An approach to minimize build errors in direct metal laser sintering.pdf

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 3, NO. 1, JANUARY 200673 An Approach to Minimize Build Errors in Direct Metal Laser Sintering Y. Ning, Y. S. Wong, J. Y. H. Fuh, and H. T. Loh AbstractThis paper discusses the effect of geometric shape on the accuracy of direct metal laser sintering (DMLS) prototypes. The percentage shrinkages due to different shapes are investigated and their empirical relationship is determined. A new speed-com- pensation (SC) method is proposed to reduce uneven shrinkage af- fected by the two-dimensional geometric shape at each layer. From case studies conducted, the optimized SC method is found to be ef- fi cient in improving the accuracy of prototypes fabricated. Note to PractitionersThis paper aims to address the problem of dimensional errors of parts built by the direct metal laser sintering (DMLS) process. Existing compensation approaches are normally based on a general relationship between the nominal dimensions and the errors after sintering. However, the effect arising from dif- ferent geometric shapes is not considered. A new approach is pro- posed using different scan speed settings to compensate for the ef- fect of geometric shapes to improve the dimensional accuracy of the entire part. During processing, the laser sinters along the tra- jectory are guided by the hatch vectors or dexel. An appropriate experimental method is used to establish the relationship for dif- ferentscanspeedswiththedexellengthtothefi nalaccuracy.When building the part, the laser scan speed is adjusted dynamically ac- cording to the dexel length which varies with the geometric shape of the part. The case study demonstrates that the proposed method can generate correct speed settings to effectively increase the di- mensional accuracy of the fi nal part. Although this method has been developed based on the DMLS process, it is also applicable to other laser sintering processes. In future research, other process parameters, such as laser power, will be considered independently, or together with the scan speed, for possible further improvement on the dimensional accuracy. Index TermsCompensation method (SC), dexel, direct metal laser sintering (DMLS), response surface method (RSM), selective laser sintering. I. INTRODUCTION S INCE the late 1980s, rapid prototyping (RP) technologies have been successfully applied in product development and manufacture. As one of the well-adopted RP technologies, se- lective laser sintering (SLS) has made very good progress in the past decade. Many materials including plastics, metals, ce- ramics, and sand can be used to produce RP objects directly by SLS. The ability of an RP process to produce an accurately Manuscript received December 26, 2004; revised January 3, 2005. This paper was recommended for publication by Associate Editor H. Ding and Editor M. Wang upon evaluation of the reviewers comments. Y. Ning was with the Department of Mechanical Engineering, National Uni- versity of Singapore, Singapore 117576. He is now with the GE Research and Development Center, Shanghai 201203, China. Y. S. Wong, J. Y. H. Fuh, and H. T. Loh are with the Department of Mechan- ical Engineering, National University of Singapore, Singapore 117576 (e-mail: .sg). Digital Object Identifi er 10.1109/TASE.2005.857656 shaped geometry is critical to its overall acceptance in the mar- ketplace 1. To achieve an accurately built part is a time-con- suming and complicated task because many factors can affect the fi nal dimensional accuracy. In general, researchers have fo- cused their attention on one or several of the following factors. Preprocessing error In the rapid prototyping (RP) process, three-dimensional (3-D) models are built by generating and stacking two dimensional (2-D) cross sections of uniform thickness. The fabricated part has a staircase or step-surface texture. Quantifi cation error can also occur when the height is not a multiple of the fi nite layer thickness. Hence, adaptive slicing algorithms 2, 3 have been developed to reduce these types of slicing errors. To process the 2-D layer data, a 3-D model is fi rst con- verted to a faceted model (in STL format). However, this conversion incurs an error during the tessellation of the faceted model. Some proposals using other data formats, such as constructive solid geometry (CSG) and NURBS- basedrepresentations,insteadoftheSTLtranslation,have been proposed 46. Machine errors Machineerrorscanbemeasured,appropriatelycalibrated, and compensated. The effect of the overall system errors can be controlled to a reasonable range. Material processing errors The dimensional errors arising from the material pro- cessing are the most complicated and have attracted much attention in RP research. In the SLS process, the temperature of the part or the powder is raised above its softening (e.g., for plastic powder) or melting (e.g., for metal powder with congruent melting point) or solidus (e.g., for prealloyed metal powders) temperature to bond and solidify the particles during the laser sintering process. As the sintered part cools, it also tends to shrink. To compensate the effect of material shrinkage, the 2-D layer needs to be scaled fi rst. In addition, an offset of the 2-D layer is processed to compensate the effect of the laser beam spot size. An experimental method to build and measure a part model to confi rm the values of scaling and offset compensation factor was described in 7. This method was based on a simple linear relationship be- tween the nominal dimensions and the errors caused after sintering. The effect arising from different geometric shapes is, however, not considered. Dimensional errors vary with different geometric shapes causing different percentage shrinkage in the entire part. To better under- stand the shrinkage mechanism, several researchers have 1545-5955/$20.00 2006 IEEE 74IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 3, NO. 1, JANUARY 2006 attempted to build thermal models to denote the heat transfer during the SLS process. These thermal models have been used to analyze the sintering of amorphous powder 1, 8, 9 and crystalline powder 10, 11. Zhang et al. 12 described a thermal model for the sintering of two-component metal powder. Because these models are complicated and deduced based on some assumptions, it is diffi cult to apply shrinkage compen- sation for a given geometric shape directly. Papadatos et al. 13 studied the effect of heterogeneous sintering on the variation of the dimensions in the Z-direction (build direction). However, the material heterogeneity caused by variation in the geometric shape in 2-D layers is not considered. A few researchers 14, 15, etc. have built the heat-transfer model using the fi nite-element analysis method. The application is limited since the model has to be developed for each part simulation with different geometric shapes. It is important to effectively analyze and compensate the effect of different geometric shapes to improve the dimensional accuracy of the entire part. Andre 16 ob- tained experimental data for measuring shrinkage values of many different geometric shapes and then applied dif- ferent shrinkage compensation factors to the computer- aided design (CAD) model for each section of a part. It is a tedious task especially for complex geometries that need a large amount of experimental data. The results are alsodiffi culttogeneralizetootherprocessconditions.The diffi cultyofusingarelativelysimplifi edmethodtodenote the shape character based on the SLS process is another problem. Thus, the geometric reasoning becomes a very diffi cult task in the case of complex geometries. Random noise Besides the error factors mentioned above, the fi nal part dimensions are not uniform in practice even when two processing environments are similar. This error is defi ned as the random noise and the relevant analysis has been provided by Jacobs 16. Thispaper proposesa method tostudytheeffectof2-D geometric shape factor on the dimensional accuracy and analyzes the effect of different geometric shapes on the dimensional accuracy of the part. A speed-compensation (SC) method has also been developed, which includes an experimental data collection method and statistical anal- ysis. Acase study is used to demonstrate the effectiveness of the optimized SC method to improve the accuracy of the fi nal part. II. MATERIALSHRINKAGE ON THELASERSINTERINGPROCESS A. Laser Irradiation In the SLS process, a moving laser source with a fi xed scan valueselectively sinters each layer of powder material line by line. According to Jacobs 17, the surface irradiation energy densityat pointwith distancefrom the laser sintering line (as illustrated in Fig. 1) can be represented as (1) Fig. 1.Laser beam sintering of continuous hatch lines. whereis the laser power,is the scan speed, andis the radius of scan beam. The energy densityabsorbed by the sintered powder is (2) whereis the absorptivity of the material related to the time and temperature. The energy densityin (1) decreases rapidly with distance . A “zone of infl uence” has been defi ned as a differential area that receives 99.99% of laser irradiation, and the radiusof it is given as 17 (3) The time for the laser to scan within the “zone of infl uence” is quite short and the powder temperature increases sharply during this period. Thereafter, the absorbed energy will be lost from the sintered surface to the outside environment by radiation and convection. Simultaneously, the heat will also be conducted through the powder bed in different directions to affect the temperature distribution of the powder bed. As a result, the temperature of the sintered powder decreases rapidly. Normally, the actual sintering time is very short so that the full-density parts cannot be achieved in such a short period. B. Material Shrinkage in the Direct Metal Laser Sintering (DMLS) Process Based on the SLS technology, DMLS has wider applications because the metal powder can be sintered directly to build func- tionalprototypes.Itcanfabricate3-Dnear-net-shapemetalparts directly in a single process, which is achieved by using a rel- atively low-power laser to sinter steel- or bronze-based metal powder layer by layer. Inthiswork,atwo-componentmetalpowdermaterialsystem, consisting of a mixture of 60 wt% pure copper powder and 40 wt% prealloyed SCuP powder, is used. The prealloyed SCuP is a Cu-based alloy with melting point at 646 C. By absorbing the laser energy, the low-melting point SCuP, which serves as a binder, is melted while the high-melting point copper, which serves as the structure powder, remains in solid state. During sintering, the structure powder is wetted and bonded together by the liquid binder through capillary action and gravity. Suffi ciency of the liquid fl ow is critical to the sintered part density and further affected the sintering quality. NING et al.: APPROACH TO MINIMIZE BUILD ERRORS IN DIRECT METAL LASER SINTERING75 Fig. 2.Translation from a layer to voxel combination. The infi ltration of the liquid binder through the pores causes signifi cant volume shrinkage. Besides these, some other factors may also affect the fi nal volume shrinkage result, such as the elastic compressive shortening during the cooling stage 1. Being primarily thermal in nature, the sintering process strongly depends on temperature variation with time. For a selected material, the physical properties of prototype parts resulting from the DMLS processing are strongly infl uenced by the temperature history during the laser-material interaction period 18. If the time of the liquid phase is prolonged, the fl ow of the liquid-phase metal will improve, fi lling up the pores and thereby increasing the densifi cation. Besides the time of the liquid phase, the highest temperature achieved at the powder surface is another important factor because it brings about a higher temperature gradient to give more surface activation energy for improved liquid-phase fl ow. III. EFFECT OFGEOMETRICSHAPE ONPARTACCURACY A. Dexel Model The2-Dlayerisgeneratedfrommaterialscreatedbythelaser scanningintheformofparallelhatchvectors.Eachhatchvector can be considered as a dexel 19, 20. During processing, the laser sinters along the trajectory guided by the hatch vectors. A voxel with the specifi c height and width can be built around each hatch vector. Each 2-D layer of a specifi c thickness can be represented as the accumulation of a list of voxels inside, as illustrated in Fig. 2. The change in the dimensional accuracy of the 2-D layer is regarded as a composite effect of the voxels. To simplify the model, the hatch vector (i.e., dexel) is used to denote the corresponding voxel. Based on the model, diverse shapes can be regarded as different combinations of dexels with different lengths. Analyzing the accuracy due to the effect of geometric shapes can be considered to be similar to analyzing the effect by the dexels and their interaction. This method is more direct and easier. From previous research 17, the material shrinkage is de- termined by the energy (temperature) change of the sintered powder and the material properties. It is feasible to control the shrinkage by adjusting the energy density of the laser with the Cu-based material system. If the temperature variation at each dexel is similarly controlled, the composite 2-D layer could show more consistent behavior in shrinkage. B. Neighboring Effect Brought by the Change of Dexel Length By regarding the 2-D geometric shape as a combination of a series of parallel dexels, the analysis of the effect of different shapes is equivalent to the analysis of the percentage shrinkage and interaction of the dexels of different lengths. During the Fig.3. Temperatureversus timecurveofpointPin2-Dinfi nitelinessintering. Fig. 4.Negative neighboring effect on the temperature versus time curve. laser sintering process, each sintered point on the surface of the powder bed receives multiple energy pulses of varying inten- sity from the neighboring dexels 1. Consider a model in an ideal situation where the length of each dexel is long enough for the temperature to decrease near to the surrounding temperature before receiving the next energy pulse. The temperature versus time curve of a pointin dexelis illustrated in Fig. 3.is the melting temperature of the binder,is the process ambient temperature, andis the time when the laser beam focuses on the sintered point. The neighboring second and third energy pulses are due to the effect of sintering the neighboring dexels (A, B, D, E). This effect is defi ned as the neighboring effect. In this situation and neglecting the dexels at the edges, the shrinkage of each dexel is similar because the temperature vari- ationwithtimeissimilarineachdexel.Theintegrated2-Dlayer shows a similar percentage shrinkage along the sintering direc- tion in different geometric regions of the layer. But in practice, the dexel lengths will be much different for different geometric shapes. In the regions with short dexels, the interval between successive irradiations is relatively short. When the interval is not suffi ciently long for the surface to cool down, the temperature in the region will gradually build up, re- sulting in higher temperature and longer liquid-phase time. The variationinthetemperaturehistorywhensinteringdexelsofdif- ferent length causes differential shrinkage in the 2-D layer and thereby reduces the sintering accuracy. The negative effect of the short-dexel sintering on accuracy is referred herein as the negative neighboring effect. In this situation, the curve of tem- perature versus time will change to the situation as shown in Fig. 4. Tontowi and Childs 21 studied the effect of different preheating temperatures of the powder bed on the part den- sity and energy density. The effect brought by the neighboring dexels can be regarded as preheating and postheating that could change the pattern of powder bed temperature distribution dy- namically. When the temperature and time features in different 76IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 3, NO. 1, JANUARY 2006 Fig. 5.Experimental setup for continuous temperature measurements. regions are not similar, the sintering quality would most likely be different.Williamsand Dickard18 havetestedtheeffecton thedensityandstrengthofdifferentdelaybetweensuccessiveir- radiation exposures during the laser-material interaction period due to a change in geometry for bisphenol-Apolycarbonate ma- terial. The results show an obvious change in the density with a delay period. Similarly, the percentage shrinkage around the re- gion of shorter dimension is expected to be different and affect shrinkageuniformityofthelayer.Moreseriously,distortionand warpage of the sintered layers may occur with such differential shrinkageinthepart.Besidestheneighboringeffectcausingun- even shrinkage, the heat-affected zones arising from the fi nite diameter of the laser beam are also different when sintering re- gions of short dexelsdue to the variation in the temperature pro- fi le. In this situation, there is limited accuracy improvementthat can be achieved just by using a fi xed offset value similar to that used in 7 to compensate the error caused by the fi nite diam- eter of the laser beam. To solve these problems, the temperature profi le of the sintered powder should be maintained as uniform as possible. C. Experimental Validation 1) Apparatus Setup: To verify the aforementioned analysis of the temperature variation on regions with different dexel lengths, several experiments have been conducted using a RAYTEK MXCF modal noncontact infrared thermometer (Fig. 5). The distance between the measuring object and the sensoroftheinfraredthermometerwassettobearound500mm and the measured spot size was around 6 mm. The temperature in the