneven sintering, arising from poor distribution of temperature results in warpage, delamination, or burnt marks. The following subsections discuss those that are specific to the DLS process case study based on the plastic powder material.

Fig. 7 DLS scrapper deposition of plastic powder

Fig. 8 Original mechanism of powder deposition in DLS system

379 Fig. 9 Effect of the temperature on the various sides of the same GBP

Warpage Figures 9a and b show the effect of temperature concentration on the front and inner side on the GBP on the DLS system. It was observed that the distribution of the temperature was uneven for the sides of the same part, i.e., the side facing the operator and toward the inside of the system. When the exhaust fan was turned on during sintering, the warpage was found to be considerably less, and this was due to more even circulation of heat in the fabrication chamber arising from the suction induced ventilation. Delamination Poor bonding strength leads to bulging, delamination, or warping. Fig. 10 shows delamination (Fig. 10a) and warpage (Fig. 10b) because of the weak bonding between layers in the part built during the case study. Therefore, it was concluded that the best setting for the part bed temperature and the laser power must be identified and tuned to achieve the suitable bonding strength to prevent part failures.

Burnt marks The powder laying system (scrapper) in the DLS system sometimes causes laying of excess powder during its motion. During normal sintering, the powder tends to cause the surface to appear burnt or uneven. It can be seen from Fig. 11 that in such cases, both the geometric accuracy and surface smoothness are affected. It was found than a proportionate increase in laser power and part bed temperature could reduce the burnt marks by melting away the excess laid powder. 2.2.5 Identifying the important factors From the screening experiments a simple cause-and-effect matrix based on quality function deployment (QFD) [22, 25] on the DLS system is shown in Table 3. The numeric values 9/3/1/0 are relative weightages based on value judgment of correlations (0- no relationship, 1- possible relationship, 3- medium relationship, 9- strong relationship) between the control factors and the geometric

Fig. 10 Delamination and warpage due to weak bonding between layers

Fig. 11 Serious effect of burnt marks on the geometric features

380 Table 3 QFD on the DLS process Process outputs Importance Geometric accuracy 10 Surface roughness 8 Total 9 102 1 38 9 102 1 18 3 34 1 18 1 18 9 102 relationship

No Process Input (control factors) Correlation: Input/output 1 Layer thickness 3 2 Part bed temperature 3 3 Laser power 3 4 Scan spacing 1 5 Scan speed 1 6 Scan pattern 1 7 Scrapper speed 1 8 Post processing 3 0- no relationship, 1- possible relationship, 3- medium relationship, 9- strong

Table 4 Identification of control factors Important control factors Laser power Scan speed Scrapper Speed Layer thickness Part bed temperature Scan pattern Significant control factors Laser power Scan speed Layer thickness Part bed temperature

accuracy and surface roughness. The correlation is partly based on the understanding of the process, from the users of the DLS system and the screening experiments. The total weightages were calculated as the sum of the correlation value (9, 3, or 1) multiplied by the relative weightage of importance of the process outputs. For example, the total weightage for layer thickness = 3 x 10 + 8 x 9 = 102. Six-sigma quality tools like the fish-bone diagram and QFD were useful in understanding and analyzing the control factors that relate directly to the output. To have a clearer interpretation of the effect, a planned design of experiments (DOE) [20, 22, 26] was used to identify the best setting of control factors that will give the desired output. Table 4 summarizes the important factors from which the significant control factors were finalized from
Table 5 Undesirable end-results and errors in the DLS process Undesirable results (Influence during fabrication process) Warpage Delamination Serious effect-temperature dependant Serious effect-temperature, layer thickness, laser power dependant Serious effect-controlled by optimized parameter setting Serious effect-material dependant

the above matrix and subsequently used to design the experiments. This selection was based on the highest total weightage from the QFD and also their ease of control during the experiments. The scraper speed and scan pattern (cross-hatch) were kept constant throughout the experiments. From the screening experiments the crosshatch scan pattern was found to be most suitable for this particular nylon material and therefore maintained consistent through out the experiments. In this case study on the DLS process the scan spacing is maintained a constant at 0.2 mm, as we were interested in changing scan speeds. Outlining of each sintered cross section was also maintained consistent thorough out the experiments as it generally improves the model quality. The plastic material is assumed to be preselected and new powder is used in all the experiments. Post processing is additionally excluded in the list of important control factors because the GBP will only be partially post processed for the sake of geometric measurements. Table 5 lists the identified undesirable end-results and errors associated with the DLS process. The aim of this investigation is to identify an effective procedure during
Table 6 Taguchi's L 9(34) orthogonal array Exp. No Control factors* PBT 1 1 2 1 3 1 4 2 5 2 6 2 7 3 8 3 9 3 * - Levels of control factors Control factors Level 1 PBT 30 LP 20 LT 0.1 SS 1200 LP 1 2 3 1 2 3 1 2 3 Level 2 35 25 0.15 1400 LT 1 2 3 2 3 1 3 1 2 SS 1 2 3 3 1 2 2 3 1 Level 3 40 30 0.2 1600

Surface Roughness Shrinkage Errors Laser beam Laser compensations if needed error System set up Optimized setting of the system in general error Post processing Based on the post processing methods error

381

the fabrication process that will minimize the undesirable results. 2.3 Design of experiments (step3) Taguchi's orthogonal arrays are used to determine the setting of the key control factors for experimental determination of the response of the DLS process so that they can be subsequently tuned accordingly to obtain the best achievable result. Two quality characteristics, namely geometric accuracy and surface roughness, are used as response factors during the experiments. Table 6 presents the L9 (34) Taguchi's orthogonal array used in the experiments for the DLS system and also the levels of the control factors. 2.4 Fabrication (step 4) The fabrication was performed on the DLS system according to the designed experiments. During the process of fabrication potentially important activities in the DLS process, such as fixing and preparation of the base plate were also recorded carefully for process evaluation and identifying the best practice. Upon completion of the experiments, measurements were performed on the coordinate measuring machine (CMM) as shown in Fig. 12. The objective was to find the deviation in the X and Y-axis accuracy in general and additionally the deviation in accuracy of the various individual geometric features on the GBP. The measurements include perpendicularity, parallelism, concentricity, angularity, coaxiality, and flatness. It is important that all geometric features on

Fig. 14 Interaction plots for laser power and part bed temperature

2.5 Measurements (step 5)

Fig. 12 CMM measurements

Fig. 13 Main plots-data means for surface roughness against control factors

382

the benchmark part and the relative measurements such as perpendicularity and parallelism be recorded for purposes of comparison between various RP&M processes. The surface roughness of the base of the GBP was measured using the Rank Taylor Hobson surface texture measuring equipment. Table 8 presents the measurement details. The use of a part program for capturing automated measurements using the CMM helps to maintain accurate, efficient, and consistent comparison across various experiments.

2.6 Analysis (step 6) A statistical analysis software [Minitab] was used to analyze the measured data and obtain the main effects plots and interaction plots. The plots in Fig. 13 show the main effects for surface roughness on the base of the GBP. The main effects plot is most useful when several factors are involved. By comparing the changes in the means level, we can identify those factors that have the most influence on the response. Since the objective is to reduce the surface roughness (SR), or to have a good surface finish, it can be interpreted from the main effect plots that the optimized setting (lowest Ra) for prototyping parts from this particular plastic material will include a layer thickness of 0.15 mm, medium

Table 7 Analysis of the feature geometry

383 Table 7 Continued

laser power of about 25 W, scan speed of about 1400 mm/s, and a high part bed temperature of 40C. The interaction plots for the laser power and part bed temperature in terms of the means of the surface roughness are as shown in the Fig. 14. It can be seen from the plots that there is a reduced mean for surface roughness, when the part bed temperature is low at medium laser power. Table 7 shows the preferred settings of control factors based on the analysis of the mean dimensional errors affected by the respective factors for the specific geometric features on the GBP. The appropriate compensations in terms of laser beam offset and scaling factors to improve the geometric accuracy of the GBP is presented in the following section.

2.6.1 Process control and tuning After identifying the best achievable process and the corresponding setting of control factors to give the most accurate geometric parts possible, further process improvement can be taken. An important process improvement procedure is error compensation to attain the best possible accuracy. The error compensations can be incorporated using a scaling factor and beam offset. The compensation for errors related to the geometric accuracy are based on the statistical data available from the measurements done on the GBP. Analysis of the CMM measurement data indicates that the fabricated benchmark part has an accuracy deviation of about 4 %. This was attributed to be partly caused by the scanner (beam offset) and partly due to the shrinkage of the material.

384 Table 8 DOE: Deviation of various features on the GBP Geometric Feature Dimensional error Exp No 1 SP1 0.1391 SP2 0.0022 SP3 0.1245 SP4 0.0834 CR1 0.1385 CR2 0.1442 CR3 0.1456 CR4 0.1536 CR5 0.1232 CH1 0.1230 CH2 0.1346 CH3 0.1380 CH4 0.1275 SC1 0.0365 HC1 0.0411 CN1 0.0889 SC2 0.0567 HC2 0.0302 CN1 0.0092 SC3 0.0531 Relational measurements HS1 (60) 0.0436 HS2 (25) 0.0206 HS1 (60) 0.0654 HS2 (25) 0.0259 SB 0.0326 Symmetry SP 0.0495 Symmetry SP 0.0563 Perpendicularity 0.0812 Perpendicularity 0.2732 Parallelism 0.2396 Parallelism 0.4609 Coaxiality 0.0041 Coaxiality 0.3047 Angularity 0.0045 X- measurements not taken Exp No 2 0.1636 0.0925 0.1928 0.1094 0.1298 0.1159 0.1272 0.1362 0.1126 0.1127 0.1397 0.1412 0.1421 0.0441 0.0663 0.0361 0.0623 0.1240 0.0094 0.0336 0.0418 0.0234 0.0641 0.0260 0.0348 0.0515 0.0558 0.0420 0.0534 0.1825 0.3226 0.0141 X 0.0123 Exp No 3 0.1025 0.0427 0.1659 0.0875 0.1851 0.1480 0.1412 0.2187 0.1300 0.1261 0.1585 0.1636 0.1723 0.0749 0.0952 0.0587 0.0056 0.1279 0.0313 0.0014 0.0473 0.0256 0.0658 0.0313 0.0394 0.0502 0.0549 0.2464 0.0811 0.3041 0.5133 0.0124 0.2657 0.0012 Exp No 4 0.0788 0.0370 0.1572 0.0685 0.1318 0.1276 0.1366 0.1405 0.1090 0.1115 0.1224 0.1496 0.1518 0.0388 0.0655 0.0190 0.0643 0.1234 0.0579 0.0939 0.0424 0.0192 0.0634 0.0185 0.0315 0.0500 0.0519 0.2934 0.1730 0.4261 0.2841 0.0209 0.2998 0.0017 Exp No 5 0.0898 0.0112 0.0908 0.0679 0.1257 0.1171 0.1263 0.1310 0.1043 0.1068 0.1292 0.1449 0.1389 0.0165 0.0838 0.0022 0.1116 0.1619 0.0865 0.0951 0.0445 0.0205 0.0605 0.0271 0.0412 0.0501 0.0517 0.3802 0.2245 0.3668 0.4248 0.0067 0.3070 0.0085 Exp No 6 0.0744 0.0823 0.1214 0.0631 0.1422 0.1515 0.1394 0.1534 0.1182 0.1110 0.1310 0.1555 0.1701 0.1691 0.1414 0.0015 0.0881 0.1725 0.0266 0.0841 0.0439 0.0193 0.0690 0.0226 0.0478 0.0516 0.0554 0.0064 0.0342 0.5343 0.1871 0.0542 X 0.0218 Exp No 7 0.0787 0.0369 0.1252 0.0548 0.1356 0.1246 0.1119 0.1142 0.1023 0.1015 0.1229 0.1324 0.1400 0.0992 0.1265 0.0982 0.1474 0.1725 0.0146 0.0441 0.0722 0.0256 0.0643 0.0234 0.0423 X X 0.0846 0.0336 0.2164 0.2487 0.0239 X 0.0021 Exp No 8 0.0942 0.0245 0.1599 0.0603 0.1620 0.1265 0.1600 0.1721 0.1272 0.1257 0.1514 0.1652 0.1722 0.1158 0.1222 0.0119 0.2224 0.1995 0.0246 0.0533 0.0469 0.0235 0.0665 0.0234 0.0358 0.0498 0.0519 0.0533 0.1120 0.4475 0.2016 0.0625 X 0.0059 Exp No 9 0.1493 0.0256 0.1218 0.0602 0.1486 0.1626 0.1691 0.1578 0.1398 0.1292 0.1486 0.1432 0.1391 0.0493 0.0541 0.0889 0.0662 0.0592 0.0374 0.0683 X X X X 0.0311 0.0504 X X X X X X X 0.0042

The error in length can be compensated using a scaling factor k and a beam offset b. The values of k and b are calculated from La obtained from GBP CAD file and Lm measured from the fabricated GBP. The maximum base method is used to process the data, as it is convenient and accurate [29]. In this method the measurement data set is sorted in decreasing order, and then the shrinkage and beam offset are calculated from the maximum differences between the maximum and minimum [30]. The procedure is as follows: 1. Sort the data set in decreasing order fLai ; Lmi ji 1; 2; 3; 4; :::ng:

2. Get the average shrinkage, s s

n 1 X Lm1 Lmi 1: n 1 i2 La1 Lai

(1)

3. Determine the average beam offset b b

n Lm1 X La1 Lai La1 : n 1 i2 Lm1 Lmi

(2)

Shrinkage in layer sintering causes the prototype to be smaller whilst beam offset makes the prototype bigger. Thus it is important to compensate for accuracy by

385

feasible to comprehensively specify all possible geometric configurations and their corresponding s and b compensation values. Therefore compensations are applied based on the different calculated values of k and b based on length along the X and Y-axis, respectively. 2.7 Experimental verification (step 7) Figures 17 and 18 show the GBP before and after the application of the proposed approach on the DLS process. Tables 9 and 10 compare the selected accuracy details (i.e., length in mm) of two GBP fabricated before and after the implementation of the proposed six-sigma approach for benchmarking RP&M processes. From the measurement data, it can be seen that the geometric accuracy has improved in the GBP fabricated using the proposed approach. As can be seen from the Table 10 the best achievable accuracy based on this material is 0.2 mm deviation as compared to the design dimensions. 2.8 Benchmarked DLS process (step 8) The proposed approach aims to identify crucial factors and their corresponding settings to enable more consistent performance of the process, which can then be used as a benchmarked process. For the DLS system, this includes reshuffling of the sequential operation of the scrapper, part bed, and the powder bed cylinders. Pre-processing of the base-plate and proper fixing of the base-plate to the build platform can also be specified. The following outline the procedure that serves as the DLS process benchmark. Pre-processing Preparation of the base plate, system parameter settings according to the material used, including the identified offset and scaling values. Fabrication A checklist for all possible considerations for a smooth process; specifically the volume of powder, preheating, and system settings. Post-processing Removal, cleaning methods based on the prototype material.

Fig. 15 Relation of La, Lm and b

incorporating appropriate scaling and beam offsets to the STL-format GBP CAD file. In practice to compensate for material shrinkage s, the CAD model is scaled by a scaling factor k, given by 1 : k 1s (3)

Thus if the material shrinkage s = 0, k = 1; otherwise k 1. To compensate for beam offset, the beam must be offset by b/2 in the X and Y directions with respect to the profiles of the CAD model as illustrated in Fig 15. From the analysis of the data it is observed that the error k is axis-dependant. Fig. 16 shows a plot of the deviation in accuracy between the actual and the measured dimensions of length in the Y-axis (as can be seen from Table 9). According to Eqs. 1, 2, 3, s,k, b, and b/2 along the Y-axis were computed to be -0.0443, 1.0463, 0.9227, and 0.4613mm, respectively. Similarly s, k, b, and b/2 along the X-axis were computed to be 0.0539, 1.0569, 0.6176, and 0.3088, respectively. Although measurement results indicate a difference in geometric accuracy for different geometric features, it is not

Table 9 Accuracy details of a fabricated GBP before the implementation of the proposed approach STL file dimensions (mm) Actual measured Dimensions (mm) X-axis Y-axis Error X-axis Y-axis

Fig. 16 Graphical plot of the deviation in accuracy

170 161.4250 163.3400 0.0504 0.0392 60 57.0760 58.4400 0.0487 0.0296 25 24.2050 24.3060 0.0318 0.0277 15 15.2100 15.4070 0.0140 0.0271 Surface Roughness, Ra: 23.3237 m

386 Fig. 17 Failed GBP before proposed approach

Fig. 18 GBP after applying the proposed approach

A test part (Fig. 19) other than the GBP was also built based on the benchmarked procedure to additionally verify the effectiveness of the proposed approach. The deviation in geometric accuracy of the part was within 0.1 0.2 mm. The part was built with clear and sharp features with negligible warpage.

3 Conclusions
A case study is presented based on the methodology of using six-sigma quality tools for benchmarking of a DLS process for prototyping using plastic nylon powder, with geometric accuracy and surface roughness as output quality characteristics. Once a tuned process is identified, it can serve as a benchmarked process for that material. Note that

a change of material used will require identification of a new process benchmark as the properties of different materials influence the fabrication parameters in the process. Generally laser-sintered models require post-processing, which also affects the geometric accuracy and surface roughness. Therefore the post-processing could also be suitably captured into the design and process planning especially when it is commonly used with the process and material used for the process. Although the benchmarking process is time consuming, the results are important for obtaining consistant quality parts achievable with the specific RP&M process. Future work includes the investigation of individual process benchmarks for the other RP&M processes. After the identification of common or popularly used benchmarks and results based on those benchmarks, a web-accessible decision support system [31] can help to select suitable RP&M processes for specific requirements based on the database of RP&M benchmarking.

Table 10 Accuracy details of a fabricated GBP after the implementation of the proposed approach STL file dimensions (mm) Actual measured dimensions (mm) X-axis Y-axis Error X-axis Y-axis

170 170.1020 170.1620 0.0006 0.0009 60 60.1840 60.2250 0.0030 0.0037 25 25.2620 25.2820 0.0104 0.0112 15 15.1020 15.1465 0.0068 0.0097 Note: Desired accuracy of the part could be further improved from post-processing Surface roughness, Ra: 10.1264 m

Fig. 19 Test part built

387

Appendix
Direct laser sintering (DLS) The high-temperature direct laser sintering process builds a part layer-by-layer on a specially coated base plate. The material generally used is a Cu-based metal powder apart from sand and plastics. A high power laser, guided by the scanner according to the layer profile, sinters the powder spread by a scraper at a pre-determined layer thickness. In this case study we demonstrate the proposed approach to identify the best achievable DLS process for prototyping using a new nylon plastic powder.

References
1. Kruth JP (1991) Material increase manufacturing by rapid prototyping techniques. CIRP Annals, 41st General Assembly of CIRP, USA, 40(2):603614 2. Gargiulo E.P (1992) Stereolithography process accuracy: user experience. Proc 1st European Conference on Rapid Prototyping, Univ. of Nottingham, pp 187207 3. Lart G (1992) Comparison of rapid prototyping systems. Proc 1st European Conference on Rapid Prototyping, Univ. of Nottingham, pp 243254 4. Juster NP, Childs THC (1994) Linear and geometric accuracies from layer manufacturing. Ann CIRP 43(1):163166 5. Juster NP, Childs THC (1994) A comparison of rapid prototyping processes. Proc 3rd European Conference on the Rapid Prototyping and Manufacturing, Univ. of Nottingham, pp 3552 6. Ippolito R, Iuliano L, De Filippi A (1994) A new user part for performances evaluation of rapid prototyping systems. Proc 3rd European Conference on Rapid Prototyping and Manufacturing, U.K, pp 327339 7. Ippolito R, Iuliano L, Gatto A (1995) A benchmarking of rapid prototyping techniques in terms of dimensional accuracy and surface finish. CIRP Ann 44(1):157160 8. Reeves PE, Cobb RC (1996) Surface deviation modelling of LMT processes-A comparative analysis. Proc 5th European conference on Rapid Prototyping and Manufacturing, Helsinki, 46 June 1996 9. Shellabear M (1999) EOS Gmbh. Benchmark study of accuracy and surface quality in RP Models. RAPTEC, Task 4.2 Report 10. Jayaram D, Bagchi A, Jara-Almonte, Cynthia C, O'Reilly S (1994) Benchmarking of rapid prototyping systems - Beginning to set standards. Solid Freeform Fabrication Proc, pp 146153 11. Wong YS, Fuh JYH, Loh HT, Mahesh M (2002) Rapid prototyping and manufacturing (RP&M) benchmarking. In: Gibson I (eds). Software Solutions for RP, PEP Ltd, UK, pp 5794

12. Mahesh M, Wong YS, Fuh JYH, Loh HT (2004) Benchmarking for comparative evaluation of RP systems and processes. Rapid Prototyp J 10(2):123135 Emerald Group Publishing Limited 13. Shellabear M, EOS GmbH (1998) Model manufacturing processes - State of the art in rapid prototyping. RAPTEC, Task 4.2, Report 1 14. Kulkarni P, Marsan A, Dutta D (2000) A review of process planning techniques in layered manufacturing. Rapid Prototyp J 6(1):1835 15. Rosen DW, West AP, Sambu SP (2001) A process planning method for improving build performance in stereolithography. Comput Aided Des 33:6579 16. Muller H, Schimmel A (1999) The decision dilemma - Assessment and selection of rapid prototyping process chains. Proc 8th European Conference on Rapid Prototyping and Manufacturing, Nottingham, pp 177192 17. Zhou J, Herscovici D, Chen C (1999) Parametric process optimization to improve the accuracy of rapid prototyped stereolithography parts. Int J Mach Tools Manuf 40:117 18. Camp RC (1989) Benchmarking: The search for industry best practices that lead to superior performance. ASQC, Milwaukee, WI 19. Arthur RT, Irving JD (1996) Process redesign. The implementation guide for managers. Engineering process improvement series. Addison-Wesley, Boston 20. Stamatis DH (2002) Six sigma and beyond problem solving and basic mathematics. St. Lucie, Boca Raton, FL 21. Stamatis DH (2003) Six sigma and beyond: statistical process control. St. Lucie, Boca Raton, FL 22. Stamatis DH (2003) Six sigma and beyond: the implementation process. St. Lucie, Boca Raton, FL 23. Jurrens KK (1999) Standards for rapid prototyping industry. Rapid Prototyp J 5(4):169178 24. Fadel GM, Kirschman C (1996) Accuracy issues in CAD to RP translations. Rapid Prototyp J 2(2):417 25. Badiru AB, Ayeni BJ (1993) Practitioners's guide to quality and process improvement. Kluwer, Dordrecht 26. Parsons R (1978) Statistical analysis. A decision making approach. Harper & Row, New York 27. Madhav S.P (1989) Quality engineering using robust design. Prentice-Hall, New York 28. Zairi M, Leonard P (1994) Practical benchmarking: the complete guide. Kluwer, Dordrecht, Ch. 9, pp 7580 29. Xiangwei W (1999) Calibration of shrinkage and beam offset in SLS process. Rapid Prototyp J MCB University Press 5 (3):129133 30. Yang TH, Jackman J (1997) A probabilistic view of problems in form error estimation. ASME J Manuf Sci Eng 119:375382 31. Fuh JYH, Loh HT, Wong YS, Shi DP, Mahesh M, Chong TS (2002) A Web-based database system for RP machines, processes and materials selection..In: Gibson I (eds). Software Solutions for RP, PEP Ltd, UK, pp 2755