design, manufacture and geometric verification of rapid prototyped microfluidic encapsulations by...
TRANSCRIPT
Computers in Industry xxx (2013) xxx–xxx
G Model
COMIND-2485; No. of Pages 14
Design, manufacture and geometric verification of rapid prototypedmicrofluidic encapsulations by computed tomography
Jorge Santolaria a,*, Rosa Monge b, Angel Tobajas a, Roberto Jimenez c, Mirko A. Cabrera b,Luis J. Fernandez b
a Department of Design and Manufacturing Engineering, Universidad de Zaragoza, Marıa de Luna, 3, 50018 Zaragoza, Spainb Group of Structural Mechanics and Materials Modelling (GEMM), Aragon Institute of Engineering Research (I3A), Universidad de Zaragoza, Mariano
Esquillor s/n, 50018 Zaragoza, Spainc Centro Universitario de la Defensa, A.G.M. Ctra. Huesca s/n, 50090 Zaragoza, Spain
A R T I C L E I N F O
Article history:
Received 1 September 2012
Received in revised form 23 June 2013
Accepted 24 June 2013
Available online xxx
Keywords:
Computed tomography
Rapid prototyping
Microfluidic encapsulations
Measurement
Threshold
Scale factor
A B S T R A C T
This paper presents the dimensional verification of encapsulations used to package microfluidic devices
manufactured using a 3D printer of photopolymerisable resin. This characterisation has been performed
by computed tomography (CT) by comparing newly manufactured encapsulations and samples that
have been subjected to test conditions. Thus, it has been possible to draw conclusions both on the
deviations of the nominal geometry of the encapsulations and on how this might affect their
performance. This paper presents a scheme of dimensional verification from the point clouds obtained
by CT. Finally, a combined threshold and scale factor correction technique of the tomography images is
shown. This method is based on the simultaneous measurement of objective and master parts with
known geometry. The results reveal the improvements achievable in the accuracy, given a particular
machine configuration. The conclusions facilitate the improvement of the geometric design of these
devices regarding their behaviour under test conditions.
� 2013 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Computers in Industry
jo ur n al ho m epag e: ww w.els evier . c om / lo cat e/co mp in d
1. Introduction
Additive manufacturing technologies, which are also known aslayered manufacturing technologies or generically, as rapidprototyping (RP) technologies, are currently fully implementedin very different applications. Since its first industrial presence in1988, more than 40 layered manufacturing technologies have beendeveloped for the market [1]. These technologies effectively coverapplications related to the production of prototypes in thedevelopment cycle of new products, obtaining various types ofprototypes in each phase of the cycle according to the require-ments of the designer or client. The latest developments onmaterials for prototyping, combined with appropriate post-processing techniques allow parts to be obtained in certainapplications for end use. Generally, these are products for shortseries production, custom products or products of complexgeometry, which cannot be manufactured otherwise than bylayered manufacturing. RP technologies group any technology thatallows a physical model to be obtained automatically from CADdata by additive layer manufacturing.
* Corresponding author. Tel.: +34976761887; fax: +34976762235.
E-mail address: [email protected] (J. Santolaria).
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
0166-3615/$ – see front matter � 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.compind.2013.06.015
From an industrial point of view, obtaining a final piece by usingadditive manufacturing technologies is known as RapidManufacturing. Although these techniques were intended origi-nally for purely industrial applications, at present, owing to theflexibility in obtaining complex shapes and free surfaces, one of themost widespread applications of additive fabrication is related tomedical applications. In combination with imaging techniques,such as computed tomography (CT) or magnetic resonanceimaging (MRI) for geometry capture and obtaining the stereolithography (STL) format files, RP can be used in the manufacture ofbiomodels for surgical planning [2–4], or in surgical guides [5] for aspecific patient and a specific clinical case. The use of RP alsoextends to applications in the field of cell biology and tissueengineering, as in the manufacture of scaffolds for tissue growth[6]. Within the field of clinical tests using microfluidic systems, it iscommon to use microchips for cell deposition obtained byphotolithography integrated in packages that provide the fluidicconnections necessary for the transport of substances into cells.The development of new tools for cell culture, based onmicrotechnologies, allows not only the control of the mechanical,chemical and electrical environment of biological samples but alsothe monitoring of their reactions in a way previously unachievable.In this way, it is possible to generate new methods for therealisation of ‘‘in vitro’’ tests in similar conditions as ‘‘in vivo’’. This
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx2
G Model
COMIND-2485; No. of Pages 14
advance is expected to promote a reduction in animal experimen-tation and the development of new drugs through high-through-put tests.
State-of-the-art microfluidic devices for cell culture are basedon soft-lithography with the use of materials such as PDMS for themicrochip manufacturing. Such devices have permitted therealisation of the first cell culture experiments on chips, provingthe excellent future prospects of this technology. However, inorder to achieve the full potential in biotechnological applications,a step forward is required towards reliability, handling and designflexibility of the microfluidic chips. Technology based on thepolymeric material SU-8 for the chip has been developedpreviously and tested for clinical diagnosis applications [7,8],making possible not only the construction of microchannels butalso the possibility of sensor integration and the creation oftridimensional channel platforms, among other interesting char-acteristics. In [9], an RP manufacturing system for tridimensionalmicrochannels is presented, together with its later integration in achip compatible with infrared spectroscopic imaging systemsanalysis. In the case of chip packaging, the geometric complexity ofthe connections and microfluidic channels, together with the needfor integration of sensors or control and detection systems, makesthis type of encapsulation difficult and costly to manufacture byconventional manufacturing technologies. It requires the use ofdifferent parts and later assembly with sealed elements to ensureadequate fluidic isolation. Often, owing to the space or geometryrequirements of the channels, these parts cannot be obtained byconventional manufacturing. RP systems with smaller layerthicknesses are able to manufacture these geometries successfully;thus, it is a manufacturing technology suitable for this type ofapplication because of both the accuracy achievable and of thebiocompatibility of the materials. However, subsequent use ofthese microfluidic devices requires both sterilisation and pro-longed biological tests under the conditions of cell culturetemperatures and under the action of loads caused by the sealsand the sealing elements of the package. Usually, technologiescapable of obtaining the required geometry work with materialsthat have high coefficient of thermal expansion, which varies withtemperature and that are orthotropic, working in plastic regime atlow temperature. Thus, there are some problems related to thefunctionality of these parts and their deformation under tempera-ture test conditions, which can result in permanent deformationthat can influence the test development or the final product.Through understanding the behaviour of the materials at thesetemperatures and during the end use of the parts in the test, it ispossible to optimise the geometry of these devices to befunctionally effective and to overcome their mechanical limita-tions.
Verification of small parts with complex internal geometry isnot possible, or is partially possible, by the use of traditionalcontact or non-contact measurement instruments. Therefore, theuse of coordinate measuring machines (CMMs), articulated armCMMs, and laser triangulation sensor or optical CMMs, is limited toexternal geometry in this type of application in non-destructiveinspection. CT metrology using X-rays [10] is suitable for capturingthe geometry and subsequent verification of these parts owing tothe ability of obtaining both external and internal geometry via anon-contact measurement. The measurement result is a cloud ofpoints that allows direct or indirect geometric analysis, afterobtaining an STL file, by post-processing and a best-fit approach tothe geometric primitives considered. Regarding the result, CTmetrology is therefore a measurement technique similar to thedigitalisation technologies used in quality control against CAD orreverse engineering with the addition that it also allows internalgeometry to be obtained. Owing to the spread of CT in industrialdimensional metrology applications, numerous studies have been
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
performed to model the error sources and to develop effectivecalibration techniques for this type of equipment [11]. Many errorsources affect measurement uncertainty, which are related mainlyto the CT X-ray emitter, the rotation of the workpiece duringcapture, the sensor that captures the attenuation in the power ofthe rays after passing through the piece, or the actual mathemati-cal algorithms for reconstruction. From the standpoint of the endresult, for a machine and specific measuring conditions, edgedetection based on thresholding is a major influence on theaccuracy of the surface reconstruction of the measured part.Furthermore, the voxel size adjustment in the reconstruction has aglobal influence as a scale factor in the STL obtained. Thus, edgedetection and voxel size calibration and rescaling are importantaspects determining the final accuracy obtained in the 2D imagesand the 3D reconstruction phase, such that various methods havebeen developed to obtain the optimum values in both cases for anaccurate reconstruction of the measured geometry. For thresh-olding, the techniques employed to obtain values that allow anaccurate separation between the material and air, or betweendifferent materials, are based on various principles, the mostcommon being the average value of grey level between the air andmaterial, local adaptive thresholding of grey level, maximum greyvalue derivatives and interpolation between voxel grey values. Agood review of the characteristics and limitations of thesetechniques can be found in [12] and a comparison of differentsegmentation algorithms used in CT measurements is presented in[13]. Regarding voxel size adjustment by scaling factor, the usualmethod to determine the factor or factors, depending on thetechnique used, is to measure by CT a calibrated object togetherwith the test part, or to measure in a CMM some geometricalcharacteristics of the test part following measurement by CT[10,14].
In this paper, the geometric verification and dimensionalcharacterisation of encapsulations for microfluidic applicationsmanufactured in photopolymerisable resin by RP in a 3D printer ispresented. This characterisation is done through a new measure-ment procedure based on CT and on the measurement of masterpieces of the same material of known geometry. This techniqueallows dimensional corrections based on scale factors andthreshold selection to be used during the reconstructing processof the 3D geometry obtained by CT analysis. It also allows theevaluation of the accuracy obtained and improves the accuracy ofthe STL model, ultimately generated following the corrections. Byusing this technique, it is possible to verify the dimensionalaccuracy of the manufactured microfluidic encapsulations and toanalyse the deformation suffered by the encapsulation due to testconditions, in order to optimise its geometric design.
2. Experimental methodology and setup
In this paper, four microfluidic chip packaging tools manu-factured in a 3D printer (Objet Eden 350V) and two master piecesobtained by the same technique in the same material (photo-polymerisable resin FullCure 720) are considered for the dimen-sional verification technique presented. The first step in theworkflow is the 3D CAD modelling of the package and once pre-processed, it is manufactured by model and support materialdeposition layer by layer. The main head moves in a plane (X-, Y-axes) with movements along the X-axis and movements in the Y-direction between passes of the print head for the same layer, if it isnecessary depending on the width of the workpiece. The materialis released by successive head passes with a mean nominal layerthickness of 16 mm on a tray, which descends between passes (Z-axis), resulting in a ZFXY kinematic structure [15]. The smallestnominal dimensions to verify in the manufactured parts are themicrofluidic channels that connect the lateral fluidic entry points
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Fig. 1. 3D CAD modelling of the microfluidic chip packaging. Channels detail and RP parts manufactured.
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 3
G Model
COMIND-2485; No. of Pages 14
to the chip, which have a nominal diameter of 500 mm (Fig. 1). Themanufacturing technique chosen is one of the techniques on themarket with lower layer thickness, which allows both a goodnominal dimensional accuracy (�100 mm) in the entire volume ofthe work piece and a good surface finish with details of the testedparts inside the tolerance range required by the application.
2.1. Microfluidic setup and test conditions. Chip packaging
A complete microfluidic setup for the realisation of cell cultureexperiments is first presented. The setup comprised an SU-8 chip,chip packaging, valves, a reservoir and a micropump. Theinterconnection between the cell culture microchamber and theoutside world is as follows (Fig. 2). On the chip, the microchamberdedicated to culture the cells is connected to a dedicated outlet(500-mm diameter) by a microchannel with a typical size of100 mm. This outlet is hermitically connected to the packaging viathe use of silicone O-rings placed on top. Then, the liquid flowsinside the package through dedicated channels (500-mm diame-ter). Finally, tube connectors, fully integrated in the package, allowconnection to standard tubes (0.8-mm inside diameter). To keepthe packaging closed, a stainless steel bolt is used. A specificelectronic control is also required for the pump in order to obtainthe desired flow. After the cell seeding, the chip is located in thepackaging created for this application. Once all the setupcomponents are linked by the appropriate tubes, the flow (ofthe order of 1 ml/min) is generated by electronic control of themicropump. Every mounting step of the final setup must beperformed inside a laminar flow cabinet to ensure sterileconditions and then moved to an incubator to maintain it atstandard cell culture conditions: 37 8C, 5%CO2 and 100% humidity.The setup must be placed in the incubator for at least 24 h. Underthese conditions, the principal components that can producedeformations in the packaging are the bolt, the O-rings and thechip.
Four packagings were manufactured for verification by CT. Eachwas subjected to different conditions in order to analyse theinfluences and geometric deviations produced at each stage of the
Fig. 2. Open and closed RP microfluidic chip
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
process: (1) manufacture in 3D printer, cleaning and assembling ofbolt, chip and O-rings; (2) manufacture and cleaning; (3) same as 2and subjected to test conditions in incubator for 24 h; and (4) sameas 1 and subjected to test conditions in incubator for 24 h.
2.2. Design, manufacturing and calibration of master pieces
We have designed a dedicated reference standard part for use asa master piece, both to evaluate the manufacturing accuracy andthe accuracy of the measurement with CT (Fig. 3), in order to applycorrection methods according to the master dimensions obtained.
Two numbered master parts were manufactured and calibratedby contact measurement in a CMM. The calibration results of tenmeasurements per geometric feature in the CMM are shown inTable 1, together with the expanded uncertainty for k = 2 in each ofthe calibrated values, taking into account the CMM uncertainty,the standard deviation of the measurements, the nominalcoefficient of thermal expansion and the range of variation ofthe laboratory temperature. The geometry considered for themaster piece, comprising calottes, parallel planes and holes,responds to the need to realise in the master piece similargeometries for verification as in the part, owing to the nature of thetwo correction techniques implemented. Table 1 shows thereference values obtained in calibration and used in the twogeometric correction techniques.
2.3. CT measurement process
Measurement of the parts was made in a micro-CT machine(General Electric). This is a general-purpose CT machinedeveloped for the measurement of tissue of small dimensionsbut with appropriate measurement capabilities for this applica-tion. It has a maximum resolution of 8 mm and an X-ray powerfrom 50 to 80 KV. Closed test chip packagings were measuredtogether with the two master parts oriented in mutuallyperpendicular directions, such that a single tomography allowsthe correction values in the three dimensions to be obtained byusing the calibration measurements of the master parts. The
packaging with seal elements and chip.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Fig. 3. CT master parts.
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx4
G Model
COMIND-2485; No. of Pages 14
arrangement of the measurement setup with fixed values ofcurrent, voltage, position and voxel size, is shown in Fig. 4 and asample of the CT image is shown in Fig. 6. Four measurementswere performed by CT for the packagings considered in differentconditions: CT1, manufacture in 3D printer, cleaning andassembling of bolt, chip and O-rings; CT2 manufacture andcleaning; CT3 same as 2 and subjected to test conditions inincubator for 24 h; CT4 same as 1 and subjected to testconditions in incubator for 24 h.
The workflow starts with data analysis in Materialise Magics.Two independent correction methods based on reference mea-surements of the master pieces are applied. The first correctiontechnique is based on determining the appropriate scale factorusing the dimensions shown in the first part of Table 1. This scalefactor may be applied locally with individual distances, or globally,according to the slope of the deviations between the calibrateddimensions of the master piece and those obtained in themeasurement. It is possible to assume that these deviations arethe combined result of a constant offset due to the threshold, plus alinear term due to the scale factor. Therefore, the first scalingcorrection is performed for subsequently applying an adaptivethreshold in order to obtain the surfaces determining theappropriate cut-off value from the deviations using the dimensions
Table 1CT Master parts calibration results in CMM.
Dimension Nominal CAD value (mm) Master 1
Measurement (mm)
Scale factor determination
Distance S_1–S_2 7000 6.9987
Distance S_1–S_3 19,310 19.3738
Distance S_1–S_4 18,000 18.0556
Distance S_1–S_5 6590 6.6526
Distance S_1–S_6 22,210 22.3117
Distance S_2–S_3 18,000 18.0573
Distance S_2–S_4 19,310 19.3558
Distance S_2–S_5 6590 6.6634
Distance S_2–S_6 22,210 22.3035
Distance S_3–S_4 7000 6.9958
Distance S_3–S_5 22,210 22.3592
Distance S_3–S_6 6590 6.5992
Distance S_4–S_5 22,210 22.3475
Distance S_4–S_6 6590 6.6043
Distance S_5–S_6 25,000 25.1819
Threshold calibration
Diameter S_1 5000 4.9573
Diameter S_2 5000 4.9581
Diameter S_3 5000 4.9573
Diameter S_4 5000 4.9606
Diameter S_5 4000 3.9710
Diameter S_6 4000 3.9790
Distance P_2–P_3 4000 3.8994
Distance P_3–P_4 7000 3.1704
Distance P_4–P_5 4000 3.8958
Distance P_5–P_6 11,000 7.0968
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
contained in the second part of Table 1. Overall, this arrangementgives a reconstruction that minimises both errors. It is complex toquantify the individual influence of the error due to scale factorand the error due to thresholding process with an inappropriatevalue. While the scale factor has a global effect on the finalaccuracy, the threshold has a different effect depending onwhether it is an outer or inner dimension of a feature. Therefore,a combined application of both sequential corrections is suitable asa global dimensional correction technique.
Thresholding is a technique used to segment an image into twoparts: the region of interest and the background. A tomographicimage is a set of pixels, each one having a value within a grey scalethat can be represented in a histogram. In an image, thebackground portion and the part are depicted in grey levels withhigh contrast between them, depending among other factors, onthe density of the materials. The final border that separates thezones, as it is a gradual transition of grey levels, will be defined bythe threshold value chosen; thus, separating the region of interestand the background portion. The range of grey levels considered inthe image determines which zone of the image corresponds to thepart of interest, such that the considered value influences directlythe dimensions of the reconstructed part and therefore, its finalaccuracy. The transition between background and part in the
Master 2
Uncertainty (mm) Measurement (mm) Uncertainty (mm)
1.3 7.0016 1.3
2.5 19.3316 2.5
2.3 18.0133 2.3
1.4 6.6632 1.3
3.3 22.2596 2.8
2.3 18.0182 2.3
2.5 19.3252 2.5
1.4 6.6735 1.3
3.3 22.2614 2.8
1.3 7.0012 1.3
2.8 22.3377 2.8
1.5 6.6003 1.3
2.8 22.3292 2.8
1.4 6.5942 1.3
3.4 25.1537 3.1
1.7 4.9530 1.3
1.4 4.9693 1.2
1.8 4.9691 1.2
1.7 4.9544 1.2
1.3 3.9750 1.2
1.5 3.9787 1.1
1.6 3.9023 1.3
1.4 3.1591 1.4
2.7 3.8811 1.4
1.6 7.0889 1.6
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Fig. 4. Measurement arrangements for CT.
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 5
G Model
COMIND-2485; No. of Pages 14
image occurs gradually in the grey scale; therefore, it is necessaryto adjust this parameter to obtain the best possible accuracy in asubsequent CAD reconstruction. Thus, it is possible to analyse theimage through its histogram in order to choose the threshold valueof the border between the part and the background. Depending onthe choice of this value, a larger or smaller size for the packaging isobtained.
2.4. Histogram analysis
The beginning of the work starts from the tomographic imagehistograms. In an image that shows different materials, thehistogram indicates each material by showing the number of pixelsthat represent them in a range of values in the grey scale. Thus, inan image where different peaks are observed, the zones betweenthem represent areas of different materials. Therefore, this zone ischosen between peaks representing the material to be tested (RPmaterial) and the threshold value of the peaks limit is observed.
Fig. 6. CT images of the untested
Fig. 5. Histogram of an image in the region of interest (tomography 1).
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
The minimum value corresponds to ISO value 0 and the maximumvalue corresponds to ISO value 100. To analyse how the packagingsize varies according to ISO values, three values are chosen. First,the average ISO value is selected, which gives the mean value of thepart size. This value corresponds to ISO-50. Second, values arechosen above and below the mean value: one value to create asmaller packaging size and the other to create a bigger one. Valuesabove and below the mean value are discussed, taking 40 and 60 asrepresentative from what would be the natural thresholdboundary. Therefore, three threshold values are selected: valuesof ISO-40, ISO-50 and ISO-60.
Fig. 5 shows a histogram belonging to CT1 and Fig. 6 presentssome sample images for two CT measurements. In this histogram,there are several peaks belonging to the different existentmaterials in the tomography. The peaks representing the RPmaterial are between the approximate values of �1024 and �205,as it is the most abundant material in the tomography. Each of thefour tomographies has a different histogram in the selected imageto set the part boundary and therefore, each has different values ofthreshold. The selected values in the tomographies are shown inTable 2. ISO values are established between the histogram peaksrepresenting the RP material. As said before, the maximum value ofthreshold is the value of ISO-100 (%) and the minimum is ISO-0.Therefore, the ISO value we want to obtain will be between themaximum and minimum threshold values. Then, a linearrelationship is established between these values to obtain thethreshold value corresponding to each predefined ISO value,searched by a simple interpolation (Eq. (1)).
100 � 0
ðVT ISO 100Þ � ðVT ISO 0Þ ¼100 � 40
ðVT ISO 100Þ � ðXÞ (1)
where VT ISO 100 is the value of threshold corresponding to ISO-100, VT ISO 0 corresponds to the threshold value corresponding toISO-0 and X is the threshold value corresponding to ISO-40 in thiscase.
microfluidic encapsulation.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Table 2Threshold values for each tomography.
CT ISO40 (HU) ISO50 (HU) ISO60 (HU)
1 �715 �637 �560
2 920 1150 1380
3 �110 75 260
4 990 1025 1060
Table 3Reconstruction results for spheres in CT1, master 1 and selected threshold values.
Diameter
(mm)
Standard
deviation (mm)
Coord. X Coord. Y Coord. Z
ISO 40
Sphere 1 4.892 22 7.353 17.593 44.879
Sphere 2 4.888 21 7.111 24.637 44.995
Sphere 3 4.874 15 7.634 24.995 26.92
Sphere 4 4.892 18 7.865 17.957 26.798
Sphere 5 3.917 23 11.538 21.194 48.659
Sphere 6 3.953 14 12.191 21.702 23.403
ISO 50
Sphere 1 4.929 15 7.363 17.593 44.882
Sphere 2 4.924 13 7.117 24.629 44.999
Sphere 3 4.915 13 7.634 24.991 26.917
Sphere 4 4.924 16 7.875 17.953 26.794
Sphere 5 3.948 16 11.544 21.194 48.643
Sphere 6 3.993 14 12.189 21.701 23.405
ISO 60
Sphere 1 4.956 16 7.367 17.598 44.888
Sphere 2 4.956 14 7.117 24.632 45.006
Sphere 3 4.943 14 7.637 24.992 26.918
Sphere 4 4.949 16 7.88 17.958 26.797
Sphere 5 3.978 16 11.542 21.196 48.638
Sphere 6 4.033 15 12.189 21.702 23.397
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx6
G Model
COMIND-2485; No. of Pages 14
3. Correction methods and results
3.1. Scale factor
The first correction method is based on the scale factor [16] andit is obtained from the reference measurement of the master parts.Thresholding unaffected measurements are wanted and therefore,for this correction method, distances between the centres ofspheres are chosen. Sphere centres are fixed points in a part andthe threshold value changes only the sphere diameter and not itscentre. In this way, distances between all the centres of spheres areconstant for all the possible threshold values of the image.
Once the 3D point clouds corresponding to the master parts foreach of the tomographies for the different values of thresholdconsidered are obtained, the data are analysed using GeomagicQualify software. The distances between the centres of the spheresare obtained to determine the scale factor of the 3D representationand finally, to compare them all to establish what relationship thisfactor has for the different threshold values considered for eachtomography (Fig. 7).
Target points are chosen in the proper zone of the cloud ofpoints and the adjustment spheres are obtained by the leastsquares method. Table 3 shows the mean diameters, standarddeviations and centre coordinates of the adjusted spheres in thedifferent threshold values for CT1 and master 1. Both reconstruc-tion and measurement work must be done individually for eachtomography and master part.
After obtaining these points, the distances between them arecalculated. Having six spheres in the master piece means that thereare 15 distances for each master and tomography (a total of 120distances are analysed). Sample results are shown in Table 4, whichcontains the actual distances, the distances considering thedifferent values of threshold, the error between the calibrateddistances and the measured ones and the equivalent scaling factorfor each distance in each threshold value.
Finally, Table 5 and Fig. 8 are a summary of all the tomographieswith each of the master pieces used. They present the mean scalefactor, the standard deviation and the variance of all the distancesobtained for each threshold value in each tomography and masterpiece.
Fig. 7. Clouds of points of the master parts obtained by CT. (a) Points selection for S4
tomography 1 and master 1.
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
It is noted that the relationship between the scale factor and thedifferent threshold values obtained in each tomography and eachmaster remains constant; therefore, it is independent of the valueof the threshold. Therefore, according to the standard deviationvalues obtained, it is possible to conclude that the calculated scalefactor with such measurements is stable and that each CT is scaledby this factor. Otherwise, in the case of not being able to extract arelationship between the scale factor and the different thresholdvalues, it would not be possible to determine which thresholdvalue would provide the obtained measurement corresponding tothe calibrated one.
3.2. Threshold value determination
The second correction method based on the threshold values, asexplained above, gives part dimensions. The scale factor wasobtained by comparing the distances between the spheres and thenominal distances obtained in the CMM. In this section, thethreshold value is determined from measurements affecting partsurfaces, i.e., using measurements that change with the consideredthreshold value. These measurements are the diameters of spheresand the distances between the planes of master objects. Once thesedimensions are measured and the scale factor is applied, corrected
in ISO-60 tomography 2 and master 2; (b) distance between S2 and S4 in ISO-40
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Table 4Scale factors in CT1 from distances between spheres. Master 1.
Nominal Value (mm) ISO 40 ISO 50 ISO 60
Dist. (mm) Error (mm) F.E. Dist. (mm) Error (mm) F.E. Dist. (mm) Error (mm) F.E.
Distance S_1–S_2 6.9987 7.0491 0.0504 0.9928 7.0413 0.0426 0.9940 7.0394 0.0407 0.9942
Distance S_1–S_3 19.3738 19.4266 0.0528 0.9973 19.4305 0.0567 0.9971 19.4336 0.0598 0.9969
Distance S_1–S_4 18.0556 18.0919 0.0363 0.9980 18.0988 0.0432 0.9976 18.1019 0.0463 0.9974
Distance S_1–S_5 6.6526 6.6910 0.0384 0.9943 6.6778 0.0252 0.9962 6.6662 0.0136 0.9980
Distance S_1–S_6 22.3117 22.3944 0.0827 0.9963 22.3926 0.0809 0.9964 22.4044 0.0927 0.9959
Distance S_2–S_3 18.0573 18.0861 0.0288 0.9984 18.0930 0.0357 0.9980 18.0991 0.0418 0.9977
Distance S_2–S_4 19.3558 19.3990 0.0432 0.9978 19.4053 0.0495 0.9975 19.4086 0.0528 0.9973
Distance S_2–S_5 6.6634 6.6991 0.0357 0.9947 6.6840 0.0206 0.9969 6.6767 0.0133 0.9980
Distance S_2–S_6 22.3035 22.3749 0.0714 0.9968 22.3741 0.0706 0.9968 22.3888 0.0853 0.9962
Distance S_3–S_4 6.9958 7.0428 0.047 0.9933 7.0432 0.0474 0.9933 7.0392 0.0434 0.9938
Distance S_3–S_5 22.3592 22.4114 0.0522 0.9977 22.3992 0.04 0.9982 22.3923 0.0331 0.9985
Distance S_3–S_6 6.5992 6.6317 0.0325 0.9951 6.6262 0.027 0.9959 6.6289 0.0297 0.9955
Distance S_4–S_5 22.3475 22.4025 0.055 0.9975 22.3907 0.0432 0.9981 22.3813 0.0338 0.9985
Distance S_4–S_6 6.6043 6.6532 0.0489 0.9927 6.6441 0.0398 0.9940 6.6442 0.0399 0.9940
Distance S_5–S_6 25.1819 25.2695 0.0876 0.9965 25.2513 0.0694 0.9973 25.2544 0.0725 0.9971
Mean 0.9959 0.9965 0.9966
Standard deviation 0.00197 0.00157 0.00160
Variance (�10—6) 3.8984 2.4541 2.5743
Table 5Summary of scale factors obtained for all measurements.
Master 1 Master 2
ISO 40 ISO 50 ISO 60 ISO 40 ISO 50 ISO 60
CT1
Mean 0.9959 0.9965 0.9966 0.9968 0.9968 0.9966
Standard deviation 0.00197 0.00157 0.00160 0.00135 0.00106 0.00098
Variance (�10—6) 3.8984 2.4541 2.5743 1.8169 1.1338 0.9664
CT2
Mean 0.9958 0.9960 0.9960 0.9966 0.9967 0.9970
Standard deviation 0.00181 0.00155 0.00202 0.00091 0.00081 0.00141
Variance (�10—6) 3.2892 2.3981 4.0827 0.8238 0.6540 1.9870
CT3
Mean 0.9962 0.9965 0.9966 0.9962 0.9963 0.9967
Standard deviation 0.00144 0.00168 0.00172 0.00098 0.00083 0.00111
Variance (�10—6) 2.0740 2.8327 2.9697 0.9581 0.6814 1.2321
CT4
Mean 0.9962 0.9965 0.9966 0.9962 0.9963 0.9967
Standard deviation 0.00144 0.00168 0.00172 0.00098 0.00083 0.00111
Variance (�10—6) 2.0740 2.8327 2.9697 0.9581 0.6814 1.2321
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 7
G Model
COMIND-2485; No. of Pages 14
measurements should be as close as possible to the calibrateddimensions of the master part (Table 1).
Consideration should be taken because the decrease ofthreshold values will reduce the internal measurements andincrease the external ones, and vice versa. Therefore, we observethat some measurements do not provide any data and othersrequire careful interpretation. Distances between planesPL2–PL4, PL3–PL5 and PL4–PL6 do not vary when decreasing
Fig. 8. Scale factor by tomography and by mast
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
threshold values; however, distances between planes PL1–PL2,PL3–PL4 and PL5–PL6 increase and distances between planesPL2–PL3 and PL4–PL5 decrease (Fig. 9).
All the measurements are analysed one by one and the scalefactor is applied to determine the threshold value corresponding tothe calibrated dimensions of the master object (Table 1). With eachof these measurements on each of the master objects in eachtomography, the threshold value that will result in the actual
er part; mean scale factor by tomography.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Fig. 9. Example of different threshold values applied in the segmentation of the
master part.
Fig. 10. Corrected measurements by ISO value for the diameter of sphere 3 (CT1,
master 1).
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx8
G Model
COMIND-2485; No. of Pages 14
measurement is predicted, establishing a linear relationship withthe obtained data. That is, it is established that the relationshipbetween the measured value and the threshold value satisfies alinear relationship and thus, any threshold value of any measure-ment can be found.
The threshold value is obtained as explained in the precedingparagraph and then it is converted to ISO scale in order to establisha comparison between the four tomographies. The obtainedthreshold value is between the peaks of the histogram represent-ing the RP material. Knowing the threshold values correspondingto the peaks, the relationship between them is established toobtain the ISO value corresponding to the calculated value bysimple interpolation (Eq. (2)).
100 � 0
ðVT ISO 100Þ � ðVT ISO ISO 0Þ ¼100 � X
ðVT ISO 100Þ � ðTHÞ (2)
where VT ISO 100 is the value of threshold corresponding to ISO-100, VT ISO 0 corresponds to the threshold value corresponding toISO-0, TH is the threshold value obtained as the final value and X isthe ISO value corresponding to the threshold value obtained asdefinitive.
Table 6 shows sample results obtained for the CT1 and masterpart 1 and specific results for sphere 3 in this situation are shown inFig. 10.
For the values presented in Fig. 10, it is possible to calculate alinear regression for the diameter of sphere 3 in this case throughEqs. (3)–(5).
Y ¼ a þ bX (3)
b ¼ nP
xy �P
xP
Y
NP
x2 �P
xð Þ2(4)
a ¼P
y
n� b
Px
n(5)
Table 6Measurements obtained for each ISO for master part 1 in CT 1.
Calibrated meas. (mm) ISO 40 I
SF = 0.9959 (Table 2) S
Meas. (mm) Corrected
Meas. (mm)
Error
(mm)
M
Diameter S_1 4.9573 4.892 4.8722 0.0851 4
Diameter S_2 4.9580 4.888 4.8682 0.0899 4
Diameter S_3 4.9573 4.874 4.8542 0.1031 4
Diameter S_4 4.9606 4.892 4.8722 0.0884 4
Diameter S_5 3.9709 3.917 3.9011 0.0698 3
Diameter S_6 3.9790 3.953 3.9370 0.042 3
P2–P3 3.8994 3.826 3.8108 0.0886 3
P3–P4 3.1704 3.339 3.3251 �0.1546 3
P4–P5 3.8958 3.786 3.7703 0.1256 3
P5–P6 7.0968 7.204 7.1748 �0.0779 7
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
where Y is the value of the actual threshold, X is the actualmeasured value, a corresponds to the trend value when X = 0 and b
corresponds to the slope of the regression line.The value of the calibrated measurement of sphere 3 and the
values obtained in Eqs. (4) and (5) are substituted into Eq. (3). Thethreshold value is then obtained. This value should be applied tothe tomography to obtain a cloud of points in which the measuredsphere diameter coincides with the calibrated diameter.
Y ¼ �11; 027:73 þ 2123:64 � X (6)
�500:1919 ¼ �11; 027:73 þ 2123:64 � 4:9573 (7)
Therefore, the obtained threshold value (�500.1919) in CT1corresponds to an ISO value of 66.74 (Eq. (2)). Table 7 contains asummary of the values obtained for all the characteristicdimensions considered.
Despite working with different values of threshold in eachtomography, the ISO values obtained are very close to each otherand it is possible to consider the correction method as stable for thefour tomographies. Table 8 contains the final threshold valuesconsidered following the method explained.
3.3. Scale factor in the final threshold value
Results have been obtained for the two correction methodspresented. It was found that the tomographies are scaled and athreshold value was obtained to separate the parts and thebackground in images, i.e., the value that defines the actualdimensions of the part. Therefore, from this final imagesegmentation we obtain the scale factor, which should matchthe factors obtained before for the considered threshold values(ISO-40, ISO-50 and ISO-60). To obtain the scale factor in this case,the same technique is used; the final cloud of points (COP) isobtained with the master parts and the encapsulation, then thedistances between the spheres of the master parts are measured inorder to obtain the scale factor for each CT. The results are shown inTables 9–11. Scale factors remain constant for the different
SO 50 ISO 60
F = 0.9965 (Table 2) SF = 0.9966 (Table 2)
eas. (mm) Corrected
Meas. (mm)
Error
(mm)
Meas. (mm) Corrected
Meas. (mm)
Error
(mm)
.929 4.9117 0.0456 4.956 4.9392 0.0181
.924 4.9067 0.0514 4.956 4.9392 0.0189
.915 4.8977 0.0596 4.943 4.9262 0.0311
.924 4.9067 0.0539 4.949 4.9322 0.0284
.948 3.9341 0.0368 3.978 3.9645 0.0065
.993 3.9790 0 4.033 4.0193 �0.0403
.876 3.8621 0.0374 3.931 3.9180 �0.0185
.223 3.2117 �0.0412 3.138 3.1274 0.043
.83 3.8169 0.079 3.957 3.9443 �0.0484
.166 7.1412 �0.0444 7.132 7.1080 �0.0111
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Table 7ISO values for the characteristic dimensions by master part and by tomography.
Sph. 1 Sph. 2 Sph. 3 Sph. 4 Sph. 5 Sph. 6 P2–P3 P3–P4 P4–P5 P5–P6 Mean CT mean
CT 1
Master 1
63.48 63.71 66.74 67.80 60.44 48.20 55.01 53.41 54.28 61.92 59.502 59.162
CT 1
Master 2
63.81 64.02 59.45 60.48 62.64 55.00 61.46 50.68 47.26 63.37 58.821
CT 2
Master 1
64.71 63.12 65.98 66.92 61.86 48.42 55.77 54.73 57.39 60.17 59.911 60.319
CT 2
Master 2
66.01 67.77 62.90 61.31 69.84 59.77 50.91 50.90 55.39 62.42 60.726
CT 3
Master 1
62.22 65.69 67.78 65.57 68.40 45.66 54.16 53.36 56.59 61.49 60.096 59.986
CT 3
Master 2
65.24 66.09 62.50 60.73 67.95 56.56 49.69 51.24 55.36 63.34 59.875
CT 4
Master 1
62.28 63.79 68.14 68.07 62.86 49.65 54.69 53.70 57.14 62.58 60.294 59.741
CT 4
Master 2
65.29 66.69 60.19 62.85 59.48 58.87 50.24 51.00 54.78 62.44 59.188
Table 8Final threshold values for each tomography.
ISO Threshold value Final threshold value
CT 1 59.1616 �566.497813 �566
CT 2 60.3187 1387.3294 1387
CT 3 59.9857 259.73479 260
CT 4 59.7409 1059.09328 1059
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 9
G Model
COMIND-2485; No. of Pages 14
tomographies and they agree with those obtained in the previoussections. This confirms again that the established scale factors areconstant throughout the entire grey-scale range corresponding tothe RP material and that the measured data in any of the parts ofthe tomographies are scaled by the corresponding factor. With this,once the measurements required from the encapsulations areobtained from the final COPs, these scale factors obtained againstthe masters on each of the tomographies are applied to obtain thecorrected measurements of the encapsulation dimensions. Itshould be noted that in this paper, a global scale factor is usedthat is obtained by the procedure described for the threedirections: X, Y and Z. It may be necessary, depending on thecharacteristics of the CT machine, to use two or three differentscale factors in cases where the error sources of the machine andprocedure affect the voxel size differently depending on the
Table 9Distances between spheres and scale factors in definitive COPs with master 1.
CT 1 master 1 CT 2 master 1
Dist. (mm) S.F. Dist. (mm) S
Distance S_1–S_2 7.0415 0.9939 7.0476 0
Distance S_1–S_3 19.4482 0.9962 19.4336 0
Distance S_1–S_4 18.0989 0.9976 18.1031 0
Distance S_1–S_5 6.6697 0.9974 6.6708 0
Distance S_1–S_6 22.3994 0.9961 22.4085 0
Distance S_2–S_3 18.1140 0.9969 18.1004 0
Distance S_2–S_4 19.4054 0.9974 19.4152 0
Distance S_2–S_5 6.6825 0.9971 6.6869 0
Distance S_2–S_6 22.3841 0.9964 22.3972 0
Distance S_3–S_4 7.0380 0.9940 7.0425 0
Distance S_3–S_5 22.4118 0.9977 22.3920 0
Distance S_3–S_6 6.6188 0.9970 6.6393 0
Distance S_4–S_5 22.3825 0.9984 22.3825 0
Distance S_4–S_6 6.6420 0.9943 6.6510 0
Distance S_5–S_6 25.2533 0.9972 25.2549 0
0.9965 0
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
direction (e.g., rotary table, linear axis, etc.). In cases where thedeviations of error are significant depending on the principaldirections, different scale factors should be chosen for use in eachdirection. In our case, the obtained results show that the use of aglobal scale factor is suitable in the proposed correction scheme.
3.4. Conclusions of correction methods
Two correction methods have been applied. The first is based onobtaining the scale factor of the images. It has been observed thatthis factor remains constant for all ISO values analysed. The secondmethod is based on obtaining an ISO value that, combined with thescale factor, will obtain the COP in which the measurementscorrespond to the dimensions of the actual part.
Usually, when representing or analysing a tomographic image,any correction method is applied using a default ISO-50 value todetermine the threshold value considered in the segmentation. Inthis case, Table 12 shows the results of the master 2 in CT4 toillustrate the efficiency of the correction method implementedversus a traditional segmentation without correction.
Table 12 shows, as a final result in each situation, the meanerror regarding the calibrated dimensions of master 2. Afterapplying the two correction methods, the error is reduced to60.37% of the initial error.
CT 3 master 1 CT 4 master 1
.F. Dist. (mm) S.F. Dist. (mm) S.F.
.9931 7.0401 0.9941 7.0434 0.9937
.9969 19.4126 0.9980 19.4410 0.9965
.9974 18.1015 0.9975 18.1096 0.9970
.9973 6.6748 0.9967 6.6765 0.9964
.9957 22.3915 0.9964 22.4194 0.9952
.9976 18.1066 0.9973 18.1036 0.9974
.9969 19.4368 0.9958 19.4153 0.9969
.9965 6.6647 0.9998 6.6869 0.9965
.9958 22.4086 0.9953 22.4015 0.9956
.9934 7.0401 0.9937 7.0425 0.9934
.9985 22.3753 0.9993 22.4100 0.9977
.9940 6.6437 0.9933 6.6381 0.9941
.9984 22.3863 0.9983 22.3994 0.9977
.9930 6.6324 0.9958 6.6502 0.9931
.9971 25.2463 0.9974 25.2784 0.9962
.9961 0.9966 0.9958
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Table 10Distances between spheres and scale factors in definitive COPs with master 2.
CT 1 master 2 CT 2 master 2 CT 3 master 2 CT 4 master 2
Dist. (mm) S.F. Dist. (mm) S.F. Dist. (mm) S.F. Dist. (mm) S.F.
Distance S_1–S_2 7.0289 0.9961 7.0281 0.9962 7.0278 0.9963 7.0284 0.9962
Distance S_1–S_3 19.4099 0.9960 19.4061 0.9962 19.4018 0.9964 19.4142 0.9957
Distance S_1–S_4 18.0804 0.9963 18.0762 0.9965 18.0963 0.9954 18.0857 0.9960
Distance S_1–S_5 6.6750 0.9982 6.6772 0.9979 6.6808 0.9974 6.6865 0.9965
Distance S_1–S_6 22.3395 0.9964 22.3455 0.9962 22.3471 0.9961 22.3526 0.9958
Distance S_2–S_3 18.0874 0.9962 18.0853 0.9963 18.0862 0.9962 18.0956 0.9957
Distance S_2–S_4 19.3940 0.9964 19.3916 0.9966 19.4159 0.9953 19.4028 0.9960
Distance S_2–S_5 6.6765 0.9996 6.6842 0.9984 6.6803 0.9990 6.6866 0.9980
Distance S_2–S_6 22.3364 0.9966 22.3452 0.9963 22.3475 0.9961 22.3524 0.9959
Distance S_3–S_4 7.0297 0.9959 7.0291 0.9960 7.0307 0.9958 7.0310 0.9958
Distance S_3–S_5 22.3896 0.9977 22.3848 0.9979 22.3798 0.9981 22.4020 0.9971
Distance S_3–S_6 6.6212 0.9968 6.6369 0.9945 6.6304 0.9955 6.6294 0.9956
Distance S_4–S_5 22.3791 0.9978 22.3725 0.9981 22.3905 0.9973 22.3920 0.9972
Distance S_4–S_6 6.6244 0.9954 6.6378 0.9934 6.6266 0.9951 6.6343 0.9939
Distance S_5–S_6 25.2053 0.9980 25.2045 0.9980 25.2037 0.9980 25.2221 0.9973
0.9969 0.9966 0.9965 0.9962
Table 11Means of scale factors in final COPs.
CT 1 CT 2 CT 3 CT 4
Master 1 Master 2 Master 1 Master 2 Master 1 Master 2 Master 1 Master 2
Scale factor 0.9965 0.9969 0.9961 0.9966 0.9966 0.9965 0.9958 0.9962
0.9967 0.9963 0.9965 0.9960
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx10
G Model
COMIND-2485; No. of Pages 14
4. Deformation analysis
During the incubator test carried out with the encapsulation for24 h at 37 8C, it is observed that it is deformed in the central zone(chip housing), undoing the hermeticity generated by the O-rings.This circumstance will cause an invalid test because of the leakingof the fluid applied. The O-rings serve the purpose of hermeticallysealing the communication microchannels of the bottom part ofthe encapsulation with those of the top part and of maintaining theseal under deformation. For the success of the incubator test, it isnecessary to control the deformation of the chip housing and tominimise it by redesigning the encapsulation. For this, it isessential to know the deformation that is obtained with the currentdesign and the different solicitations; therefore, it is necessary toknow which deformation belongs to each stress solicitation,characterising the behaviour of the encapsulation in the area ofinterest.
The deformation of CT1 is due solely to the deformationresistance of the silicon O-rings, as the encapsulation of this
Table 12Error correction in CT 4, master 2.
Nominal ISO-50 without scale factor
Diameter (mm) Diameter (mm) Error (mm)
Sphere 1 4.9530 4.9110 0.0420
Sphere 2 4.9693 4.9140 0.0553
Sphere 3 4.9691 4.9440 0.0251
Sphere 4 4.9544 4.9170 0.0374
Sphere 5 3.9750 3.9430 0.0320
Sphere 6 3.9787 3.9590 0.0197
P_2–P_3 3.9023 3.9118 0.0095
P_3–P_4 3.1591 3.1774 0.0183
P_4–P_5 3.8811 3.8723 0.0088
P_5–P_6 7.0889 7.1636 0.0747
0.0323
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
tomography has not been tested in an incubator. The deformationin CT2 is practically negligible because it belongs to a newlymanufactured encapsulation without testing or mounting com-ponents. In contrast, CT3 will only give us the deformationgenerated by the test in the incubator and finally, CT4 will giveprovide the deformation caused by both, i.e., by the O-rings and bythe test for 24 h at 37 8C.
The deformation zone analysed is the peripheral area of theencapsulation near the chip housing owing to the diversity ofmaterials present in the central area (chip, O-rings and theencapsulation itself), which contributes a considerable decrease inthe measurement accuracy regarding the analysis in the areawhere there is only RP material. As explained in previous sections,the COP dimensions depend on the chosen threshold value. In thiscase, the choice of the threshold value corresponds to the RPmaterial area. This value does not match the ideal threshold valuesof elements of different materials, such that tomographic images 1and 4, in which more than just RP materials are present, areobtained with more noise in the zone where these elements are
ISO-59 without scale factor ISO-59 with scale factor
Diameter (mm) Error (mm) Diameter (mm) Error (mm)
4.9570 0.0040 4.9372 0.0158
4.9650 0.0043 4.9452 0.0241
4.9780 0.0089 4.9582 0.0109
4.9710 0.0166 4.9512 0.0032
3.9820 0.0070 3.9661 0.0089
4.0010 0.0223 3.9850 0.0063
3.9097 0.0074 3.8941 0.0082
3.1414 0.0177 3.1289 0.0302
3.9490 0.0679 3.9333 0.0522
7.1530 0.0641 7.1245 0.0356
0.0220 0.0195
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Fig. 11. Deformation measured by corrected CTs in the region of the chip housing of the four encapsulations tested.
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 11
G Model
COMIND-2485; No. of Pages 14
located. Thus, this circumstance reduces drastically the measure-ment accuracy. As deformation also exists in the peripheral zone,the work is to analyse the most accurate way in which to minimiseit in later designs; thus, reducing it in the middle zone. Fig. 11shows the measured deformation after corrections in the region ofinterest of the four analysed encapsulations. In case of multi-material measurements, the main problem is to obtain theoptimal threshold for each material, which may be differentbetween each material and air and between two differentmaterials. Generally, in a CT measurement with the sameacquisition parameters (e.g., voltage, current and time ofexposure), the presence of different materials will cause artefactsor noise in the intermediate zone. Thus, it is not possible toreconstruction accurately the geometries of different materialswith the same threshold. Different multi-energy techniques havebeen developed for optimal threshold selection and accurateseparation of materials. They combine CT measurements withhigh-energy exposure for materials with high absorption and low-energy exposure for materials with low absorption. This allowsthe combination of information from both measurements in the2D image phase or in the 3D voxel reconstruction to separateaccurately the two materials [17–19]. For practical purposes,depending on the density difference of the materials and thecapabilities of the CT machine available, a possible solution toovercome the pointed limitation is to make the master parts withcalibrated interfaces between the materials involved, in order todefine later the optimal threshold for the multi-material areas.This would allow accurate separation of materials in the centralarea of the packaging when the density of the materials involved issimilar.
Table 13 describes the deformation measured in eachtomography in a grid, such that it is possible to analyse specificpoints and to compare them through the different tomographies.It should be noted that the deformation shown belongs tothe height difference between the lower face of the upper part ofthe encapsulation and the upper face of the bottom part(Fig. 12a).
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
As seen in Table 13, the deformation belongs to the peripheralpoints of the central zone and it shows the results without applyingthe scale factor and the results with the correction applied. In CT 4,the point is indicated where the maximum deformation ismeasured in the four grids. The obtained results are coherentbecause the tomography presenting the greatest deformation isthe one corresponding to the encapsulation fully assembled andtested in the incubator, followed by the encapsulation fullyassembled and not tested. Subsequently, it is observed that theencapsulation measured in CT 3, which carries no component andwas tested, presents a greater deformation than the untested andunassembled encapsulation measured in CT 2. Therefore, asexpected, the observed deformations are the result of both thecomponents loads and test in the incubator, due to the viscoelasticbehaviour of the material.
Once guaranteed of an accurate CT measurement of thedeformation in the region of the chip housing, it is possible toaddress a finite elements (FE) simulation work to predict thedeformation in a particular geometric configuration and for testtemperature conditions. This can lead to a geometric optimisationof the encapsulation and its components to control the deforma-tion, in order to avoid exceeding the allowable limits of the test,once the Fe model is validated with the experimental resultsobtained in this work. These limits are those that make the O-ringswork in the range of hermetical sealing of the connections betweenthe microchannels of the top and bottom parts of the encapsula-tion. Therefore, the aim will be to design an encapsulation whosemaximum deformation may make the O-rings remain compressedthroughout the test. Furthermore, it is also possible to obtaincorrected dimensions of interest of the encapsulation geometry.Thus, it is possible to characterise accurately the influence andrelationship of these dimensions with the biological results of thetests. This can result in a dimensional optimisation process of theencapsulation, not only to prevent deformations but also from afunctional standpoint. As an example, Table 14 shows themeasurement results of two simple dimensions of width andthickness (Fig. 12b) for the four tested encapsulations.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
Table 13Grid of deformation values by tomography (Fig. 12b).
X (mm) Y (mm)
1 2 23 24 1 2 23 24
Without scale factor With scale factor
CT 1
1 0.3707 0.3741 0.4758 0.4749 0.3695 0.3729 0.4742 0.4733
2 0.3942 0.4132 0.5031 0.5051 0.3929 0.4118 0.5014 0.5035
8 0.4713 0.4977 0.5979 0.5838 0.4698 0.4960 0.5959 0.5818
9 0.4756 0.4924 0.6002 0.5775 0.4741 0.4908 0.5982 0.5756
10 0.4672 0.4952 0.5883 0.5585 0.4656 0.4936 0.5863 0.5567
16 0.2757 0.3056 0.4036 0.4058 0.2748 0.3046 0.4023 0.4044
17 0.2383 0.2610 0.3789 0.3573 0.2375 0.2601 0.3776 0.3561
CT 2
1 0.1567 0.1576 0.0963 0.0963 0.1561 0.1570 0.0960 0.0960
2 0.1462 0.1593 0.0963 0.0963 0.1456 0.1588 0.0960 0.0960
8 0.1139 0.1212 0.0595 0.0255 0.1135 0.1207 0.0593 0.0254
9 0.1438 0.1136 0.0595 0.0204 0.1433 0.1132 0.0593 0.0204
10 0.1438 0.1190 0.0198 0.0483 0.1433 0.1186 0.0197 0.0482
16 0.1304 0.1465 0.0837 0.0617 0.1299 0.1460 0.0834 0.0615
17 0.1517 0.1592 0.0836 0.0723 0.1511 0.1586 0.0833 0.0720
CT 3
1 0.1399 0.1366 0.0463 0.0376 0.1394 0.1361 0.0461 0.0374
2 0.1363 0.1493 0.0529 0.0426 0.1359 0.1488 0.0527 0.0425
8 0.1276 0.1553 0.0835 0.0595 0.1272 0.1548 0.0833 0.0593
9 0.1314 0.1618 0.0802 0.0565 0.1310 0.1613 0.0799 0.0563
10 0.1350 0.4952 0.0881 0.0617 0.1345 0.4935 0.0878 0.0615
16 0.1682 0.1786 0.0952 0.0880 0.1676 0.1780 0.0948 0.0877
17 0.1778 0.1855 0.1101 0.0939 0.1772 0.1849 0.1097 0.0936
CT 4
1 0.4258 0.4446 0.6284 0.6319 0.4241 0.4428 0.6259 0.6293
2 0.4627 0.4849 0.6674 0.6578 0.4608 0.4829 0.6648 0.6552
8 0.5397 0.5892 0.7905 0.7731 0.5375 0.5868 0.7873 0.7701
9 0.5315 0.5671 0.7872 0.7463 0.5294 0.5648 0.7841 0.7433
10 0.5267 0.5741 0.7853 0.7535 0.5246 0.5718 0.7821 0.7505
16 0.3194 0.3533 0.5514 0.5477 0.3181 0.3519 0.5492 0.5455
17 0.2839 0.3116 0.5194 0.5128 0.2828 0.3104 0.5173 0.5107
Fig. 12. Corrected cloud of points of one encapsulation. (a) Top and bottom parts; (b) width and thickness dimensions defined for the bottom part and reference XY grid for
sampling deformation values.
Table 14Sample dimensional measurements of the bottom part of the encapsulations.
CT1 CT2 CT3 CT4
Width (mm)
Without scale factor 18.1243 18.1297 18.1288 18.1229
With scale factor 18.0645 18.0627 18.0654 18.0504
Thickness (mm)
Without scale factor 5.9836 5.9819 5.9816 5.9783
With scale factor 5.9638 5.9598 5.9607 5.9544
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx12
G Model
COMIND-2485; No. of Pages 14
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
5. Conclusions
Nowadays, RP technologies play an important role in thedevelopment of microfluidic devices for biomedical testing. Owingto the geometric characteristics, the fabrication of these devices isextremely complex by conventional manufacturing techniqueswithout incurring the high development costs of dedicatedinjection moulds. In many cases, these devices are designed forsingle or few uses in specific clinical cases. Given the accuracy thatcan be achieved today with some of the available additivemanufacturing technologies, it is possible to obtain end-use partsfor these applications. Generally, the characteristics of RPmaterials, together with the temperature testing conditions with
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx 13
G Model
COMIND-2485; No. of Pages 14
the microfluidic systems, lead to several problems related to thedeformation of these devices. Therefore, it is necessary tocharacterise the geometric accuracy that can be achieved indevices fabricated using these technologies and their relationshipwith future test conditions. On the other hand, it is critical topredict the deformation suffered by the encapsulation and therelationship of this deformation with the success of the clinicaltests, in order to feed back the encapsulation geometric designalternatives, to ensure the success of tests performed with piecesobtained by these materials and by layer manufacturing.
This paper shows the capacity to manufacture encapsulationsfor microfluidic chips using RP technologies. It has shown thepotential that RP technologies have in the manufacture of devicesfor biomedical applications as end pieces. The use of thesetechnologies offers complete freedom to the designer of the testsetup and the possibility of systematic and rapid designoptimisation of its components in relation to cell behaviour in aparticular test.
Although some of the current RP technologies are able to obtainsmall parts within the required tolerance, the materials commonlyused in these manufacturing technologies, together with the testconditions, brings problems of deformation of the parts thatencapsulate the chips that affect the running of these tests. In orderto obtain an optimum design for the packaging of the chips, it isnecessary to know the exact mechanics of these deformations andtheir effect on the dimensions of the pieces.
CT has been presented as a suitable dimensional verificationtechnique for these types of small parts with complex geometry,owing to its non-contact measurement capability and the ability tocapture outer and inner geometry versus conventional measure-ment techniques, resulting in 3D reconstruction of the measuredpiece. The analysis presented shows the benefits of using masterpieces of the same material and geometric characteristics similarto those of the piece to be checked. Two correction techniques havebeen presented. These techniques are applicable directly on CTimages and globally, in the cloud of points or STL file ultimatelyrebuilt in order to reduce the measurement uncertainty byimproving the overall accuracy of the reconstruction. Finally,the workflow in the dimensional characterisation through pointcloud analysis has been shown.
References
[1] M. Koc, T. Ozel, Micro-Manufacturing: Design and Manufacturing of Micro-Products, John Wiley & Sons, Inc., 2011.
[2] P.S. D’Urso, et al., Stereolithographic biomodelling in cranio-maxillofacial sur-gery: a prospective trial, Journal of Cranio-Maxillofacial Surgery 27 (1) (1999) 30–37.
[3] A. Muller, et al., The application of rapid prototyping techniques in cranialreconstruction and preoperative planning in neurosurgery, Journal of CraniofacialSurgery 14 (6) (2003) 89–914.
[4] H. Zenha, et al., The application of 3-D biomodeling technology in complexmandibular reconstruction – experience of 47 clinical cases, European Journalof Plastic Surgery 34 (4) (2011) 257–265.
[5] C. Herlin, M. Koppe, J.L. Beziat, A. Gleizal, Rapid prototyping in craniofacialsurgery: using a positioning guide after zygomatic osteotomy – a case report,Journal of Cranio-Maxillofacial Surgery 39 (5) (2011) 376–379.
[6] W.-Y. Yeong, C.-K. Chua, K.-F. Leong, M. Chandrasekaran, Rapid prototyping intissue engineering: challenges and potential, Trends in Biotechnology 22 (12)(2004) 643–652.
[7] V. Calvo, et al., A highly integrated vertical SU8 valve for stepwise in-seriesreactions, Journal of Micromecanics and Microengineering 21 (2011) 065037.
[8] L.J. Fernandez, et al., Study of functional viability of SU-8-based microneedles forneural applications, Journal of Micromecanics and Microengineering 19 (2009)025007.
[9] K.L.A. Chan, X. Niu, A.J. de Mello, S.G. Kazarian, Rapid prototyping of microfluidicdevices for integrating with FT-IR spectroscopic imaging, Lab on a Chip 10 (2010)2170–2174.
[10] K. Kiekens, et al., A test object with parallel grooves for calibration and accuracyassessment of industrial computed tomography (CT) metrology, MeasurementScience and Technology 22 (2011) 115502.
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
[11] J. Kumar, A. Attridge, P.K.C. Wood, M.A. Williams, Analysis of the effect of cone-beam geometry and test object configuration on the measurement accuracy of acomputed tomography scanner used for dimensional measurement, Measure-ment Science and Technology 22 (2011) 035105.
[12] J.P. Kruth, et al., Computed tomography for dimensional metrology, CIRP Annals –Manufacturing Technology 60 (2) (2011) 821–842.
[13] H. Bulu, A. Alpkocak, Comparison of 3D segmentation algorithms for medicalimaging, in: Twentieth IEEE International Symposium on Computer-Based Medi-cal Systems, 2007, 269–274.
[14] K. Kiekens, et al., A test object for calibration and accuracy assessment in X-ray CTmetrology, in: Proceedings of the IMEKO 10th International Symposium onMeasurement and Quality Control, 2010, B6_86_1-4.
[15] A.K.M. Elshennawy, Performance evaluation of coordinate measuring machines,(PhD thesis), The Pennsylvania State University, University Park, PA, 1987.
[16] C. Reinhart, Industrial computer tomography – a universal inspection tool, in:17th World Conference on Nondestructive Testing, 2008.
[17] Ph. Kramer, A. Weckenmann, Multi-energy image stack fusion in computedtomography, Measurement Science & Technology 21 (2010) 045105.
[18] Ch. Heinzl, J. Kastner, E. Goller, Surface extraction from multi-material compo-nents for metrology using dual energy CT, IEEE Transactions on Visualization andComputer Graphics 13 (6) (2007) 1520–1527.
[19] M. Knaup, Ph. Stenner, M. Kachelrieß, Rawdata-based dual energy CT (DECT) frominconsistent scans, in: IEEE Nuclear Science Symposium Conference Record 26-268, 2007, 4457–4459.
Jorge Santolaria received the B.S., M.S., and Ph.D.degrees in mechanical engineering from the Universi-dad de Zaragoza, Zaragoza, Spain, in 1998, 2000, and2007, respectively. He is currently an Associate Profes-sor in the Dept. of Design and Manufacturing Engineer-ing at the Universidad de Zaragoza. He is the head of theManufacturing Engineering and Advanced MetrologyGroup (GIFMA) which is a member of the AragonInstitute of Engineering Research (I3A). His researchinterests include precision engineering, microtechnol-ogy, modelling and calibration of coordinate measuringsystems, laser-based and noncontact measurementsystems, optimisation and error correction computa-tional methods for robot and portable measuringmachines kinematic calibration, gear geometry andgear manufacturing and inspection processes andadditive manufacturing technologies.
Rosa Monge received the B.S. and M.S. degrees inmechanical engineering at Zaragoza University (Spain)in 2010 and 2011, respectively. At the moment, she isworking in her PhD thesis under the supervision of Dr.Luis J. Fernandez in the development of microtechnol-ogies in the field of microfluidic devices for cell cultureapplications.
Angel Tobajas received the B.S., M.S. degrees inmechanical engineering from the Universidad deZaragoza, Zaragoza, Spain, in 2006 and 2012, respec-tively. He works currently at AlphaSIP as MechanicalEngineer in the R&D department.
Roberto Jimenez received the B.Sc. and M.Sc. degreesin industrial engineering in 2002, and the PhD degree in2009 from the Universidad de Zaragoza (Spain). He iscurrently an Associate Professor in the Centro Uni-versitario de la Defensa, Zaragoza, Spain. He is memberof the Manufacturing Engineering and AdvancedMetrology Group (GIFMA) which is a member of theAragon Institute of Engineering Research (I3A). Hisresearch interests include precision engineering, addi-tive manufacturing technologies, process optimisation,computerised tomography for industrial applicationsand other noncontact measurement systems.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015
J. Santolaria et al. / Computers in Industry xxx (2013) xxx–xxx14
G Model
COMIND-2485; No. of Pages 14
Mirko A. Cabrera received the B.S. degree in mechani-cal engineering from the Universidad de Zaragoza,Zaragoza, Spain, in 2012. His degree thesis was aboutthe design of specific packaging for microfluidic devicesfor cell culture applications.
Please cite this article in press as: J. Santolaria, et al., Design, manufaencapsulations by computed tomography, Comput. Industry (2013)
Luis J. Fernandez received the M.S. degree in Physics atZaragoza University (Spain) in 2001. He then joined theTransducers Science and Technology department atTwente University (The Netherlands) where heobtained the PhD degree in 2005 for the developmentof MEMS based power sensor for radio frequencysignals. During the next 5 years he was involved in thedevelopment of low cost Micro Total Analysis Systems(mTAS), microfluidic control elements and biosensorsat Ikerlan research institute (Spain), becoming a formalpartner of the centre. In 2010 he received the ‘‘Ramon yCajal Fellowship’’ to join the GEMM group at ZaragozaUniversity (Spain) and lead scientific advances in thefield of microfluidic devices for cell culture applications.He has published 22 scientific papers with SCI, 39articles in proceedings of international conferences,and holds 9 patents.
cture and geometric verification of rapid prototyped microfluidic, http://dx.doi.org/10.1016/j.compind.2013.06.015