Application of Vacuum Techniques in Shell Moulds Produced by Additive Manufacturing

This research shows the feasibility of the additive manufacturing technique (AM), Binder Jetting (BJ), for the production of shell moulds, which are filled by vacuum suction in the field of aluminium parts production. In addition, this study compares the gravity pouring technique and highlights the advantages of using vacuum techniques in AM moulds. A numerical simulation was carried out to study the behaviour of the liquid metal inside the moulds and the cooling rate of parts was analysed. The results show that in the gravity-pouring mould, the velocity in the gate causes moderate turbulence with small waves. However, vacuum suction keeps the velocity constant by eliminating waves and the filling process is homogeneous. Regarding dimensional accuracy, the staircase effect on the surface of the 3D moulds was the most critical aspect. The vacuum provides very homogeneous values of roughness across the entire surface of the part. Similarly, 3D scanning of castings revealed more accurate dimensions thanks to the help of vacuum forces. Finally, the microstructure of the cross section of the moulded parts shows that the porosity decreases with the vacuum filled. In both cases, the origin of the pores corresponds to gas entrapment and shrinkage during the filling process, the binder vaporization and nucleation points creation, leading to pores by shrinkage, gas entrapment or a mixture of both. This is the first study that uses vacuum filling techniques in moulds created by BJ, demonstrating the feasibility and advantages of AM using vacuum techniques, as an alternative to traditional casting.


Introduction
Additive manufacturing technology (AM) is based on the manufacture of parts or products by adding layers of material. It was originally developed to produce rapid prototypes. Today, this technology has many applications in different industrial sectors [1].
One of the AM techniques that is receiving more attention lately is the binder jetting (BJ) technique [2]. This technique is based on joining powder by means of a binder [3]. The process starts by spreading a layer of powder on a platform using a roller. Then, micro-droplets of a liquid binder are selectively injected through a print-head. This creates a two-dimensional layer where the powders that receive the binder bond with each other. The platform then descends a distance equivalent to the thickness of the layer and the process is repeated according to the slicing software. Finally, a three-dimensional part called the "green part" is obtained, which is usually post-treated to improve the final properties [4,5].
The applications of this technique are present in diverse sectors from aeronautics [6] to biomedical [7][8][9]. However, much effort is being made to use the BJ technique in the foundry vacuum-assisted high-pressure injection (HPDC). In both studies, they concluded that as absolute pressure decreases, average porosity and pore sizes are reduced and tensile strength and elongation are significantly improved.
Vacuum suction is a technique commonly used in the industry to improve the properties of cast parts. The purpose of this research is to demonstrate the improvement of properties in moulds manufactured by AM, being the first known study in which the vacuum suction technique is applied to moulds manufactured by BJ. To demonstrate the improvements of the vacuum suction technique, two moulds were manufactured using the BJ technique and filled using the different techniques (gravity and vacuum assisted pouring). The casting parts were compared and it was possible to quantify the improvement that the vacuum offers to the casting parts and the feasibility of using vacuum processes in the BJ technique.

Mould Design
A part was designed specifically to compare the quality of both casting techniques (gravity and vacuum). The design has a high volume/area ratio that maximizes the defects due to shrinkage and internal porosity. An ellipsoidal central geometry was chosen with two cylindrical parts at both ends ( Figure 1a). On the other hand, the curved surfaces of the part facilitate the evaluation of the characteristic staircase effect of AM technology. The theoretical weight and volume of the Computer Aided Design (CAD) part is 0.6 kg and 223.4 cm 3 , respectively. The casting material was a cast alloy from the AlSi series. The chemical composition is shown in Table 1.
(gravity and vacuum assisted pouring). The casting parts were compared and it was possible to quantify the improvement that the vacuum offers to the casting parts and the feasibility of using vacuum processes in the BJ technique.

Mould Design
A part was designed specifically to compare the quality of both casting techniques (gravity and vacuum). The design has a high volume/area ratio that maximizes the defects due to shrinkage and internal porosity. An ellipsoidal central geometry was chosen with two cylindrical parts at both ends ( Figure 1a). On the other hand, the curved surfaces of the part facilitate the evaluation of the characteristic staircase effect of AM technology. The theoretical weight and volume of the Computer Aided Design (CAD) part is 0.6 kg and 223.4 cm 3 , respectively. The casting material was a cast alloy from the AlSi series. The chemical composition is shown in Table 1. The filling system for both moulds was designed and calculated considering the limitations and advantages of the BJ technique. The moulds are a 10 mm thickness shell, sufficient to withstand the pressure of the molten metal [18]. For the gravity-pouring mould, Figure 1b, the filling system was composed of a sprue with a decreasing circular section describing an arc large enough to prevent cavitation and turbulence during the filling process. This sprue was connected to two vertical gates [39] leading to the cavity of the casting part. Finally, an open top riser was placed to prevent shrinkage defects due to solidification. In addition, it allows a reduction in the metallostatic pressure during the filling process [34]. With this design, the total volume of the 3D Printing (3DP) mould for gravitypouring is 449 cm 3 , the metal yield is 49.75%, and the weight of the mould is 0.55 kg.   The filling system for both moulds was designed and calculated considering the limitations and advantages of the BJ technique. The moulds are a 10 mm thickness shell, sufficient to withstand the pressure of the molten metal [18]. For the gravity-pouring mould, Figure 1b, the filling system was composed of a sprue with a decreasing circular section describing an arc large enough to prevent cavitation and turbulence during the filling process. This sprue was connected to two vertical gates [39] leading to the cavity of the casting part. Finally, an open top riser was placed to prevent shrinkage defects due to solidification. In addition, it allows a reduction in the metallostatic pressure during the filling process [34]. With this design, the total volume of the 3D Printing (3DP) mould for gravity-pouring is 449 cm 3 , the metal yield is 49.75%, and the weight of the mould is 0.55 kg.
For the vacuum-assisted mould, Figure 1c, the filling system consists of a conical cup connected directly to the gate. The gate has a cylindrical geometry with a calculated diameter to control the maximum velocity at the entry of the cavity. In addition, the suction force avoids the turbulent flow during the filling process. Finally, a small overflow was placed at the bottom of the mould to prevent the damage due to the molten metal initial jet. With this design, the total volume of the vacuum-assisted mould is 249.8 cm 3 , the metal yield is 89.43% and the mould weight is 0.31 kg. A slot was designed in both moulds to place a thermocouple to measure the temperature gradient during the pouring and solidification process. Both moulds were calculated based on the Campbell design rules [40] and were subsequently optimized using the simulation software.

Simulation
AnyCasting software (AnyCasting CO., LTD, Seoul, Korea) was used to simulate and optimize the filling system for both pouring techniques.
The basic equations to simulate free surface liquid flow were: (i) the complete Navier-Stokes equation, (ii) the continuity equation and (iii) the Volume Fluid method (VoF). For the convergence of the simulation, the Successive Over-Relaxation (SOR) iteration method was used. In addition, to discretize and solve these equations, the AnyCasting software automatically generated a structured tetrahedral mesh of 10,500,000 cells. This number of cells allowed a good adaptation of the mesh to the geometry and it was a sufficient number to represent effectively the flow behaviour inside the moulds. The highest density of cells was accumulated in the places where major changes in the geometry occurred in order to capture in more detail the changes in the flow variables. The initial parameters shown in Table 2 were introduced as boundary conditions to obtain reliable simulation results. The Heat Transfer Coefficient (HTC) between the casting and the mould was chosen as a linear function based on the properties of the mould. These parameters were automatically taken from the database provided by the software. HTC were compared to other studies [41][42][43] and the results were very similar to the values offered by the software. In this way, it was possible to predict the behaviour of the liquid metal and the turbulence created inside the moulds during the filling process. The two optimized models are shown in Figure 2. For the vacuum-assisted mould, Figure 1c, the filling system consists of a conical cup connected directly to the gate. The gate has a cylindrical geometry with a calculated diameter to control the maximum velocity at the entry of the cavity. In addition, the suction force avoids the turbulent flow during the filling process. Finally, a small overflow was placed at the bottom of the mould to prevent the damage due to the molten metal initial jet. With this design, the total volume of the vacuumassisted mould is 249.8 cm 3 , the metal yield is 89.43% and the mould weight is 0.31 kg. A slot was designed in both moulds to place a thermocouple to measure the temperature gradient during the pouring and solidification process. Both moulds were calculated based on the Campbell design rules [40] and were subsequently optimized using the simulation software.

Simulation
AnyCasting software (AnyCasting CO., LTD, Seoul, Korea) was used to simulate and optimize the filling system for both pouring techniques.
The basic equations to simulate free surface liquid flow were: (i) the complete Navier-Stokes equation, (ii) the continuity equation and (iii) the Volume Fluid method (VoF). For the convergence of the simulation, the Successive Over-Relaxation (SOR) iteration method was used. In addition, to discretize and solve these equations, the AnyCasting software automatically generated a structured tetrahedral mesh of 10,500,000 cells. This number of cells allowed a good adaptation of the mesh to the geometry and it was a sufficient number to represent effectively the flow behaviour inside the moulds. The highest density of cells was accumulated in the places where major changes in the geometry occurred in order to capture in more detail the changes in the flow variables. The initial parameters shown in Table 2 were introduced as boundary conditions to obtain reliable simulation results. The Heat Transfer Coefficient (HTC) between the casting and the mould was chosen as a linear function based on the properties of the mould. These parameters were automatically taken from the database provided by the software. HTC were compared to other studies [41][42][43] and the results were very similar to the values offered by the software. In this way, it was possible to predict the behaviour of the liquid metal and the turbulence created inside the moulds during the filling process. The two optimized models are shown in Figure 2.

Binder Jetting Manufactured Moulds
A ProJet CJP 660 Pro (3DSystems, Rock Hill, SC, USA) machine was used to manufacture the moulds. The ceramic material used as core was Visijet PXL. The powder is made up of a mixture of calcium sulfate, dehydrate, hemihydrate and anhydrous phases. In addition, it has some impurities or additives such as aluminium, sodium, magnesium, potassium, silicon [33]. The distribution of powder particles is D90 = 70.12 µm and the bulk particle density is 0.912 g/cm 3 [44]. In addition, the binder used was PXL, a liquid composed of 99% water and 1% of 2-pyrrolidone (C 4 H 7 NO).
The moulds were placed in the same zone on the machine table, as shown in Figure 3. This zone was chosen because it is the one with the highest powder compaction, as indicated by the manufacturer [45,46]. In addition, the moulds were placed in a horizontal position and in the same direction, to optimize the manufacturing time. The manufacturing parameters used are shown in Table 3.
The moulds were placed in the same zone on the machine table, as shown in Figure 3. This zone was chosen because it is the one with the highest powder compaction, as indicated by the manufacturer [45,46]. In addition, the moulds were placed in a horizontal position and in the same direction, to optimize the manufacturing time. The manufacturing parameters used are shown in Table 3. Once the manufacturing process was completed, the moulds were extracted from the machine and the unbound powder was removed with a brush and pressurized air. Finally, the moulds received a 2-h heat treatment at 250 °C in a JP Selecta 2001255 (JP Selecta, Barcelona, Spain) furnace to remove the moisture and volatile elements. This process improved the permeability and mechanical strength of the moulds.

Pouring and Solidification
The casting process started with melting of the aluminium alloy in an AGATRONIC R induction furnace (Argenta, Kielce, Poland). The pouring temperature was 750 ± 5 °C. The holding time was 2 min. The temperature sensor that controls the furnace is housed inside the crucible, therefore the temperature that the sensor reads is the real temperature of the metal liquid. The casting had a small volume and the handling and pouring time was very short, therefore the temperature of the metal liquid during the pouring did not vary significantly. For the gravity-pouring mould it was necessary to rest the mould on a sand base to prevent movement during the casting. In the case of the vacuumassisted mould, it was placed in the vacuum chamber at the bottom of the furnace.   Once the manufacturing process was completed, the moulds were extracted from the machine and the unbound powder was removed with a brush and pressurized air. Finally, the moulds received a 2-h heat treatment at 250 • C in a JP Selecta 2001255 (JP Selecta, Barcelona, Spain) furnace to remove the moisture and volatile elements. This process improved the permeability and mechanical strength of the moulds.

Pouring and Solidification
The casting process started with melting of the aluminium alloy in an AGATRONIC R induction furnace (Argenta, Kielce, Poland). The pouring temperature was 750 ± 5 • C. The holding time was 2 min. The temperature sensor that controls the furnace is housed inside the crucible, therefore the temperature that the sensor reads is the real temperature of the metal liquid. The casting had a small volume and the handling and pouring time was very short, therefore the temperature of the metal liquid during the pouring did not vary significantly. For the gravity-pouring mould it was necessary to rest the mould on a sand base to prevent movement during the casting. In the case of the vacuum-assisted mould, it was placed in the vacuum chamber at the bottom of the furnace. The vacuum system consisted of a GRIÑO ROTAMIK VS 21 EW (GRIÑO ROTAMIK, Rubí, Spain). A vacuum pump with a flow rate of 24 m 3 /h and a maximum vacuum pressure of 0.5 mbar in absolute pressure was used. For the successful performance of the operation, 150 ± 1 mbar of vacuum in absolute scale were applied. The thermocouple K-type connected to EasyLog USB (LASCAR electronics, whiteparish, UK) was placed in the mould housing to measure its temperature during the pouring and solidification process. The process used is shown in Figure 4.
Metals 2020, 10, x FOR PEER REVIEW 7 of 20 was used. For the successful performance of the operation, 150 ± 1 mbar of vacuum in absolute scale were applied. The thermocouple K-type connected to EasyLog USB (LASCAR electronics, whiteparish, UK) was placed in the mould housing to measure its temperature during the pouring and solidification process. The process used is shown in Figure 4. Once the moulds' temperature reached 50 ± 5 °C, the parts were extracted to start the analysis stage. Figure 5 shows the castings ( Figure 5a) and final parts ( Figure 5b). Three features were selected to compare the castings: surface quality, dimensional quality and porosity.

Surface Roughness Analysis
The surface quality of the parts was obtained using a Mitutoyo Surftest SJ-500 profilometer (Mitutoyo, Kawasaki-shi, Japan). The cutting length was 2.5 mm and the evaluation length 12.5 mm, according to ISO 4288: 1996 [47]. The measurements were taken in three areas of the part, designated as Zone I, II, and III. In Zones I and III, the measurements were made in two directions: (i) in the  Once the moulds' temperature reached 50 ± 5 • C, the parts were extracted to start the analysis stage. Figure 5 shows the castings ( Figure 5a) and final parts ( Figure 5b). Three features were selected to compare the castings: surface quality, dimensional quality and porosity.
Metals 2020, 10, x FOR PEER REVIEW 7 of 20 was used. For the successful performance of the operation, 150 ± 1 mbar of vacuum in absolute scale were applied. The thermocouple K-type connected to EasyLog USB (LASCAR electronics, whiteparish, UK) was placed in the mould housing to measure its temperature during the pouring and solidification process. The process used is shown in Figure 4. Once the moulds' temperature reached 50 ± 5 °C, the parts were extracted to start the analysis stage. Figure 5 shows the castings ( Figure 5a) and final parts ( Figure 5b). Three features were selected to compare the castings: surface quality, dimensional quality and porosity.

Surface Roughness Analysis
The surface quality of the parts was obtained using a Mitutoyo Surftest SJ-500 profilometer (Mitutoyo, Kawasaki-shi, Japan). The cutting length was 2.5 mm and the evaluation length 12.5 mm, according to ISO 4288: 1996 [47]. The measurements were taken in three areas of the part, designated as Zone I, II, and III. In Zones I and III, the measurements were made in two directions: (i) in the

Surface Roughness Analysis
The surface quality of the parts was obtained using a Mitutoyo Surftest SJ-500 profilometer (Mitutoyo, Kawasaki-shi, Japan). The cutting length was 2.5 mm and the evaluation length 12.5 mm, according to ISO 4288: 1996 [47]. The measurements were taken in three areas of the part, designated as Zone I, II, and III. In Zones I and III, the measurements were made in two directions: (i) in the direction of the print-head (X direction) and (ii) in the direction of the movement of the re-coater Metals 2020, 10, 1090 8 of 20 (Y direction). Zones and directions are shown in Figure 6. The measurements made in the Y direction registered surface irregularities due to the staircase effect characteristic of the 3DP technique. In Zone II, only measurements in the Y direction were made.
Metals 2020, 10, x FOR PEER REVIEW 8 of 20 direction of the print-head (X direction) and (ii) in the direction of the movement of the re-coater (Y direction). Zones and directions are shown in Figure 6. The measurements made in the Y direction registered surface irregularities due to the staircase effect characteristic of the 3DP technique. In Zone II, only measurements in the Y direction were made.

Quality Dimensional Analysis
A Breuckmann Smart SCAN3D-HE (Breuckmann, Heiligenhaus, Germany) structured light scanner with a working field of 125 mm and an accuracy of 5 µ m was used to analyse the geometrical and dimensional quality of the moulds and parts. The obtained point-clouds were processed by Geomagic Control X software (3D Systems, Rock Hill, SC, USA) for comparing to nominal CAD by Best Fit alignment. The average number of points in each scan was 45,000 points. The average and standard deviations of the total number of points from the CAD was obtained. The tolerance range used was 0.05 mm, divided into upper and lower deviation and all calculated along the maximum deviation, set as 1 to −1 mm. In addition, to observe the differences between the different states, the average deviation of the upper and lower deviation was compared with the percentage of total points in each type of deviation.
The analysis of the dimensional quality was divided into three stages: (i) the moulds were scanned as built. This allowed the analysis of errors induced by the machine and the manufacturing process itself (bleeding effect); (ii) the moulds were scanned after the heat treatment. In this way, the error produced by the elimination of volatiles and the microstructural change suffered during the process was quantified, and (iii) the cast parts were scanned to observe the errors introduced by the casting process (shrinkage). Therefore, the dimensional deviation was the sum of the errors produced by the mould manufacture, the heat treatment and the casting process.

Porosity Analysis
The volumetric porosity was evaluated on the castings. The control Zones (I, II and III) were sectioned and the hydrostatic weighing method was used [38]. A CB-Complet electronic balance (Cobos Precisión, Barcelona, Spain) with an accuracy of 0.001 g was used to weight the parts.
To analyse the internal porosity and obtain a high resolution of the pores, specimens were extracted in each part at the central and external area of each Zone (I, II and III). The specimens were polished according to standard [48]. A total of 50 images were taken from each specimen, with a Tecmicro metallographic microscope (OLYMPUS BHM, Barcelona, Spain). Finally, the images were evaluated with ImageJ 2016-0205 software (National Institutes of Health, Bethesda, MD, USA). This tool allowed the brightness and contrast of the images to be stabilized, isolating the contour of the holes by applying a threshold filter. In this way, the number of pores, the maximum and minimum Feret diameter (the longest and shortest distance between two points along the selection boundary, respectively) and the roundness of the pores in each region were obtained.

Zone (III) Zone (I)
Zone (II) Figure 6. Zones and directions for measuring the roughness of castings.

Quality Dimensional Analysis
A Breuckmann Smart SCAN3D-HE (Breuckmann, Heiligenhaus, Germany) structured light scanner with a working field of 125 mm and an accuracy of 5 µm was used to analyse the geometrical and dimensional quality of the moulds and parts. The obtained point-clouds were processed by Geomagic Control X software (3D Systems, Rock Hill, SC, USA) for comparing to nominal CAD by Best Fit alignment. The average number of points in each scan was 45,000 points. The average and standard deviations of the total number of points from the CAD was obtained. The tolerance range used was 0.05 mm, divided into upper and lower deviation and all calculated along the maximum deviation, set as 1 to −1 mm. In addition, to observe the differences between the different states, the average deviation of the upper and lower deviation was compared with the percentage of total points in each type of deviation.
The analysis of the dimensional quality was divided into three stages: (i) the moulds were scanned as built. This allowed the analysis of errors induced by the machine and the manufacturing process itself (bleeding effect); (ii) the moulds were scanned after the heat treatment. In this way, the error produced by the elimination of volatiles and the microstructural change suffered during the process was quantified, and (iii) the cast parts were scanned to observe the errors introduced by the casting process (shrinkage). Therefore, the dimensional deviation was the sum of the errors produced by the mould manufacture, the heat treatment and the casting process.

Porosity Analysis
The volumetric porosity was evaluated on the castings. The control Zones (I, II and III) were sectioned and the hydrostatic weighing method was used [38]. A CB-Complet electronic balance (Cobos Precisión, Barcelona, Spain) with an accuracy of 0.001 g was used to weight the parts.
To analyse the internal porosity and obtain a high resolution of the pores, specimens were extracted in each part at the central and external area of each Zone (I, II and III). The specimens were polished according to standard [48]. A total of 50 images were taken from each specimen, with a Tecmicro metallographic microscope (OLYMPUS BHM, Barcelona, Spain). Finally, the images were evaluated with ImageJ 2016-0205 software (National Institutes of Health, Bethesda, MD, USA). This tool allowed the brightness and contrast of the images to be stabilized, isolating the contour of the holes by applying a threshold filter. In this way, the number of pores, the maximum and minimum Feret diameter (the longest and shortest distance between two points along the selection boundary, respectively) and the roundness of the pores in each region were obtained.

Results of Simulation
The results of the simulation made it possible to know the velocity profile, to avoid a high turbulent flow of the liquid metal during the filling process. Campbell [40] defined the maximum filling speed as 0.5 m/s; increasing this speed value can lead to gas entrainment and oxide films. The colour map in Figure 7 shows the velocity profile of both moulds during the filling process. The vacuum-assisted technique has a more controlled filling, avoiding the possible defects of gas entrapment, inclusions or oxides.

Results of Simulation
The results of the simulation made it possible to know the velocity profile, to avoid a high turbulent flow of the liquid metal during the filling process. Campbell [40] defined the maximum filling speed as 0.5 m/s; increasing this speed value can lead to gas entrainment and oxide films. The colour map in Figure 7 shows the velocity profile of both moulds during the filling process. The vacuum-assisted technique has a more controlled filling, avoiding the possible defects of gas entrapment, inclusions or oxides. The flow behaviour of the liquid metal inside the gravity-pouring mould was progressive (Figure 7a), with moderate turbulence and a relatively homogeneous velocity profile. Although the filling velocity remained at values very close to 0.5 m/s at some points, this velocity was exceeded, creating waves in the early stages of the filling process. This phenomenon was due to small differences in the velocities at the gates. These differences were in turn imposed by the metallostatic pressure of the sprue.
The waves did not cease until the sprue was saturated with liquid metal. From that moment on, the metallostatic pressure acted uniformly on each gate. The waves disappeared to give way to a progressive filling as shown in Figure 7a. In addition, the velocity in the gates area remained stable throughout the filling process, providing moderate turbulence that minimized the risk of gas entrainment. Finally, the filling time was 5.2 s.
The behaviour of the liquid flow for the vacuum-assisted mould was more stable and homogeneous (Figure 7b). In the early stages of the filling process, the liquid entered the mould cavity at a high velocity; this was due to the vertical pouring configuration of the mould. The initial jet impact generated a turbulence, which was absorbed by the overflow. A slight wave effect was produced, but this effect was mitigated by the suction forces applied by the vacuum. Subsequently, a uniform filling process was obtained. The filling time in this case was 1.9 s.
The vacuum-assisted technique showed a more controlled filling, avoiding the possible defects of gas entrapment, inclusions or oxides. The flow behaviour of the liquid metal inside the gravity-pouring mould was progressive (Figure 7a), with moderate turbulence and a relatively homogeneous velocity profile. Although the filling velocity remained at values very close to 0.5 m/s at some points, this velocity was exceeded, creating waves in the early stages of the filling process. This phenomenon was due to small differences in the velocities at the gates. These differences were in turn imposed by the metallostatic pressure of the sprue.
The waves did not cease until the sprue was saturated with liquid metal. From that moment on, the metallostatic pressure acted uniformly on each gate. The waves disappeared to give way to a progressive filling as shown in Figure 7a. In addition, the velocity in the gates area remained stable throughout the filling process, providing moderate turbulence that minimized the risk of gas entrainment. Finally, the filling time was 5.2 s.
The behaviour of the liquid flow for the vacuum-assisted mould was more stable and homogeneous (Figure 7b). In the early stages of the filling process, the liquid entered the mould cavity at a high velocity; this was due to the vertical pouring configuration of the mould. The initial jet impact generated a turbulence, which was absorbed by the overflow. A slight wave effect was produced, but this effect was mitigated by the suction forces applied by the vacuum. Subsequently, a uniform filling process was obtained. The filling time in this case was 1.9 s.
The vacuum-assisted technique showed a more controlled filling, avoiding the possible defects of gas entrapment, inclusions or oxides.
The velocity profile of the gate cross-section during the filling process was obtained using the simulation software. Figure 8 shows the different velocities at the gates during the filling process. This velocity was divided in two stages: (i) up to 15% of filling (the metallic liquid was not stabilized because the gate has not been saturated) and (ii) up to 100% of moulds filling (the metal liquid was stabilized at the gate).
Stage (ii) represents the filling process when the metal liquid was stabilized at the gate. As shown in Figure 8, the gravity-pouring mould showed a filling with fluctuations over an average value of 0.48 m/s. The initial source effect dragged these fluctuations during the filling process. In addition, a damped harmonic state was observed in which the flow-generated waves decreased in amplitude as the filling process was completed. This is consistent with Figure 7a, where waves generated by the source effect tended to decrease and the flow became more stable. High runner pressures and large areas in the cavity result in higher gate velocities [49,50]. Even so, only in specific cases did the average velocity exceed the critical value of 0.5 m/s. Therefore, there was low risk of gas entrapment.
For the stage (ii) in the vacuum-assisted technique, a constant gate velocity of 0.43 m/s was observed. The filling velocity only depends on the gravity force, the vacuum force and the gate area. These three parameters were constant during the process and therefore the velocity was constant in this second filling stage.   For the gravity-pouring mould, in stage (i), a strong increase in velocity was observed, which descended rapidly, with the highest value being 0.56 m/s. The velocity fluctuations generated the waves previously described.

Result of Solidification
In contrast, the vacuum-assisted technique presented two phases within the first stage. In the first phase, a sudden increase in velocity was observed. The liquid entrance was placed at the top of the mould; therefore, the gravity force acted in favour of the filling. In the second phase, the filling cup was saturated with liquid metal and the velocity gradually increased to the optimum filling velocity.
Stage (ii) represents the filling process when the metal liquid was stabilized at the gate. As shown in Figure 8, the gravity-pouring mould showed a filling with fluctuations over an average value of 0.48 m/s. The initial source effect dragged these fluctuations during the filling process. In addition, a damped harmonic state was observed in which the flow-generated waves decreased in amplitude as the filling process was completed. This is consistent with Figure 7a, where waves generated by the source effect tended to decrease and the flow became more stable. High runner pressures and large areas in the cavity result in higher gate velocities [49,50]. Even so, only in specific cases did the average velocity exceed the critical value of 0.5 m/s. Therefore, there was low risk of gas entrapment.
For the stage (ii) in the vacuum-assisted technique, a constant gate velocity of 0.43 m/s was observed. The filling velocity only depends on the gravity force, the vacuum force and the gate area. These three parameters were constant during the process and therefore the velocity was constant in this second filling stage. Figure 9 shows the temperature distribution measured in both moulds and the exponential regression of the cooling rates. For the gravity-pouring mould, the temperature increased to a maximum of 389 ± 5 • C. The curve of the temperature had a large amplitude, due to the volume/area ratio of the part. The thermal energy had to be evacuated through the walls of the mould. When the mould heated up, the temperature difference between the casting surface and the mould became smaller and smaller, so the thermal evacuation decreased. The average cooling rate was −2.56 • C/s. The exponential regression shows a R 2 = 0.96 and an exponent of 0.00023. process; once the part is solidified, the vacuum system does not apply. The cooling rate was −9.08 °C/s. The exponential regression shows a R 2 = 0.95 and an exponent of 0.0003.

Result of Solidification
As can be seen in Figure 9, the exponent of the equation was higher in the vacuum-assisted mould, therefore the vacuum-assisted cooling was quicker than for the gravity-pouring mould. The higher thermal dissipation was due to the vacuum suction, as it removed the hot gases that were generated during the casting process, evacuating a higher heat flow. In addition, the proximity of the hot runner in the gravity-pouring casting could prevent the heat evacuation. Therefore, the turning point on the curve and the subsequent cooling process in the vacuum-assisted mould occurred earlier.
The thermal behaviour of both moulds was quite similar, as both moulds were built with the same 3DP technique, the same material and the same wall thickness. In addition, it can be observed that the total time cycle was very similar for both moulds, concluding that the application of vacuum did not excessively affect the solidification process. Despite that, the vacuum considerably improved the homogeneity in the filling, extracting most of the gases formed during the casting process, which is advantageous for reducing porosity.

Roughness
The principal parameters that influence the surface roughness of castings are (i) the mould material: type, morphology, size and grain distribution, (ii) the mould manufacturing process: compaction and layer height, (iii) the pouring of the aluminium alloy: pouring temperature, (iv) the vacuum applied: degree of vacuum imposed. On the other hand, the temperature of the vacuum-assisted mould reached 374 ± 5 • C. As can be seen in Figure 9, the cooling distribution can be divided in two phases, before and after 3100 s. After this point, the vacuum system was turned off. The vacuum only has an effect during the solidification process; once the part is solidified, the vacuum system does not apply. The cooling rate was −9.08 • C/s. The exponential regression shows a R 2 = 0.95 and an exponent of 0.0003.
As can be seen in Figure 9, the exponent of the equation was higher in the vacuum-assisted mould, therefore the vacuum-assisted cooling was quicker than for the gravity-pouring mould. The higher thermal dissipation was due to the vacuum suction, as it removed the hot gases that were generated during the casting process, evacuating a higher heat flow. In addition, the proximity of the hot runner in the gravity-pouring casting could prevent the heat evacuation. Therefore, the turning point on the curve and the subsequent cooling process in the vacuum-assisted mould occurred earlier.
The thermal behaviour of both moulds was quite similar, as both moulds were built with the same 3DP technique, the same material and the same wall thickness. In addition, it can be observed that the total time cycle was very similar for both moulds, concluding that the application of vacuum did not excessively affect the solidification process. Despite that, the vacuum considerably improved the homogeneity in the filling, extracting most of the gases formed during the casting process, which is advantageous for reducing porosity.

Roughness
The principal parameters that influence the surface roughness of castings are (i) the mould material: type, morphology, size and grain distribution, (ii) the mould manufacturing process: compaction and layer height, (iii) the pouring of the aluminium alloy: pouring temperature, (iv) the vacuum applied: degree of vacuum imposed.
All parameters were kept constant in this research, except the degree of vacuum applied during filling (no vacuum for the gravity-pouring mould and 150 ± 1 mbar in absolute pressure for the vacuum-assisted mould).
As shown in Figure 10a, the roughness in the gravity-pouring part showed a considerable dispersion of values in all zones and directions. The roughness in Zone I in both directions was a little more stable than in the rest. This was due to the higher metallostatic pressure in the lower area of the part (Zone I), favouring a more accurate reproduction of the mould surface. For the upper zones of the part, the metallostatic pressure was lower and therefore the values of the measurements were more dispersed. part (Zone I), favouring a more accurate reproduction of the mould surface. For the upper zones of the part, the metallostatic pressure was lower and therefore the values of the measurements were more dispersed.
The roughness values for the vacuum-assisted part, Figure 10b, were a little higher than for the gravity-pouring, but significantly more uniform in all zones and directions. This was because the vacuum pressure created a suction effect that attracted the molten metal against the walls of the mould, helping the liquid metal to penetrate more into the capillaries formed between the calcium sulphate particles.
The gravity-pouring part had an average value of Ra = 3.52 µ m in the direction of the printinghead movement (X direction) and Ra = 7.02 µ m in the re-coater direction (Y direction). The average roughness values of the vacuum-assisted part were Ra = 4.63 µ m in the X direction and Ra = 7.76 µ m in the Y direction.
The overall roughness values could be reduced if the machine could compact the powder during the manufacturing process [51]. The results obtained in the normal direction to the layer construction (re-coater direction) can be reduced by using a lower layer height. Creating a thinner layer would reduce the roughness of the mould, but increase the printing time. Another option could be to apply an infiltration post-process to the moulds to increase the surface quality. However, the values obtained are tolerable compared to traditional sand casting (10 µ m < Ra < 25 µ m) [52].

Dimensional Quality
One of the most critical parameters for metal casting is dimensional accuracy. The moulds showed dimensional and geometrical deviations due to two factors. Firstly, the bleeding effect, caused by capillary action, which attracts the liquid binder to the adjacent calcium sulphate particles. This effect caused deformations in each two-dimensional layer, affecting the final dimensions of the moulds [28]. Secondly, the heat treatment applied to the moulds caused a variation in the bonding bridges among the particles, resulting in volumetric shrinkage in the mould cavity.
As shown in Figure 11, step 1 reflects the comparison between the as-built mould cavity and the nominal CAD. The gravity-pouring mould showed a low standard deviation (0.087 mm). The The roughness values for the vacuum-assisted part, Figure 10b, were a little higher than for the gravity-pouring, but significantly more uniform in all zones and directions. This was because the vacuum pressure created a suction effect that attracted the molten metal against the walls of the mould, helping the liquid metal to penetrate more into the capillaries formed between the calcium sulphate particles.
The gravity-pouring part had an average value of Ra = 3.52 µm in the direction of the printing-head movement (X direction) and Ra = 7.02 µm in the re-coater direction (Y direction). The average roughness values of the vacuum-assisted part were Ra = 4.63 µm in the X direction and Ra = 7.76 µm in the Y direction.
The overall roughness values could be reduced if the machine could compact the powder during the manufacturing process [51]. The results obtained in the normal direction to the layer construction (re-coater direction) can be reduced by using a lower layer height. Creating a thinner layer would reduce the roughness of the mould, but increase the printing time. Another option could be to apply an infiltration post-process to the moulds to increase the surface quality. However, the values obtained are tolerable compared to traditional sand casting (10 µm < Ra < 25 µm) [52].

Dimensional Quality
One of the most critical parameters for metal casting is dimensional accuracy. The moulds showed dimensional and geometrical deviations due to two factors. Firstly, the bleeding effect, caused by capillary action, which attracts the liquid binder to the adjacent calcium sulphate particles. This effect caused deformations in each two-dimensional layer, affecting the final dimensions of the moulds [28]. Secondly, the heat treatment applied to the moulds caused a variation in the bonding bridges among the particles, resulting in volumetric shrinkage in the mould cavity.
As shown in Figure 11, step 1 reflects the comparison between the as-built mould cavity and the nominal CAD. The gravity-pouring mould showed a low standard deviation (0.087 mm). The measured values indicated results very close to the nominal (0 value). For the upper points (55.52%) the average value did not exceed 0.07 mm. The lower points (44.48%) also had an average value of −0.07 mm. On the other hand, vacuum-assisted mould showed an even smaller standard deviation (0.054 mm). Moreover, the average values are even closer to the nominal values, 0.04 mm for the upper (46%) and lower (54%) points, respectively. Therefore, both moulds have a similar accuracy, with a standard deviation around ±0.05 mm, which indicates that the AM machine has a good geometrical accuracy and high repeatability during the manufacturing process.

65.31%.
The heat treatment for both moulds increase in the standard deviation (Table 3) and average values for both moulds indicates a higher dispersion in the points and, therefore, an increase in the geometric error before the pouring process. The volume reduction was also reflected through the increase in the percentage of positive points. This meant that the castings had smaller dimensions than the nominal CAD. Therefore, the heat treatment added a geometric error and caused a reduction in the volume of the cavity in both moulds. Finally, step 3 shows the comparison of the part and the nominal CAD. For the casting part, the positive values represent the negative values of the mould. Therefore, in the gravity-pouring part, the positive points decreased to 28.67% with an average value of 0.17 mm. Consequently, 71.83% of the points are in negative values, with an average of −0.37 mm. In addition, the standard deviation for the casting part achieved the highest value in all of the complete process, with a result of 0.35 mm. However, the vacuum-assisted part presented less geometric defects than the gravity-pouring alternative. Figure 11. Scanning the three steps of the casting process for both techniques. Table 4, the standard deviation was 0.25 mm and 92.69% of the points were at 0.44 mm.

As shown in
The vacuum-assisted was able to improve the geometric quality of the final part, since the error was similar to the sum of the aluminium shrinkage (linear thermal shrinkage for this aluminium alloy is 1.2%, which for this geometry is −0.3 mm) plus the geometric error of the heat-treated mould.
Step 1: Scanning result of the moulds as built

Gravity Vacuum
Step 2: Scanning result of the moulds with heat treatment  Figure 11. Scanning the three steps of the casting process for both techniques.
Step 2 shows the comparison between the mould cavity after heat treatment and the nominal CAD. The average value for the gravity-pouring mould became 0.13 mm and −0.09 mm in the upper and lower deviation, respectively ( Table 4). The number of positive points also increased from 44.48% to 62.95%. In the case of the vacuum-assisted mould, the values in the upper and lower deviation were 0.13 and −0.10, respectively. In addition, the number of positive points was redistributed to 65.31%. The heat treatment for both moulds increase in the standard deviation (Table 3) and average values for both moulds indicates a higher dispersion in the points and, therefore, an increase in the geometric error before the pouring process. The volume reduction was also reflected through the increase in the percentage of positive points. This meant that the castings had smaller dimensions than the nominal CAD. Therefore, the heat treatment added a geometric error and caused a reduction in the volume of the cavity in both moulds. Finally, step 3 shows the comparison of the part and the nominal CAD. For the casting part, the positive values represent the negative values of the mould. Therefore, in the gravity-pouring part, the positive points decreased to 28.67% with an average value of 0.17 mm. Consequently, 71.83% of the points are in negative values, with an average of −0.37 mm. In addition, the standard deviation for the casting part achieved the highest value in all of the complete process, with a result of 0.35 mm. However, the vacuum-assisted part presented less geometric defects than the gravity-pouring alternative.
As shown in Table 4, the standard deviation was 0.25 mm and 92.69% of the points were at 0.44 mm.
The vacuum-assisted was able to improve the geometric quality of the final part, since the error was similar to the sum of the aluminium shrinkage (linear thermal shrinkage for this aluminium alloy is 1.2%, which for this geometry is −0.3 mm) plus the geometric error of the heat-treated mould.

Porosity
Porosity of castings directly affects the mechanical properties of the castings. Large pores reduce the loading area and more easily induce crack propagation.
To evaluate the volumetric porosity of the parts, it was necessary to calculate the bulk density of the parts using the following expression [38]: where: ρ p and ρ w are the specimen density and water density, respectively; m 1 and m 2 are the mass of the specimen in air and water, respectively. Finally, the following expression was used to calculate the volumetric porosity: where: ρ r is the actual density of aluminium ENAB46000 (2760 kg/m 3 ) according to EN1706 [53]. Table 5 shows the volumetric porosity results for the different part zones in both techniques. Volumetric porosity increased with the zone. This was due to the vertical filling in both moulds. The density of the bubbles entrapped was lower than the metallic liquid, so the bubbles ascended and accumulated in the upper part of the casting. The vacuum pressure reduced the porosity of the parts in each zone, resulting in an average porosity 2.212% lower than in the gravity-pouring part.
To analyse porosity in depth, a study of the number, size and shape of the pores was carried out. Figure 12 shows the statistical distribution of the pores according to their size for each zone and location. Pore size was evaluated using Feret s maximum diameter.
The following can be deduced: (i) the pore size is in general between 10 and 14 µm for the gravity-pouring casting and between 0 and 14 µm for the vacuum-assisted casting. Therefore, the vacuum pressure decreases the pore size. (ii) The number of pores diminished in all areas with the vacuum-assisted casting. (iii) For both parts, the number of pores increased in Zones II and III, as explained in the previous paragraph. On the other hand, Figure 12 reveals that the number of pores increased on the outside of the parts regardless of the zone. The reason is that the gas entrapped by the metal liquid could not be evacuated before solidification. Preheating of the mould is recommended to avoid this phenomenon.
where: C is the circularity of the pore; a is the area occupied by the pore and p is the perimeter of the pore.
Circularity in the vacuum-assisted part increased in all zones and locations. Therefore, it can be stated that when the size of the pores decreases, the circularity increases. In contrast, the larger pores had more irregular shapes, due to the defects associated with contraction by solidification, among others. The circularity of these particles was very low. Finally, the trend in morphology is very similar in all the analysed samples. For the gravity-pouring part, the circularity is grouped around 0.3; this indicates sharper and more irregular pores. However, for the vacuum-assisted part, the circularity tends to be higher (0.6), indicating more rounded pores.  Figure 13 shows the pores' morphology. Circularity of pores was calculated according to the following expression: where: C is the circularity of the pore; a is the area occupied by the pore and p is the perimeter of the pore. Circularity in the vacuum-assisted part increased in all zones and locations. Therefore, it can be stated that when the size of the pores decreases, the circularity increases. In contrast, the larger pores had more irregular shapes, due to the defects associated with contraction by solidification, among others. The circularity of these particles was very low. Finally, the trend in morphology is very similar in all the analysed samples. For the gravity-pouring part, the circularity is grouped around 0.3; this indicates sharper and more irregular pores. However, for the vacuum-assisted part, the circularity tends to be higher (0.6), indicating more rounded pores.  Figure 14 shows the morphology of pores in every zone in the parts: (i) Pores with sharp corners associated with shrinkage ( Figure 14a); (ii) Oval-shaped pores with less sharp corners, resulting from a combination of shrinkage and gas entrapment ( Figure 14b) and (iii) rounded-shaped pores associated with gas entrapment (Figure 14c). These three types of pores appeared in all specimens [24,34,38].   Figure 14 shows the morphology of pores in every zone in the parts: (i) Pores with sharp corners associated with shrinkage ( Figure 14a); (ii) Oval-shaped pores with less sharp corners, resulting from a combination of shrinkage and gas entrapment ( Figure 14b) and (iii) rounded-shaped pores associated with gas entrapment (Figure 14c). These three types of pores appeared in all specimens [24,34,38].  Figure 14 shows the morphology of pores in every zone in the parts: (i) Pores with sharp corners associated with shrinkage ( Figure 14a); (ii) Oval-shaped pores with less sharp corners, resulting from a combination of shrinkage and gas entrapment ( Figure 14b) and (iii) rounded-shaped pores associated with gas entrapment (Figure 14c). These three types of pores appeared in all specimens [24,34,38].  Table 6 shows the summary of the porosity results for each technique, zone and location of specimens. The application of vacuum results in a significant reduction in the average pore size (47.3%). Meanwhile, the maximum diameter, the minimum diameter and the number of pores also decrease with the application of vacuum.