Investigating the Dimensional Accuracy and Surface Roughness for 3D Printed Parts Using a Multi-jet Printer

The shortcoming of conventional manufacturing (CM) is that it cannot manufacture geometrically complex parts with high repeatability and good surface properties. In order to overcome these shortcomings of CM, additive manufacturing (AM) is the major alternative to the CM. However, the usefulness and performance of parts manufactured through AM are closely correlated with dimensional accuracy and surface roughness, SR (Ra). Therefore, an investigation was carried out in this study for dimensional accuracy and surface roughness of 3D printed parts fabricated in different orientations. In the investigation, four orientation patterns are considered. The part is lying on the base (A), part is lying on the long edge (B), part is lying on the short edge (C), and the part is inclined to 45°(D) to the surface of the base plate (refer to Fig. 2). Orientations, i.e., A, B, C, and D, were explored for the variations in dimensional deviation and SR. In addition, an analysis was carried out using scanning electron microscopy (SEM) on fabricated parts. The results obtained exhibit a variation in dimensional accuracy and change in SR with different part orientations. Among all orientations, the largest surface area of the component in contact with the base plate (A) was the most suitable.


Introduction
Additive manufacturing (AM) is defined as the layer-bylayer addition process to obtain a 3D printed part (Ref 1, 2). It is not much more beneficial than subtractive manufacturing in terms of surface finish and dimensional accuracy ( Ref 3,4). The parts were designed with the Computer-Aided Designing (CAD) software ( Ref 5,6).This CAD software provides the appropriate 3D printing file ( Ref 7). Every 3D printer has software that is used as an interface between CAD and 3D printers, with which a layer thickness can be varied (Ref 8,9). The ability to produce complex geometry parts and reduced production time have contributed to the popularity of AM processes ( Ref 10,11). The 3D printing process can be used for rapid prototyping, direct manufacturing, jewelry industry, health, dental, architecture, and quick fixtures ( Ref 12,13).
As per the application of the mechanical parts, interchangeability is one of the critical requirements. Due to the limited life cycle of the working parts, they have to be replaced with new features. So whenever the 3D printing fabricates these parts, there is no fixed orientation on the base plate to place these parts. Many of these parts are placed on the base plate to use the maximum amount of the base plate. When these mechanical parts are positioned on the base plate, they may be arranged horizontally, vertically, or inclined at a given degree to minimize the printing cost. In order to attain interchangeability, when the parts which are fabricated in different orientations will be replaced, their performance and life will differ. The life of these parts will be different because they will have different coefficients of friction (due to different roughness levels) with the adjusted mechanical product. The outcome of this article is to obtain the best orientation with the least dimensional deviation and surface roughness. On the same set of exposures, some authors have investigated the variation in the mechanical properties ( Ref 14).
Multi-jet modeling (MJM) and material jetting (MJ) have the same working principle. But manufacturers of 3D printers have patented with different names. When comparing FDM & MJP, the latter is preferred because it provides better interlayer bonding. In the current investigation, the effect of orientationbased deposition on dimensional variation and surface roughness is studied. In MJP, a print head deposits liquid drops on the platform as per the geometry embedded in the CAD model. The table on which the layer-by-layer deposition is to be done mainly consists of the photopolymer material for the part (build material) and wax (support material). Wax is used as the support material, and then, it is removed from the part material (photosensitive resign) by post-processing. During the postprocessing, the part fabricated from the 3D printer is heated in saturated refined oil for 20 min under different sets of vibrations at 35-40°C temperature. After the post-processing, the wax is dissolved in the oil bath and the solid required part is cleaned with fresh water and becomes ready for use.
Accuracy of a surface is important for friction between two surfaces, contact deformation, heat and electronic current conduction, tightness of the contact joints, and positional accuracy of the surface. The SR is the subject of experimental and theoretical investigation ( Ref 15,16). SR parameters can be calculated in the two-dimensional (2D) and three-dimensional (3D) domains. For more than half of the century, 2D profile analysis is used for engineering and scientific applications ( Ref 17,18). In the current study, amplitude parameters are considered as the main parameter to characterize surface morphology. The amplitude parameters are used to measure the vertical characteristics of surface variation. In general quality control, the average height (R a ) parameter or the center line average (CLA) is universally used roughness parameters (Ref 19). It is defined as the average absolute variation in the roughness from the mean line over the one sampling length. It is easy to define and measure, and a general description of height variation makes it more user-friendly. The challenge among the R a is that it is not sensitive to changes in wavelengths (Ref 20, 21).

Research Question (RQ)
The bonding of the layers in FDM happens due to the polymer-based filament heating up to the plastic stage under gravity's effect. The bonding area between the two layers is less, leading to the formation of staircase defects and high thermal contraction. In MJP, however, the liquid material is used for both the primary and support material. Due to fluid layer interaction, the chances of thermal contraction and slow cooling rate require attention for exploring dimensional accuracy and surface properties.

Background
A substantial research contribution has been made toward the benchmarking of various AM processes. Singh et al. (Ref 22) in this context investigated the effect of FDM process parameters on the dimensional accuracy of the parts produced. They examined the volume, density, and a number of coatings in the investment casting (IC). They proposed an alternative method to develop nylon-6 from waste and tailored properties for aluminum matrix composite (AMC). Beniak et al. (Ref 23) considered the FDM process to know the effect of layer thickness and printing temperature on the shape and dimensional accuracy of the fabricated parts. They concluded that the printing temperature was most significant to control the variations in the shape and the dimensions. Hyndhavi et al. (Ref 24) studied the effect of process parameters (layer thickness, raster angle, and build orientation) and their interaction on the dimensional accuracy of the specimen fabricated through the FDM process with acrylonitrile butadiene styrene (ABS) and polylactic acid (PLA) materials. They applied Grey relational analysis (GRA) method to optimize the parameters, which yielded a minimum percentage change in the parts' length, width, and thickness. Kozak et al. (Ref 25) considered the selective laser sintering / selective laser melting (SLS/SLM) process to investigate the effect of process parameters (laser power, spot diameter, and exposure time) on the shape and dimensional accuracy and surface quality of manufactured parts. A new dimensional thermal model was developed for the estimation of process parameters along with a sensitivity analysis to validate the results. Sahu et al. (Ref 26) presented a fuzzy decision-making logic in integration with the Taguchi method to improve the dimensional accuracy of FDM processed ABS parts. The process parameters used in this study were: layer thickness, orientation, raster angle, raster width, and air gap to observe variation in dimensions of length, width, and thickness of the parts. It was reported that shrinkage was pre-dominant in length and width directions, but dimensions expanded in the thickness direction. Noriega et al. (Ref 27) proposed the neural network approach to reduce dimensional error for the parallel face of the ABS prismatic part fabricated on FDM. The method increased the accuracy of manufactured parts. Tronvoll et al. (Ref 28) investigated the dimensional accuracy of the thread manufactured through the FDM process. They proposed the narrowing of thread profiles which increased the performance of printed threads. Mou et al. (Ref 29) considered three different AM processes to fabricate parts with protruded surfaces (like holes, flat surfaces, micro-scale ribs, and long channels). They used non-contact methods to record the dimensional accuracy and surface roughness. It was found that the FDM process was the least accurate in terms of dimensional accuracy and surface roughness, whereas MJM (ProJet MJP 3600 3D printer used) produced the most precise parts. Garg  After going through the literature, some of the important points are as follows. Much study has been done on the influence of parametric variation, various materials, different material combinations, and fabrication orientations on FDM parts' dimensional deviation and surface behavior. However, there has been less research on the effect of component orientations on surface behavior in the case of multi-jet printers. As a result, the present research investigates the impact of part orientation on dimensional accuracy and surface roughness behavior. Figure 1 depicts a schematic representation of the study's approach.

Dimensional Analysis
One of the primary benefits of 3D printing is the capacity to manufacture complicated forms and geometries. But accuracy and repeatability are required for interchangeability. So in the present investigation for the MJP-based 3D printer, precision and repeatability are investigated. A part is placed in different orientations to check the precision and reparability. The objective of arranging comparable parts in various orientations is to maximize the base platesÕs limited space utilization. The cost of each unit will increase if the whole surface of the base plate is not used for similar components. However, to optimally use the base plate, dimensional deviations due to different orientations of components on the base plate will not be accepted to fulfill industry requirements for interchangeability.
In order to examine the geometries, dimensional accuracy, and surface roughness of MJP-based produced components (SR), the part is designed which is a combination of a rectangle, cylinder, and fillets. These shapes provide an opportunity to examine the linear and radial dimensions related to different orientations of the part. In orientations A, B, C, and D, the product is aligned with the faces II, IV, and I and inclined at 45°t o the face I and II, as shown in Fig. 2. The references for the dimensions for comparison between the actual and nominal values are represented in Fig. 3. Four parts subjected to different orientations as illustrated in Fig. 3(a), (b) and (c) were fabricated by using MJP (MJP-2500 modeled). All parts are fabricated with the same material, that is, VisiJet M2R-WT (3D systems). The properties of the material are given in Table 1 (Ref 32).
The MJP 3D printer was used for the fabrication of the 3D part. In this model of the MJP-3D printer, there is no such provision for changing the operating parameters. Some tech-nical specifications of the printer are printing mode-high definition (HD), net build volume (XYZ) 294 9 211 9 144 mm, resolution (XYZ) 800 9 900 9 790 DPI (Dots Per Inches), 32-l (micron) layers. Support material (VisiJet-M2 -SUP) is the only completely compatible support material for Project MJP2500, for all the range of primary materials that can be used on the printer (Ref 32). After fabricating the part on the MJP 3D printer, the post-processing of the printed part was done. In the post-processing, the part was put into the oil bath and preheated at 35-45°C for ten minutes. Due to post-processing, the support material will dissolve in the oil bath, and the required part will be left behind. After taking out part from the oil bath, the part was washed with tap water and is ready to use. The printed parts were measured by the Coordinate Measuring Machine (CMM) (Quick Vision WLI Series 363-CNC video measuring system with white light interferometry), which can measure the product dimensions without contact (measuring range of 300 9 200 9 190 mm). Here, the height (Z-print axis) is measured at four different points (H 1 , H 2 , H 3 , and H 4 ). Two points are examined along the X-axis (length), namely (L5 and L6). Six points are considered along the Y-Axis (width) (W 1 , W 4 , W 5 , W 7 , W 8 and W 10 ). The radius (R 1 , R 2 and R 3 ) and diameter (D 4 , D 5 and D 6 ) were also measured. End-to-end measurements were taken for height, length, breadth, and diameter. The radial distance from the geometric center to the outer surface was considered, and the average of five observations was regarded as a single observation. Table 2 reflects the nominal values of various parameters considered in the investigation.
The SR value for the fabricated parts on different faces was measured by the roughness tester (IN-SIZE-ISR-S400 model). It can measure R a , R q , R t , R z , R c , R max , R rsm , R pc . The device has a range of R a 0-100 lm, accuracy ± 3%, resolution of R a is 0.001 lm (Micro Meter), diamond type of the probe is used,  Qanta FEG 250, the quanta line scanning electron microscope, was used. It is a versatile, high-performance instrument with three modes (high vacuum, low vacuum, and field emission scanning electron microscope). The samples for the SEM were taken out from fabricated parts, and the top surfaces of all the four samples were investigated.

Surface Roughness Analysis
The actual behavior of any manufactured part with the environment is the function of its surface roughness. The parts having high surface roughness will wear out more and have a high coefficient of friction. The life of the mechanical part is also a function of its tribology. If the surface roughness is high, the chances of the formation of cracks will also be more. The high roughness values are undesirable in the field of 3D printing because subsequent operations are required to reduce SR which will lead to the high part cost. The present work measures the SR for the fabricated parts on eleven faces of all four orientations (A, B, C & D). The eleven faces are face I, face II, face III, face IV, face V, face VI, solid curved face, cylindrical part, inclined part, top rectangular part-I, and top rectangular part-II as shown in Fig. 3(d) and (e). These measurements were carried out at four different locations along X-axis and four locations along Y-axis. This process was repeated five times on each face, and the final SR obtained was the average of all the readings. The final SR shown in the graphs is average of all these readings.

Investigations for Dimensional Accuracy
This work explores the differences in nominal and actual heights values for various orientations, H 1 , H 2 , H 3, and H 4 . Nominal height represents the part height given by the design software, whereas actual height is the measured height of the fabricated part (refer to Table 3). Figure 4(a), (b), (c) and (d) shows how the height values change by changing orientations. Figure 4 (a) and (b) depicts the height H 1 and H 2 variation. The minimal divergence between the measured and actual values is discovered for orientation A. In contrast, the largest deviation is found for orientation D.
The fluctuation in height H 3 is shown in Fig. 4(c). The most negligible difference between measured and actual values is observed for orientation B. On the other side, ientation D has the most significant variation. Figure 4(d) illustrates the variability in height H 4 . The variation between measured and actual values is the smallest for orientation D. On the other hand, orientation B has the most diversity. When the mean average deviation is taken into account, the ideal orientation is A, whereas the least optimal orientation is D (see Table 14 for further information).
The variations in heights and standard deviation of heights are shown in Fig. 4 and Table 4. It can be established here that, in the case of height, the best orientation is A, B, and D, and the worst is C (refer to Fig. 4a, b, c, and d). It is suggested for fabricating the part, it should not be placed such that the part is resting on the short edge. It is related to the rate of heat   transmission. Heat transfer from the outermost layer to the base plate is not consistent from layer to layer. This is because fluids (the primary material and the supporting material (wax)) are incompressible. Because of incompressible fluids, they do not mix along their boundaries; voids emerge when the wax is removed from the manufactured component. Additionally, due to the high temperature during post-processing, the rate of heat transmission becomes non-uniform, resulting in oversizing and under-sizing parts (Ref 34). The 3D printed parts are postprocessed at 30-35°C in an oil bath. The voids are left behind as the support material is removed from the 3D printed item and dissolved in the oil bath during post-processing. Simultaneously, the primary material will attempt to fill the gaps, while the printed part will become undersized and distorted (Ref 34). Figure 5(a) illustrates the variability in length L 5 . The variation between measured and actual values is the smallest for orientation D. On the other hand, orientations B and C have the most diversity. Figure 5(b) illustrates the variability in length L 6 . The variation between measured and actual values is the smallest for orientation A. On the other hand, Orientation C has the most diversity. When the mean average deviation is taken into account, the ideal orientation is A, whereas the least optimal orientation is D (see Table 14 for details).
Tables 5 and 6 represent variation for various lengths and standard deviation. From Fig. 5, it was found that if the profile of the 3D printed part has maximum surfaces parallel to the base plate, the dimensional deviation will be less and the best orientation is A and B, and C will be the worst. On the other hand, if the inclined plane dimensions are more important, then the parts should be fabricated in such a manner that the inclined plane will remain parallel to the base plate. Figure 6(a) depicts the variation in width W 1 . The inequality between measured and actual values is the smallest for orientation D. Orientations A, B, and C, on the other hand, have the greatest variety. Figure 6(b) represents the variation in width W 4 . The disparity between measured and actual values is the smallest for orientation C. Orientations A, B, and D, on the other hand, have the greatest variety. Figure 6(c) shows the variation in width W 5 . The difference between measured and actual values is the smallest for orientation B. Orientation D, on the other hand, has the greatest variety.
The fluctuation in width W 7 is shown in Fig. 6(d). The divergence between measured and actual values is the lowest for orientation A. Orientation B, on the other hand, is the most diverse.
The variation in width W 10 is shown in Fig. 6(e). The divergence between measured and actual values is the lowest  for orientation A. Orientation D, on the other hand, is the most diverse. When all dimensions are considered, mean average deviations are taken into account, the optimum orientation is A, whereas the least optimal orientation is D (Table 14 details  further). For the width, the best orientation is A, and the worst orientation is B. The W 1 widths are undersized (refer to Table 7). For width, W 4 , W 5 , W 7 , and W 10 dimensions are sometimes oversized and undersized with different orientations (refer to Table 7), and the standard deviation of width is enlisted in Table 8.
The variation in radius is shown in Fig. 7, whereas Tables 9 and 10 contain the actual value, nominal value and standard deviation for the radius, respectively. It can be observed that, for all the orientations A, B, C, and D, as the semicircular part height increases from the base plate to the farthest layer, the dimensional deviation in radius is also increased due to nonuniform heat dissipation to the base plate. Hence, it can be concluded here that, in the case of the radius, orientation B is the best, and the worst is A and C. The actual value of the radius is found to be undersized for all the orientations. Figure 8(a) illustrates the variation in D 4 diameter. The discrepancy between measured and actual values is the smallest for orientation B. On the other side, orientation D is the most diversified.
The fluctuation in D 5 diameter is shown in Fig. 8(b). The disparity between measured and actual values is the lowest for orientation A. Orientation B, on the other hand, is the most diverse.
The variation in D 6 diameter is shown in Fig. 8(c). The disparity between measured and actual values is the lowest for orientation A. Orientation D, on the other hand, is the most diverse. A is the best orientation when all dimensions are considered, while D is the least optimal orientation (Table 14 details further).
The dimensional variation for various diameters (the comparison of actual values, nominal values, and standard deviation) is shown in Fig. 8, Tables 11, and 12, respectively. Here it can be observed that for the diameter of various circular parts, as the height from the base plate increases, the more dimensional deviation was observed. The reason for this is the non-uniform heat flow rate from different heights of geometries (Ref 34). In brief, it can be inferred here that for various diameters, the best orientation is A and the worst is D. Overall, it can be concluded that for both linear and radial dimensions, orientation A should be preferred to fabricate the 3D part on the MJP. It will be able to provide minimum dimensional variations. All sets of diameters are found to be undersized for all the orientations.

Investigation of Surface Roughness
The four figures for the eleven faces and four orientations are discussed in Fig. 9(a), (b), (c), and (d), and Table 13 reflects the measured values of SR. In the case of the orientations A, B, C and D, the value of SR is higher for the faces II, IV, and I. These are those faces (base) from where printing started. While removing the component from the base plates, it was heated from underneath. The heating of these surfaces resulted in an increase in the roughness on the bottom side of the part. In the case of orientation A, face II was at the base, for orientation B, face IV was the base, and for orientations C and D, faces I and II were at the bottom, respectively.
From a set of Fig. 9(a), (b), (c) and (d), it is found that as the layer-by-layer fabrication is moving away from the reference plane X-Y plane (base plate) in the Z-axis, the magnitude of SR   is increasing. The roughness of the surfaces is due to the overlapping of the build material and support material. During the 3D printing, printing jets inject both the primary and support materials to make the part dense. These materials are incompressible fluids, so they will not fuse during the overlapping (layering) mode. During the post-processing, the incompressible support material will come out thereby forming pits and voids. As the height of the part is increased, the roller and the plagiarizer will spread the material in an X-Y plane direction; more material will be collected along the walls of the  part. That will cause an increase in roughness along the walls of the parts fabricated. Due to high material collection at the ends, the rate of heat dissipation from the top layer to the base plate is decreased (due to the formation of voids) (Ref 34, 35). This may cause a lack of fusion of layers and an increase in roughness. This is observed clearly in Fig. 9(a), (b), (c), and (d) for orientation D. Moreover, for orientation D, the roughness R a decreases because the part is inclined at an angle, and the machine head had to travel less distance along the Z-axis.

SEM Analysis of Fabricated Components
The SEM was performed to investigate the cause of surface roughness in different orientations. In order to have a comparison of the SEM, the magnification power of 500 X is selected as a common platform. In the case of orientation A Fig. 10(a), the printer marks were observed to have tiny pits and they were found at a fixed distance in the X-Y plane. The movement of the printing head during the printing process is   the cause of these markings. Figure 10(b) shows that several fractures and inclusions with certain heights were found on the orientation B surfaces. This resulted in an increase in surface roughness. The cracks and inclusions were due to non-uniform heat transfer from the layers due to the incompressible primary material and support material (refer to Fig. 10c) for the orientation C; many voids, cracks, unfused layers, and clusters of the unfused material were observed. The reason behind this is the height of the part in the Z-Axis. Due to this, the combination of the uncompressible primary material and support material leads to the formation of air voids between the layers. These voids will interrupt the heat flow from the top to bottom layers. It will cause an increase in the roughness of the material and thermal contraction during post-processing. Considering Fig. 10(d) for orientation D, since the part is inclined at 45°, that contact area with the base plate is more as compared to orientation C. The rate of heat transfer from the layers will be more uniform with respect to orientation C. Due to this reason, the roughness of the part in orientation D is reduced with respect to C. In the current investigation, the traction height (TH) is considered as criterion to explain the cause of roughness.   During the comparison of images (Fig. 10a, b, c, and d), it is observed that the traction height (height of unfused layers) (TH) increases from orientation A to C (TH 1 < TH 2 < TH 3 ) and after that decreased with respect to C (TH 1 < TH 2 < TH 3 > TH 4 ). Hence, SR (R a ) is less in orientation A (closely packed) as compared to orientation C. The SEM images given in Fig. 10(b) and (c) are reflecting a similar trend. In Fig. 10(a), it is observed that the cause of the roughness is the 3D print head marks of the printer along with the unfused material and unfused material cluster on the surface of the fabricated part.
On the other hand in Fig. 10(c) for orientation C, it is observed that the surface is rough and the lack of fusion of the successive layers was observed. It was because the height of the part increases from the base plate.
The orientation B has the minimum SR, and the images reflect the same. SR in orientation B is due to some inclusions  and increased TH (TH 1 < TH 2 ) when compared with orientation A. The distribution of these inclusions is not identical in all the directions, due to which the roughness varies. Figure 10(d) shows that the lack of fusion of layers is nearly the same in all directions. The unfused layers of material create the waviness that is the cause of relatively higher roughness compared to orientations A and B.

Cause of Dimensional Deviation and Surface Roughness in Different Orientations
The surface area of the part in orientations A and B has more surface area in contact with the base plate with respect to orientations C and D. Hence, rate of heat transfer will be more uniform in orientations A and B with respect to orientations C and D. Concerning the non-uniform heat flow from the top layer to the base plate leads to the formation of voids; these voids were initially occupied by the wax (support material). When the post-processing of the parts is done in the saturated oil heat baths (30-35°C) with a different set of random vibrations due to the low melting temperature of the wax, it will come out and the voids will be left behind. After postprocessing when the parts will solidify, these different sizes of voids provide a different rate of cooling. This becomes the cause of contraction in build material, refer to Fig. 11(b). These different contraction levels lead to the formation of rough surfaces that are in direct contact with air and become the cause of the shrinkage and dimensional deviation ( Ref 34,35 and 14). These dimensional deviations and SR can also be seen in SEM, refer to Fig. 10 (a to b).

Obtaining best orientation for dimensional deviation and SR
During the investigation in terms of dimensional deviation for parts in different orientations, it was found that the orientations A and B provide the better compaction of the material in comparison with orientations C and D. These  Table 14). Table 14 represents the mean dimensional deviation between the measured value and true value. It is minimum for orientation A and maximum for orientation D. Hence, orientation A should be preferred for fabrication of the complex geometries with a multi-jet 3D printer. Table 15 represents the mean surface roughness for all the faces considered during the investigation. It is minimum for orientations B and A and maximum for orientation C. Hence, orientations A and B should be preferred to fabricate the complex geometries with a multi-jet 3D printer.
When dimensional deviation and surface roughness are combined, the optimal orientation is A and B. C and D are the worst and should be avoided.

Conclusions
Experimental investigations were conducted to study the effect of different orientations on dimensional accuracy and SR(R a ) of the 3D printed parts fabricated using a multi-jet machine. From this investigation, the following conclusions were found.
• Orientation A provides less dimensional deviation than other part orientations on the base plate. So it is suggested to place the part on the base plate in order to have the maximum area in contact with the base plate.
• Orientations A and B will result in greater material compaction and reduced surface roughness than other part orientations. Consequently, it is advised that the part should be positioned on the base plate to maximize the contact area with the base plate. • According to SEM pictures, the orientation A and B fabricated parts show superior interlayer bonding to the others. As a result, it has less dimensional variation and superior surface qualities when compared to other orientations. From the investigation of SEM, it was found that due to non-uniform heat transfer from the far layers, the post-processing of part led to the formations of voids and cracks (due to thermal contraction) on the surface that decreased the surface properties. • While considering dimensional deviation and surface roughness together, part should be organized in orientations A and B rather than the orientations C and D of the printer's base plate surface.
In general, it is suggested that the parts for the MJP-based 3D printing should be placed along the maximum area in contact with the base plate. It will provide better heat dissipation from the hot layers to the base plate. This will lead to uniform fusion between the layers and causes less dimensional deviation and better surface properties. It is a matter of investigation that different orientations will directly affect other mechanical properties (tensile strength, compressive strength, etc.) too. The findings are useful for the aerospace, jewelry, and automotive industries, which demand high-dimensional precision and appropriate surface properties from additively formed parts.

Open Access
This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.