*Article* **E**ff**ects of Eutectic Modification and Grain Refinement on Microstructure and Properties of PM AlSi7 Metallic Foams**

#### **Dirk Lehmhus 1,\*, Daniela Hünert 2, Ulrike Mosler 3, Ulrich Martin <sup>4</sup> and Jörg Weise <sup>1</sup>**


Received: 14 October 2019; Accepted: 13 November 2019; Published: 20 November 2019

**Abstract:** For AlSi7 foams, microstructure modification by variation of solidification rates and addition of Sr, B and TiB2/TiAl3 was investigated and its transfer to powder metallurgical metal foaming processes demonstrated. Microstructural characterization focused on grain size and morphology of the eutectic phase. Cooling rates during solidification were linked to secondary dendrite arm spacing, establishing a microstructure-based measure of solidification rates. Effects of refining and modification treatments were compared and their influence on foam expansion evaluated. Studies on foams focused on comparison of micro- and pore structure using metallographic techniques as well as computed tomography in combination with image analysis. Reference samples without additives and untreated as well as annealed TiH2 as foaming agent allowed evaluation of pore and microstructure impact on mechanical performance. Evaluation of expansion and pore structure revealed detrimental effects of Sr and B additions, limiting the evaluation of mechanical performance to the TiB2 samples. These, as well as the two reference series samples, were subjected to quasi-static compression testing. Stress-strain curves were gained and density-dependent expressions of ultimate compressive strength, plateau strength and tangent modulus derived. Weibull evaluation of density-normalized mechanical properties revealed a significant influence of grain size on the Weibull modulus at densities below 0.4 g/cm3.

**Keywords:** aluminum foam; metal foam; aluminum alloys; grain refinement; modification; microstructure; mechanics of materials; metallurgy; melt treatment; powder metallurgy

#### **1. Introduction**

#### *1.1. Foam Fundamentals*

Metallic foams based on the powder compact melting or Fraunhofer process have by now reached sufficient levels of maturity to allow series production, e.g., for applications in the transport, machine tool and even the building industry [1–4]. The underlying process is based on hot compaction of a mixture of matrix metal, usually aluminum or an aluminum alloy, and foaming agent powders, usually TiH2. The resulting precursor material is then heated above its melting point and expands to yield a liquid foam, which is stabilized by solidification. The method was patented in 1990 by Baumeister et al. and has since been described in several publications [5,6]. A recent overview contrasting this

process in terms of foam structure, performance and cost with alternative approaches has recently been published by Lehmhus et al. [7].

As a natural consequence of the material's success, the call for statistically well founded design criteria as a prerequisite for making best use of the material's capabilities gains urgency. One major step to this end is an improved understanding of the role of different structural features in determining the mechanical properties, as well as their scatter. Both, however, are influenced by several characteristics, among which Mosler et al. and Martin et al. suggest the following hierarchy [8,9]:


Mu et al. introduce friction between cell walls as a further mechanism influencing behavior under compressive load, which, however, becomes effective only at high strain levels beyond the stress-strain curve's typical plateau [10]. Density as the dominating aspect may be subdivided into global density and systematic density variations such as density gradients as typically induced by foam drainage, or by solidification shrinkage. Mechanical testing parallel to such density gradients will cause the lowest density cross sections to fail first, leading to a lowered yield point and a steeper plateau region than observed in samples of matching average density either tested perpendicular to any density gradient, or showing a homogeneous overall structure [11].

To shed additional light on the relative importance of the above features in determining mechanical performance, the present study concentrates on microstructure variation achieved via modification and grain refinement of the foam matrix alloy, AlSi7. Part of the methodology is to eliminate as far as possible the effects of the global density (a) by normalizing the results of mechanical testing. This was done according to a Gibson-Ashby type formulation of strength and stiffness as a function of global density [12]. Foam structural characteristics (b) were documented for all samples subjected to mechanical testing based on computed tomography (CT) scans in order to have a further basis for explanation of potential outliers among compression test results. To independently study the significance of cell structure, a second reference sample series was introduced based on thermally treated TiH2 as foaming agent. Treatments of this and similar kind have been demonstrated to allow tailoring of decomposition kinetics and can thus be employed to modify or improve pore structure and morphology [13–18]. Moreover, documentation of structural features is required for investigating any change induced by additives in this respect. The matrix alloy (c) itself is kept the same in all sample series, while the expression of the microstructure (d) is deliberately modified between series—either by means of appropriate additives (Sr, B, TiB2) or by varying the cooling rate in solidification.

Background to this approach are earlier observations suggesting that in Al-Si foams, coarse forms of the eutectic (lamellar or needle-shaped Si phase) can lead to Al-Si interface planes similar in size to the typical cell wall thickness, the latter being approximately 80 μm for the alloy system in question [19]. Such interfaces have been suggested as preferred initial failure sites in aluminum alloy foams [20]. The notion that Si particles and their geometry influence failure is also supported by several fracture mechanical studies on bulk Al-Si and closely related alloys, though mostly focusing on failure under conditions of fatigue. Among these, Gall et al. suggest that fatigue cracks preferably grow along the Al-Si interface [21]. Su et al. also observed particle debonding under conditions of wear based on experimental and numerical studies, while alternative mechanisms include particle fracture and plastic deformation of the Al matrix. Spherical shapes of Si particles are shown to reduce susceptibility to fracture [22,23]. Lados et al. confirmed the role of both primary Al dendrite and Si phase shape and size in this respect [24]. Chan et al. suggested that crack growth in fatigue of B319 Al alloy is mostly via fractured and debonded Si particles, while interdendritic grain boundaries provide preferred fracture paths in later stages of rapid crack growth [25]. In contrast, Xia et al. derived quantitative values of considerable magnitude (namely an interface shear strength of 240 ± 6 MPa and a normal strength of

247 MPa) from nanoidentation-based experimental studies combined with finite element analysis and claimed a good match with certain atomistic simulations [26].

Grain refinement and modification are thus directed at eliminating potential weaknesses caused by the foam matrix alloy microstructure. The relevance of such investigations is stressed by the fact that due to their good processing characteristics, near-eutectic Al-Si alloys in a composition range from approximately 7 wt.% Si upwards and related systems, e.g., containing further additions of Mg or Cu, have retained their role as backbone both in classic metal foam and metal foam sandwich production [2,27] and in more recent developments such as Advanced Pore Morphology (APM) foams [28–31]. In its concentration on additive-based microstructure modification, the approach complements studies on heat treatment of foams [32–34] and variation of the matrix alloy [35–37]. Of these, when it comes to identifying the role of matrix alloy and specifically microstructure, the latter will suffer greatly from coincidental differences in expansion characteristics and thus pore structure, as has recently been shown in much detail by Helwig et al. [38]. The former, in contrast, allows control of matrix material properties at constant pore structure, but is limited to alloys that show a notable response to heat treatment, such as the 6000 and 7000 series Al alloys susceptible to precipitation hardening.

#### *1.2. Grain Refinement and Modification of Al-Si Alloys*

While grain refinement and modification of Al-Si and related alloys are established techniques in metal casting, neither has yet been evaluated in the context of powder metallurgically produced foams. Additions of TiB2, a substance known for its refining capability, have as yet only been considered in the context of particle stabilization, e.g., by Kennedy et al. Grain refinement was also not investigated, nor was any such effect to be expected in these studies as a result of TiB2 particle size, which greatly exceeded dimensions of nucleation sites effective in the microstructural refining effect [39].

Depending on the understanding specifically of grain refinement, transfer of this essentially liquid phase technique to powder metallurgy may seem contradictory, and, in fact, the relevant literature is mostly associated with casting processes during which the liquidus line is passed with a considerable margin. An exception to this rule is a study by Nafisi and Ghomashchi, who also considered the semi-solid region in their work on A356, i.e., AlSi7 Mg, alloys, though only with respect to casting billets for further, semi-solid processing. Thus, the effects of, e.g., inhomogeneous distributions of refiners in the semi-solid state, is not reflected in their investigation [40]. In any case, though metal foams of the type covered here do start as a powder metallurgical (PM) precursor, they cross solidus and usually also liquidus temperature for a limited time during foaming. It has repeatedly been shown that unless specific measures are taken to ensure that expansion occurs solely in the semi-solid region [38,41,42], the final microstructure of the foam is entirely formed during solidification [36]. A remaining concern is the question how effective treatments can be if modifiers are contained only in a limited Al powder fraction, which is diluted by addition of conventional Al and alloying element powders. For economic reasons (cost of specially prepared powders with refining/modifying agents), such a processing route seems mandatory if modification and refinement are to be established on a commercial basis.

Grain refinement generally relies on increasing the number of available nucleation sites in a melt. This can either be achieved by influencing the constitution of the melt in a way that the critical radius above which nuclei may grow is reduced, and thus using mechanisms of homogeneous nucleation, or via heterogeneous nucleation by offering additional nucleation sites. Current state of the discussion suggests that standard treatments with TiB2/Al3Ti and AlB2 combine effects of both kinds, as is outlined in a dedicated review on the underlying principles of grain refinement mechanisms provided by Easton and St. John. According to them, the models proposed thus far fall into two main categories, termed the Nucleant and the Solute Paradigm. The Nucleant Paradigm itself encompasses nucleant particle theories, which stress the role of TiB2 and isomorphous AlB2 particles as primary nucleation sites, and phase diagram theories, which consider the properitectic Al3Ti phase as the main nucleation

site. The problem of the former theories is that based on crystallographic considerations and some experimental evidence, both TiB2 and AlB2 have to be considered poor nucleants and only really show their benefits in the presence of additional Ti. In comparison, Al3Ti is a powerful nucleant due to a number of beneficial orientation relationships with α-Al and its peritectic reaction forming α-Al. However, pure phase diagram theories once again fail to deliver an explanation for the superior performance of combination of certain Ti levels with TiB2, AlB2 and (Al,Ti)B2 as used in commercial grain refiners of the so-called 5-1 Ti-B-Al kind. This deficiency is reflected in theories combining aspects of both views, including a possible higher Al3Ti stability in Al melts in the presence of borides or the peritectic hulk theory, which assumes a preferred nucleation of Al3Ti on the melt-TiB2 interface [43,44]. More recently, Schumacher et al. have demonstrated that the effectiveness of TiB2 particles in fact relies on their being covered by Al3Ti layers, which only form at Ti content levels exceeding 0.15 wt.% [45].

For modification of the eutectic Si phase, Sr- and Na-based treatments are common practice. Explanations of their effectiveness consider the influence of additives on Si nucleation and crystal growth. Among the latter, a distinction is made between kinetic effects caused by reduction of Si diffusion coefficients [46] and surface tension [47,48] in Sr-containing Al melts and blocking of Si lattice planes preferred in further crystal growth by local Na enrichment [49]. More recent studies by Srirangam et al. confirm significant influence of Sr additions on the liquid state structure of Al-Si alloys seen in simulations using synchrotron radiation, thus supporting a role of the additive in nucleation of the Si phase [50]. Further contributions to this topic by Timpel et al. based on atom-probe tomography and transmission electron microscopy suggest parallelism of two mechanisms, of which growth restriction of the eutectic Si phase is just one. Their measurements indicate the existence of two types of co-segregation of Sr with Al and Si, of which type I causes multiple twin formation in the Si crystal ("impurity-induced twinning"), which further facilitates multiple direction crystal growth, while larger size type II segregations restrict Si crystal growth [51].

Common to all theories, be they directed at grain refinement or eutectic modification, is that they require additional nucleation sites or solved additives in homogeneous distribution once solidification starts. Obviously, this is not necessarily a given thing in a PM material in which (a) refining/modifying agents may only be present in part of the metal powder components, and which (b) reaches the liquid state solely in the course of foam expansion and thus for seconds rather than minutes before solidification is initiated to stabilize the developing foam. Stability of the modified foam is an issue in itself, as earlier investigations indicate that B and Sr, but also TiB2 additions alter characteristics such as surface tension and viscosity [50,52–54], parameters that are instrumental in determining a metal foam's susceptibility to detrimental phenomena such as cell wall rupture, coalescence or drainage [15,55,56]. For this reason, initial studies of refinement and modification have been performed on powder compacts without foaming agent and varied levels of additive-containing Al powders.

The present study will thus elucidate the possibility and the effects of transferring the aforementioned approaches for microstructural optimization to PM aluminum foams by evaluating microstructural features, expansion characteristics and mechanical behavior of such materials with and without additives.

#### **2. Materials and Methods**

As motivated above, the work described here is divided in three major sections, namely (a) the general evaluation of grain refinement and eutectic modification of powder compacts, (b) the evaluation of expansion characteristics of such powder compacts with addition of 0.5 wt.% TiH2 as blowing agent and (c) the microstructural and mechanical evaluation of foams produced from these precursor materials.

In this, steps (a) and (b) turned out to effectively narrow down the set of samples suitable for detailed studies along the lines of step (c).

#### *2.1. Materials and Sample Production*

Production of precursor materials and foams followed the principles of the Fraunhofer or powder compact melting process. Alloying was based on mixing of elementary Al and Si powders. Table 1 lists the powders used and their composition, including, where applicable, levels of modifying agent content, as well as measures of particle size according to specification and measurement. Powders containing modifying or grain refining additions were specifically prepared for this purpose by Alpoco and match in constitution materials commonly employed in preparation of powder compacts used in melt treatment of aluminum casting alloys. The choice of Sr as modifying agent as well as B and TiB2 for grain refinement was based on the dominant role of these additives in melt treatment in casting of aluminum in general, and specifically of hypoeutectic Al-Si alloys [57–60].

Particle sizes were measured using a Coulter LS 130 laser particle size analyzer (Beckman Coulter Inc., Brea, California, USA). Powders were dispersed in ethanol and subjected to ultrasonic agitation to break agglomerates before measurement.

Mixing of powders for hot pressing was done in a tumble mixing device. Compaction was realized as two-step process at a furnace temperature of 450 ◦C, using a Zwick 1474 universal testing machine with added furnace (ZwickRoell GmbH & Co. KG, Ulm, Germany), the initial load being 60 kN or 74.6 MPa during the first and a constant 90 kN (111.9 MPa) during the final stage. Holding time was 20 min for each step. The hardened steel dies with circular cross-section were filled at room temperature. During the first compaction stage, settling effects and plastic deformation in the powder bed led to a load decrease to below 20 kN. Hot extruded precursor material based on identical Al, Si and TiH2 powders was acquired from Entwicklungsgemeinschaft Schunk-Honsel for use as reference. Powder mixtures were cold isostatically pressed to a relative density of approximately 0.75 to 0.85. The billets gained were heated to a temperature of 450 ◦C and extruded using a die with rectangular cross-section of 160 × 40 mm. The thickness of 40 mm allowed producing cylindrical samples for expansion measurements and foaming of specimens for compression tests from the extruded precursor material with the central axis oriented perpendicular to the extrusion direction and the larger transversal extension of the extruded geometry.

Axially compacted samples were turned to 31 mm diameter and 28.5 mm height, corresponding to a maximum global sample of 0.66 g/cm3 considering the foaming molds used. Higher levels of foam density were achieved by interrupting foam expansion prior to complete filling of the mold. Both extruded and hot pressed materials show a preferred direction of expansion. In both types of samples, this direction coincides with the cylinder axis.

Foaming of all samples took place in a specially designed furnace configuration with an attached cooling device allowing both water and forced convection cooling. Furnace and attached cooling rig were designed to be raised and lowered automatically to avoid moving the then unstable sample prior to solidification. Foaming molds were machined from high temperature oxidation resistant steel grades (1.4713, 1.4828) with internal dimensions of D45 × 55 mm. The setup, including a foaming mold, is depicted in Figure 1.


Composition, production method and supplier of powder variants.

> **Table 1.**

1 Ecka Granules. 2 Not applicable.

**Figure 1.** Foaming furnace used for sample production—(**a**) general layout including control units for heating (left) and movement (right), (**b**) nozzle arrangement for forced convection (FC) cooling (below furnace, with cylindrical foaming mold).

For initial evaluation of refining and modification effects achievable in powder compacts, altogether nine different types of samples without blowing agent were produced containing three different levels of B, TiB2 and Sr, respectively. These specimens, as well as reference samples containing neither grain refining nor modifying additives, were subjected to thermal treatments at 660 and 680 ◦C, meant to simulate the foaming process. For those powder compacts representing the medium level of additives, a further variation of the quenching step following thermal treatment was foreseen. Full details of all sample series are reported in Table 2.


**Table 2.** Overview of powder compact sample series for evaluation of grain refinement and modification of the Al-Si eutectic.

<sup>1</sup> Ti as Al-5Ti-1B.

Evaluation of the expansion characteristics of a subset of the above sample types, including all additives, led to three different compositions of foamed samples being produced for structural, microstructural and mechanical evaluation. These are listed in Table 3. Further foam samples were produced to substantiate the observations on microstructural features made on powder compact samples.


**Table 3.** Overview of foam sample series for mechanical testing. As also implied by the series' designations, forced convection cooling was employed in all cases.

<sup>1</sup> Forced convection using compressed air.

#### *2.2. Microstructural Characterization*

Foam sample preparation followed general guidelines laid down by Müller et al. and Mosler et al. [61,62]. To clearly distinguish microstructural features such as the individual primary Al grains, Barker etching was employed. The method provides grain-level contrast under polarized light based on local crystallographic orientation via this feature's influence on the growth of thin oxide layers during anodic etching [63]. For quantitative image analysis, the software package A4i-Analysis developed by Acquinto was used.

Grain size was established via two alternative approaches. For Barker etched samples, the largest extension of several grains' was measured on approximately 100 grains per sample. Since larger grains showed better contrast, grain size tends to be slightly overestimated in this case. As an alternative, scanning electron microscopy and electron backscatter diffraction (EBSD) were employed for image acquisition. The method proved critical in eutectic regions, where presence of several Al-Si transitions in close vicinity did not allow for unambiguous detection of orientations. To compensate for this effect, a dilation of clearly identified Al grains was performed, leading to a single phase microstructure with polyhedral grains. General consideration of the alloy's characteristics suggests that these approximate to the circumference of the complex shaped dendritic grains which make up the true microstructure. As a consequence, it is possible to derive the grain sizes of the true microstructure from its processed counterpart. In the present study, this has been done by measuring linear intercepts of grain boundaries for approximately 250 grains per sample. The second advantage, besides the increased database, is the fact that the method is capable of distinguishing and thus evaluating even very small grains, while the light microscopy approach effectively stressed larger grains showing better contrast due to the subjective selection step involved. In the following text, whenever grain sizes are given, the fundamental principle used in determining these is given.

Analysis of the degree of modification was based on metallographic sections images of which were acquired using light microscopy. In following Ohser and Lorz's recommendation, the specific interface area SV between Al and Si phase in the eutectic as well as the integral of the average curvature of the Al-Si boundary MV in the two-dimensional (2D) section were determined [64].

As additional parameter, values of secondary dendrite arm spacing were established relying on light microscopy images of metallographic sections. Two methods were employed, namely selection of dendrites followed by interactive measurement of arm spacing and an automated approach based on the assumption that the thickness of dendrite arms matches the arm spacing. The latter method was used on anodically etched bright field light microscopy images providing high levels of contrast between the primary aluminum grains and the eutectic phase, thus allowing straightforward binarization. Evaluation first singled out isolated dendrite arms not connected (in the respective 2D section) to a dendrite stem, then determined and averaged the minimal Feret diameter of these. Generally, good agreement between both methods was observed; however, less effort and greater numbers of dendrite arms included in the analysis make the latter the principle of choice.

#### *2.3. Pore Structure and Morphology*

For evaluation of the pore structure, both cutting of foam samples to produce photographic images and computer tomography (CT) were employed. In the former case, preparation relied on wire EDM cutting as a means to avoid exerting mechanical loads on foam structural members. Part of the CT measurements was executed by Dr. Illerhaus at the Bundesanstalt for Materialforschung und -prüfung (BAM). Further CT investigations were performed on reference compression test samples using a Procon CT-MINI device, and on foamed samples using modified Yxlon equipment. Resolution was 60 μm, 65.2 μm and approximately 200 μm, respectively. The latter exceeds minimum cell wall thickness by an approximate factor of two, but still allows reconstruction of cellular structure based on averaging effects.

#### *2.4. Foam Expansion Measurements*

Foam expansion characteristics were determined using the so-called mechanical expandometer to simultaneously measure furnace and sample temperature as well as volume expansion of the foam by means of Ni-CrNi thermocouples and position encoders based on inductive sensors [65]. Samples tested were of 29 mm diameter and 9 mm height. Per sample type, at least three measurements were performed. Furnace temperature control was set to 750 ◦C, leading to an approximate 770 ± 5 ◦C within the furnace just beside the steel tube containing the sample. To eliminate side effects attributable, e.g., to thermal expansion of the device itself, a correction curve was calculated based on three measurements under the same conditions but without sample. This curve was subtracted from each individual expansion measurement. A further, temperature based correction relied on matching of solidus temperatures determined by differential scanning calorimetry (DSC, Netzsch STA 409 C) with temperature profiles measured during foam expansion.

#### *2.5. Mechanical Testing and Evaluation of Test Results*

General concerns about viability of mechanical test results in cellular materials were put to the test by Andrews et al. and more recently Yu et al. Andrews et al., as a consequence, proposed the principal dimensions of samples to be at least seven times the average pore size [66] and Yu et al. suggested specimen heights to exceed this parameter at a factor of six [67]. Alkheder and Vural basically confirmed these observations in a recent numerical study and add considerations of boundary layer influence, which they found to be insignificant once sample edge length equaled 10 times the average pore size [68]. With representative pore size values for the type of foam studied here between 2 and 3 mm, a diameter of 30 mm at a height of 25 mm was considered acceptable. Compliance with the criterion was checked, with positive results, based on CT measurements.

To eliminate effects of the foams' typical solid skin, samples for mechanical testing were machined to final dimensions from foam cylinders originally measuring 45 mm in diameter and 55 mm in height by means of turning or wire EDM cutting. The main axis of these specimens coincides with that of the original foam cylinder, the main direction of expansion and gravity during foaming and the direction of the applied compressive load. Samples tested parallel to the direction of expansion have previously been shown to exhibit lower strength than those tested perpendicular to it, most likely due to a density gradient of identical direction [11]. In the present case, limitation of sample height to 25 mm allowed selecting a region of constant density in sampling.

Compression tests were performed in the quasi-static regime at a constant strain rate of 0.1 s−<sup>1</sup> and stopped once densification had clearly been reached for all density levels, i.e., at a total strain of at least 80%, or at a load exceeding 100 kN (corresponding to 203.7 MPa). Modified foams were tested at TU Bergakademie Freiberg using a MTS 810 universal testing system. Reference samples were tested at Fraunhofer IFAM using a Zwick 1476 universal testing device and a 100 kN load cell.

As principal characteristics of the stress-strain response on which further evaluation was based, tangent modulus (defined as the maximum slope in the elasto-plastic region), ultimate compressive

strength and plateau strength were selected. The latter was determined as the stress value associated with the intersection of a linear fit to the plateau region with a tangent to the elasto-plastic region of the stress-strain curve at the point of maximum inclination.

The strain interval for linear plateau region fits has an arbitrarily chosen lower boundary at 0.1 engineering or technical strain, while for the upper limit, a density dependent formulation was chosen in accordance to the initiation of densification as proposed by Gibson and Ashby [12]:

$$
\varepsilon < (1 - 1.4 \cdot \rho\_{\rm rel}) \cdot (1 - \text{D}^{-1}) \tag{1}
$$

The value of the parameter D depends on, e.g., the characteristics of the matrix material. For metallic foams, Gibson and Ashby suggested a value of 2.3 [12], which has been taken over for the present study. Both definitions ascertain that for the whole density range studied neither initial stress peaks nor densification affect plateau stress values.

Ultimate compressive strength (UCS) was defined as the peak stress reached immediately after the elasto-plastic region. Identification of such a peak, which does not occur in very ductile foams [32], proved possible for all samples.

Results of this kind were evaluated using Weibull statistics [69]. For this purpose, all mechanical characteristics derived were normalized based on Gibson's and Ashby's fundamental models of strength and elastic modulus as function of density, which reads as follows, in its simplified form [12]:

$$\sigma\_{\rm UCS/pl} = \mathbb{C}\_{\rm UCS/pl} \cdot \rho\_{\rm rel}^{3/2} \tag{2}$$

$$\mathbf{E} = \mathbf{C}\_{\mathrm{E}} \cdot \mathbf{p}\_{\mathrm{rel}}\,^2\tag{3}$$

Normalization can be realized by dividing the respective property value by the relative density to the applicable power, i.e., 1.5 for plateau and ultimate strength and 2 for the tangent modulus, seen here as elastic materials characteristic. In the present study, a twofold deviation from this straightforward approach was followed: first, the dependence of density on strength was formulated based on absolute rather than relative density. Both descriptions are equivalent when assuming that in the former case, the respective power of the matrix density's reciprocal forms part of a redefined constant CE. Second, instead of using the standard values introduced in Equations (2) and (3), normalization relied on the exponent of foam density found when fitting a power law function to the experimental density-dependent compression test data.

#### **3. Results**

#### *3.1. Microstructural Modification of Powder Compacts*

Figure 2 summarizes data on the influence of cooling rate and levels of different additives on grain size. Results from sample series 1 (holding temperature/time 660 ◦C/5 min.) are included. The diagram confirms that refining and modifying additions can reduce grain size significantly even at the highest cooling rate that could be realized by water quenching. Refining effects are visible even at lower additive levels, i.e., for an initial constitution which contains additives only in part of the aluminum powder fraction. Nevertheless, highest additive levels lead to smallest grains, but in size these still retain an order of magnitude which slightly exceeds the minimum cell wall thickness.

Figure 3 shows selected microstructures complementing the above diagram. In these, Barker etching is employed to achieve distinction between grains. Lines hold reference, Sr-, TiB2- and B-containing samples from top to bottom, while from column 1 to 3 the cooling rate increases from natural via forced convection to water quenching. Additive content levels are 250, 350 and 140 ppm for the Sr, B and TiB2 treated samples, respectively. Specifically the water-cooled Sr-based samples appear to show an exceedingly fine microstructure, additional quantitative data on which is related in Figure 4. A similar observation is made in Figure 5, which contrasts non-etched metallographic sections corresponding to the same sample types. The finding is, however, not directly reflected in the quantitative data gathered in Figure 2; grain sizes are almost identical for water-quenched samples containing additives, with Sr assuming an intermediate position between B and TiB2. A possible explanation for smaller grain sizes to be found in Sr-containing samples is the fact that Sr boosts the level of undercooling in solidifying Al alloys, and thus, promotes formation of additional crystallization nuclei [70].

**Figure 2.** Average grain size of powder compacts, (**a**) as a function of Sr, B and TiB2 additive content when quenched in water, (**b**) depending on cooling rate for reference material without additives and samples containing medium levels of Sr, B and TiB2 addition as given in the diagram.

**Figure 3.** *Cont.*

**Figure 3.** Metallographic sections of powder compacts containing (**a**–**c**) no additives (AlSi7 reference material), (**d**–**f**) Sr, (**g**–**i**) TiB2 and (**j**–**l**) B additions after thermal treatment (melting) for 5 min at 660 ◦C and cooling via (**a**,**d**,**g**,**j**) natural convection (NC), (**b**,**e**,**h**,**k**) forced convection (FC) and (**c**,**f**,**I**,**l**) water quenching (WQ), revealing differences in grain size and morphology (Barker etching). Content levels of Sr, TiB2 and B are 250, 350 and 140 ppm, respectively.

Figure 4 displays graphically the influence of Sr, TiB2 and B additions on the morphology of the eutectic phase for the two holding times and temperatures compared. For B-refined samples, only data for sample series 1 is available. The diagrams show that both TiB2 and B additions have a coarsening effect on the eutectic phase. With rising additive content, both the integral of average curvature and the specific boundary layer increasingly fall below the reference data set by the additive-free samples. For Sr, the situation is different, as a coarsening is observed only for the lowest content level, whereas specifically the data from sample series 2 treated for 10 min at 680 ◦C testifies to the expected effect, i.e., a significant rise of both parameters scrutinized. Nevertheless, the focus of the investigation remains on sample series 1, as the analysis of temperature and expansion versus time curves for the various materials suggests that the respective conditions are a better approximation of the thermal history of foam than the increased time and temperature values adopted for sample series 2.

**Figure 4.** *Cont.*

**Figure 4.** Influence of holding time and temperature on expression of geometrical characteristics of the eutectic phase in powder compacts after melting and resolidification, (**a**) with Sr additions, (**b**) with TiB2 and (**c**) with B additions. Cooling is done by water quenching in all cases. For samples containing B additions, experimental data is available only for materials held for 5 min at 660 ◦C (series 1). The legend provided in Figure 4b refers to all diagrams.

Micrographs corresponding to the diagrams in Figure 4 are presented in Figures 5 and 6. While the former reflects the influence of cooling rate without as well as at the medium additive levels considered, the latter relates the impact of the amount of additives under conditions of water quenching, and thus, directly corresponds to the quantitative data presented above.

**Figure 5.** Powder compacts, morphology of the eutectic phase, influence of cooling rate and additives. Columns from left to right correspond to (**a**,**d**,**g**,**j**) natural convection, (**b**,**e**,**h**,**k**) forced convection and (**c**,**f**,**I**,**l**) water quenching, lines from top to bottom showing (**a**–**c**) reference samples without additives, followed by (**d**–**f**) 250 ppm Sr, (**g**–**i**) 350 ppm TiB2 and (**j**–**l**) 140 ppm B. All images taken at identical magnification.

**Figure 6.** Powder compacts, morphology of the eutectic phase, influence of additive content at high cooling rates. In columns from left to right, additive level is stepped up from 130 to 250 to 465 ppm (Sr, top row, (**a**–**c**)), 260 to 350 to 465 ppm (TiB2, center row, (**d**–**f**)) and 70 to 140 to 280 ppm (B, bottom row, (**g**–**h**)). Note the twofold increase in magnification compared to Figure 5.

Figure 5 underlines the dominant effect of cooling rate on grain refinement as opposed to that of moderate additive levels. Contrast between the third column and the two others exceeds the difference between the reference sample in the top row and the additive-containing ones at matching cooling conditions. Still, both a refining and a modification effect are discernable: Specifically at forced convection, i.e., at slightly elevated cooling rates, the eutectic phase seems finest in B and TiB2-containing samples. Figure 6 optically confirms the data represented in Figure 4, as specifically the coarsening of the eutectic structures is clearly visible in lines 2 (d–f) and 3 (g–i), which correspond to TiB2 and B additions, while the finest structure is revealed in Figure 6c for highest Sr levels.

Since direct measurement of cooling rates in a solidifying foam is difficult, as a result, it may differ significantly depending on whether or not a thermocouple was in actual contact with the matrix material during data acquisition, and literally impossible if the macroscopic structure development must not be influenced (which is the case when using thermocouples within the sample); thus, alternative ways of establishing correct values have to be developed. Correlating microstructural features with solidification rates thus becomes attractive. Secondary dendrite arm spacing (SDAS) can be used for this purpose, with an expression of the type:

$$
\lambda\_{\mathbf{u}} = \mathbf{A} \text{ v}^{-1/3} \tag{4}
$$

Describing the relationship between both parameters according to Sahm et al., where λ<sup>a</sup> describes the actual SDAS value in μm and ν the cooling rate in Ks−<sup>1</sup> [71]. While the overall size of grains, be they dendritic in nature or not, is clearly influenced by refining additions, literature shows that secondary dendrite arm spacing (SDAS) is not. This allows SDAS measurements to be performed on reference samples containing neither additives nor foaming agents. For the current study, calibration experiments have been performed based on the melting of powder compacts without foaming agent and cooling via natural convection, forced convection using a pressurized air supply and a water jet. During all experiments, cooling rates were measured using thermocouples, while associated SDAS values were determined metallographically. Following suggestions from Schumann, the measured "solidification rate" was determined as the average cooling rate between 550 ◦C and 400 ◦C [63]. As a result, the value A = 45.75 has been gained by fitting an equation of the aforementioned type to the measured points, see Figure 7 as well as Equation (6). In doing so, the original equation has been modified to the two parameter form given in Equation (5):

$$
\Lambda\_\mathbf{a} \text{ [}\mu\text{m]} = \mathbf{A} \cdot \mathbf{v}^\mathbf{n} \tag{5}
$$

$$
\Lambda\_{\text{a}}\,\,\text{[\mu m]} = 45.75 \cdot \text{v}^{-0.3295} \text{, R}^2 = 0.94211 \,\tag{6}
$$

**Figure 7.** Measured values for secondary dendrite arm spacing (SDA) and cooling rate and fit curve based on Sahm's equation (Equation (4), [71]).

For the second parameter, the exponent n, regression results in a value of −0.3295, which almost matches the suggested theoretical value of −1/3, and thus, supports the viability of the underlying measurements. Based on this, it is possible to ascribe actual cooling rates to the foamed compression test samples via a microstructural feature.

Figure 7 underlines the fact that SDAS is independent of Sr, B or TiB2 additions, as no systematic deviation of the various sample series from the fit curve can be asserted. Study of the powder compacts has provided insights into refining and modification effects of the various additives considered. Furthermore, the investigations allowed to compare influence of additives with the effect of cooling rate specifically on grain size. The findings provide a microstructure-based measure of cooling rate which can be transferred from fully dense materials to foams, as well as a quantification thereof (see Equation (6)).

#### *3.2. Expansion Characteristics and Pore Structure of Precursor Material Variants*

Figure 8 sums up the expansion measurements performed on different material variants. Measurements on B-containing samples have not been included based on the results of the previous evaluation of grain refinement performance, which favors further study of TiB2 in the composition ranges accessible. The error bars denote the standard deviation observed within the sample series

tested (three to six samples evaluated in each case). Comparison of the maximum porosities achievable shows that both Sr and B additions adversely affect foamability. The effect is most evident for B additions, where levels as low as 70 ppm suffice to reduce maximum expansion from an average of 438.8% to 342.2%, and thus, by 22.0%, whereas Sr addition levels of 250 ppm and 465 ppm lead to a decrease of 8.5% and 21.8%, respectively.

**Figure 8.** Maximum expansion of various precursor material variants, highlighting the influence of additive type and content.

This finding is contrasted by improvements in maximum achievable porosity achieved by addition of TiB2. Ti contents of 350 ppm result in a maximum expansion which equals both the standard material (sample series Al-1/TiH2 as received) at an average expansion of 430.9% and an otherwise identical material based on pure aluminum powders with reduced oxygen content (sample series Al-2/TiH2 as received, maximum expansion 425.1%), the latter matching the respective values of the additive-containing powders. This sample series has explicitly been included to rule out a major influence of the oxygen content of Al powders on deviations in expansion characteristics. The parameter as such is known to significantly influence this property, as has, e.g., been shown by Weigand, who determined the effective oxygen content range for pure Al alloys to be between 0.2 and 0.72 wt.% [72], while Asavavisichai reported high expansion for the same material, though in a cold rather than hot compacted state, at oxygen contents 0.24, 0.3 and 0.33 wt.%, with a decrease observed for 0.73 wt.% [73]. The fact that the oxygen content of all powders used falls into this range supports the observation that no significant differences in foamability were found between the two standard sample series without additives and different oxygen content levels. Thus, the results of this comparison underline that for the present study, any influence of oxygen content on expansion is secondary to phenomena associated with the various additives. This point is stressed by the fact that a further rise in Ti content to 465 ppm brings about a major increase in expansion by 18.5% to 520.1%. This superior performance of the TiB2 sample series matches similar observations by Kennedy et al. on pure Al foam, though TiB2 addition was studied in terms of stabilization and strengthening instead of grain refinement, and thus, used a much higher content level of 10 wt.% TiB2 [39].

Sr additions to Al-Si alloys differ in their effects on characteristics such as melt viscosity and surface tension based on the Si content. Limited reduction in viscosity is reported for eutectic compositions, while an increase has been observed for hypoeutectic alloys according to Song et al. [54].

Lower melt viscosity is generally considered to adversely affect foam stability. Static metal foam stabilization concepts suggest increasing melt viscosity, e.g., by means of adding non-surface-active ceramic particles, to suppress cell wall drainage, thinning and rupture of cell walls [55,56,74]. In contrast, concepts of foam stability based on dynamic influences indicate that low viscosity might favor healing of local defects which could otherwise develop into rupture sites during stretching of these membranes in foam expansion [15]. The latter point is of interest when considering the superior performance

exhibited by TiB2 based samples. For hypoeutectic compositions and specifically AlSi7, as investigated here, Yan et al. suggest a reduction in viscosity at least for the semi-solid state when TiB2/TiAl3 is employed for grain refinement [53].

On the other hand, no effect of TiB2 additions on surface tension is known, whereas Sr is reported by [50] to reduce this properties' value in Al-Si alloys. Considering once again static foam stability criteria, a decrease of surface tension should influence stability to some advantage, since it would mean a reduction of free surface energy. However, the dynamic concept suggested by Lehmhus assumes that high surface tension increases the driving force for compensation of neck formation during stretching of cell walls in the course of foam expansion, and may thus increase foam stability in conditions where dynamic effects represent the major threat to foam stability [15]. Similarly, Nadella et al. correlate reduced stability in Al-Si-Mg foams with the lowering effect of Mg additions on the surface tension [75].

The overall result of expansion measurements is fundamental for the decision to concentrate on Sr and TiB2 as grain refining/modifying agent in the course of the present study, and solely on TiB2 in terms of sample production for foam microstructure evaluation, and specifically for mechanical testing.

Figure 9 contrasts the cell morphology of samples containing no additives and as received and thermally treated titanium hydride as foaming agent with that of a sample containing untreated TiH2 in combination with TiB2 additions for grain refinement. The images reveal no noticeable influence of this additive on the cellular structure, whereas it is clearly visible that thermal treatment of the foaming agent leads to a more regular structure with increased average pore size.

**Figure 9.** Foam pore structure, alloy variants AlSi7 with "as received" TiH2 as foaming agent (left), thermally treated TiH2 as foaming agent (center), "as received" TiH2 as foaming agent and Ti/TiB2 modification (right). Left and center image represent scans of cut samples, while the right image is derived from computer tomography (CT) data.

#### *3.3. Microstructural Modification of Foams and Influence on Mechanical Properties*

Microstructural evaluations of foams have been performed for materials containing no additives, 465 ppm of Sr and 465 ppm of Ti as TiB2. The parameters evaluated include the SDAS value, from which actual cooling rates could be derived. Data was gathered in top, central and bottom locations within the foam. The findings suggest that top and center locations experience similar thermal conditions, whereas the bottom of the foam samples is subjected to slower cooling. This observation may be linked to effects of drainage, the influence of the thermal mass associated with the mold support, and the shielding of this part of the mold from forced convection cooling and water quenching. Table 4 lists the cooling rates determined from SDAS in conjunction with Equation (6) for the different positions and sample compositions.


**Table 4.** Local cooling rates determined for foam samples containing no as well as Sr and Ti as TiB2 additives (465 ppm each) determined via evaluation of SDAS.

The listing in Table 4 shows that cooling rates reached within the foam may even exceed the values actually measured in solid powder compacts. This may be explained by reductions in heat capacity per unit volume which coincide with the increased porosity. However, the finding must be seen critical as it implies that the derived cooling rates are in many cases based on an extrapolation of Equation (6) to values not covered by the original data set, which only extends to about 33 K/s (see Figure 7).

Figure 10 summarizes the studies on grain refinement in foams and should be matched with Figures 2 and 4 for comparison with results obtained on powder compacts. Figure 11 adds the corresponding metallographic sections. The top line contains images of Barker etched samples giving an indication of grain size, while the bottom line images allow evaluation of the eutectic structure.

**Figure 10.** *Cont.*

(**c**)

**Figure 10.** Foam microstructure, diagrams: (**a**) grain size as function of additive level and (**b, c**) geometrical characteristics of the eutectic phase as a function of cooling rate.

**Figure 11.** Foam microstructure, (**a**–**c**) Barker etching emphasizing grain sizes, (**d**–**f**) expression of the eutectic phase. (**a**,**d**) Reference material AlSi7 with conventional, as received foaming agent, (**b**,**e**) with 465 ppm Sr and (**c**,**f**) with 465 ppm Ti as TiB2 added.

Data in Figure 10a confirms that grain refining additions are effective and further add to the effect of cooling rate. TiB2 containing samples undercut the reference material by a margin of approximately 20% at low and in excess of 30% at high cooling rates, at which grain sizes well below 40 μm are observed. For Sr additions, given the prevailing level of scatter, no effect on grain refinement can be substantiated. It is noteworthy that these results fall below those of measurements on powder compacts as summarized in Figure 2 by approximately one order of magnitude.

In contrast, Figure 10b,c reveal that under conditions of foaming, in stark divergence from comparable findings based on non-foamed powder compacts and irrespective of the cooling rate, addition of Sr has very limited influence on the expression of the eutectic. Values of both the specific boundary layer SV and the integral of average curvature MV roughly match corresponding measurements on powder compacts, but show a more significant dependency on cooling rate than on additive content.

Figure 11 illustrates the findings expressed in the diagrams: Sr-, but even more so TiB2-containing samples show regions with fine and coarse eutectic morphologies that do not differ greatly from those observed in the reference samples.

For all density levels compared, foams produced with the thermally treated foaming agent show highest levels of strength. Moreover, they are characterized by a clearly defined stress peak immediately following the elasto-plastic region (encircled in Figure 12). As Alkheder and Vural pointed out, based on general considerations and 2D simulation results, such peaks can be interpreted as expression of stored elastic energy, which is released once collapse of the structure starts. Naturally, occurrence of this effect is favored in regular structures, in which the capacity of resistance to an external load is distributed more homogeneously, and thus localization of failure only arises at higher levels of global load [68,76].

When comparing grain refined and non-refined material variants based on untreated foaming agents, ultimate compressive strength levels nearly match for densities of approximately 0.3 and 0.4 g/cm3. At higher densities, the non-refined foams outperform the refined ones. Initial peaks can be identified, though they are less clearly distinguished in these sample series, and specifically so in the grain refined one.

**Figure 12.** *Cont.*

**Figure 12.** Compression test results—comparison between stress-strain curves associated with different density levels: 0.3 g/cm<sup>3</sup> (**a**), 0.4 g/cm3 (**b**), 0.53-55 g/cm3 (**c**), 0.58-0.61 g/cm3 (**d**). Note the difference in the scaling of the x- (i.e., stress-) axis in Figure 12 (**a**,**b**) compared to (**c**,**d**).

A notable distinction between refined and non-refined samples is a postponed onset of densification observable in the former.

Figure 13 shows exemplarily the density dependence of plateau strength, ultimate compression strength and tangent modulus for the two unmodified reference series samples and the grain refined sample series. The diagram includes the fit curves derived for the various properties and specimen series using a variant of the standard Gibson/Ashby approach for describing the dependence between density and mechanical characteristics in foams already introduced as Equation (2) (plateau and ultimate compressive strength) and 3 (tangent modulus) above. Fitting is done according to a least squares method with variation of the factor C and the exponent n. Table 5 sums up all values derived for the parameters C and n and the three sample series. As all evaluations are based on the absolute density of the foam, the parameter C as given in the table differs in definition from the form introduced in Equations (2) and (3).

(**a**)

**Figure 13.** *Cont.*

(**c**)

**Figure 13.** Compression test results: ultimate compressive strength (**a**), plateau strength (**b**) and tangent modulus (**c**) versus density.


**Table 5.** Compression test results, comparison of fit curve parameters for density dependence of plateau and ultimate strength as well as tangent modulus for different sample series (standard series with TiH2 as received and thermally treated, TiB2 grain refined with TiH2 as received—all series produced using forced convection cooling with air).

<sup>1</sup> Indicates the associated constant quantified in column 3.

These data clearly show that among all sample series, those based on the thermally treated foaming agent show by far the best adherence to the general dependency of strength and stiffness on density postulated by Gibson and Ashby, even though the values of the fitting curve's exponent do neither meet expectations for stiffness (with an expected value of 2 compared to 1.47 for the respective sample series) nor for strength (1.5 versus 1.98 and 1.56, respectively).

Thus, since this is the major difference specifically between both sample series without modification, but different foaming agents, it must be assumed that this phenomenon is based on the respective expression of the pore structure, which has already been shown to be favorable in the case of thermally treated TiH2 in Figure 9 as well as various earlier publications [13,14,16,17,77].

Closer observation, however, reveals that the expression of this structural deviation is alleviated at higher levels of porosity. In the present case, a boundary of this kind can be identified at a density of about 0.4 g/cm3, which corresponds roughly to the earlier observations related to the stress strain curves in Figure 12, which also tend to deviate more significantly at lower porosity and level of expansion. The observation as such is in line with earlier studies suggesting a healing of initial deficiencies in pore structure following an extended period of expansion, i.e., an expansion to a higher final porosity [14,77]. This finding is of considerable importance, as it may serve to define a limit in density above which pore structure can be effectively controlled to adapt global mechanical performance, which in turn implies that microstructural effects to this end should primarily be sought for—and aimed at—below this density level.

A fundamental assumption of the present study was that failure of foam is considerably affected by local failure initiation sides either in the form of Al-Si interfaces or grain boundaries approaching in their dimensions those of cell walls and struts. Such a dependence of failure initiation on local, randomly distributed microstructural weaknesses suggest a description of strength based on a Weibull distribution function. General recommendations for this type of analysis suggest a number of at least 30 samples/data points. At 38 and 60 for grain refined and untreated reference samples, this condition is fulfilled in both cases. Moreover, approximation via a Weibull distribution function is considered justified if an asymmetric shape of the probability density function is observed. This has qualitatively been verified for the three parameters plateau strength, ultimate compressive strength and tangent modulus via the observed histograms of these characteristics in their normalized form. The respective diagrams are contrasted in Figure 14, which contains data for non-refined, reference and TiB2-refined foams. Essentially, the non-refined foams do show a slight asymmetry, while the grain refined samples tend to exhibit a more symmetric distribution. Note that the parameters compared here are normalized using the observed density dependence as described in Section 2 of this manuscript.

**Figure 14.** Distributions of normalized properties for the reference material AlS7 expanded using "as received" TiH2 and the same with added TiB2 grain refiner—(**a**,**b**), plateau strength, (**c**,**d**) ultimate compressive strength and (**e**,**f**) tangent modulus.

The following Figures 15–17 depict Weibull evaluation plots of the same compressive properties—namely plateau strength, ultimate compressive strength and tangent modulus. The Weibull distribution parameters for derived from this evaluation for both sample series are given in Table 4. The underlying cumulative distribution function is given by:

$$F(\rho) = 1 - \exp\left(-(\sigma\_0 \cdot \rho^{-1})^m\right) \tag{7}$$

(**b**)

**Figure 15.** Compression test results—Weibull evaluation of plateau strength for AlSi7 with (**a**) and without TiB2 grain refiner (**b**), as-received TiH2 as foaming agent.

(**a**)

**Figure 16.** *Cont.*

**Figure 16.** Compression test results—Weibull evaluation of ultimate compressive strength for AlSi7 foam with (**a**) and without TiB2 grain refiner (**b**), as-received TiH2 as foaming agent.

(**b**)

**Figure 17.** Compression test results—Weibull evaluation of tangent modulus for AlSi7 foam with (**a**) and without TiB2 grain refiner (**b**), as-received TiH2 as foaming agent.

In general, the plots in Figures 15–17 show a good match of the experimental data with the expected linear relationship in the double-logarithmic plot and thus support the possibility of expressing material performance via a Weibull distribution. For both plateau and ultimate compressive strength, the respective scale parameter reaches higher values in the case of the non-refined reference material. The tangent modulus, however, clearly deviates from this behavior (Table 6). Common to all material properties, however, is the fact that the shape parameter, i.e., the modulus m, is increased by the refinement. As higher values of this parameter indicate a narrower distribution, the conclusion is that grain refinement can indeed reduce the level of scatter observed.

**Table 6.** Compression test results, comparison of Weibull distribution parameters for AlSi7 foam with and without TiB2 grain refiner with as received TiH2 as foaming agent.


#### **4. Discussion**

The present study shows that grain refinement of PM aluminum foams made from hypoeutectic Al-Si matrix alloys is a viable approach. Both the microstructural evaluation of powder compact samples without foaming agent as well as the corresponding studies on actual foam samples show a significant influence of the refining agent chosen, TiB2. However, contrasting data gathered from powder compacts and foam samples shows that though the relative influence of the refining treatment on grain size is similar in both cases, the absolute grain sizes differ significantly. For powder compacts, values roughly between 150 and 900 μm were found, while foam samples show grains of approximately 40 to 150 μm. Differences in cooling rate may be excluded as underlying cause of this phenomenon, as the evaluation of secondary dendrite arm spacing (SDAS) and the establishment of a correlation between this microstructural feature and the cooling rate on powder compacts has facilitated the association of cooling rates with the grain sizes measured on foams, too. As expected, the respective data do not support any large deviation of the cooling rates prevalent in foams from the range covered in the experiments performed on powder compacts. However, there is still room for several possible explanations: first, Ti content in foam samples is naturally increased, as the blowing agent, TiH2, will add to it. The result may be a secondary grain refining effect superimposed to that of the deliberately added TiB2. Second, the topology of the foam itself, and specifically the fact that the dimensions of its main structural elements, cell walls and struts fall short of the grain sizes measured on powder compacts by up to one order of magnitude, will prohibit the formation of large grains in a foam. This sterical hindrance is amplified further by the fact that the smooth surface of the solidified foam's cell walls suggest that solidification and thus nucleation of grains occurs not only within these membranes, but may even start at the interface between liquid and gas, potentially helped by the fact that, e.g., oxide particles present in the matrix alloy [78] can accumulate here and act as heterogeneous nucleation sites, leaving once again less room for grains to grow within the cell walls and struts.

In contrast to grain refinement, modification of the eutectic structures has not been achieved in foams. The reason for this maybe deduced from the comparison of the two powder compact sample test series, which differ in temperature and in the time of exposure to the respective temperature. As mentioned earlier, comparison of initial powder compact sample series clearly shows that successful modification of the eutectic structure in a material produced by compaction of a powder mixture in which only one component brings in the modifying agent apparently requires an extended process window for distribution of the dissolved modifier. 10 min at 680 ◦C serve this purpose, while 5 min at 660 ◦C do not entirely; however, the latter is a better representation of the actual conditions during foaming as performed in the present study, which uses unmodified foaming agents in additive-containing samples. Since these conditions cannot be altered without putting the stability of the foam at risk—which, as this

study has shown, is compromised by Sr addition anyway—the only way to achieve Sr-based eutectic modification in a metal foam of the type studied here would seem to be choosing an Al powder that contains the optimum level of modifying agent from the start. This would effectively eliminate the need for distributing Sr by the time- and temperature-dependent process of diffusion within the matrix material. In the present study, this approach was not adopted, since working with a master alloy-like powder containing increased levels of Sr allowed facile production of samples covering a range of Sr contents rather than a single value. In any case, the adverse effect of Sr on foam stability practically ruled out the production of compression test samples from this material variant and thus limited the studies on modification to microstructural investigations.

Comparison of compression test data between samples with and without grain refiner but identical, non-optimized foaming agents suggests that macroscopic deficiencies dominate mechanical response. The effect of grain refinement is thus somewhat obscured in stress-strain curves as well as strength-versus-density plots. This finding is stressed further by the obvious increase in strength observed for the third sample series based on thermally treated foaming agents. As a consequence, further studies on grain refinement should be based on samples with comparable regularity in macroscopic structure to facilitate a clearer discrimination of the influence of refinement and modification. Meanwhile, the fact that the optimization of the macroscopic foam structure achieved by means of using a thermally treated, oxidized and depleted blowing agent [14,77] not only results in the highest strength of all sample types compared, but also shows least scatter (see fit quality as expressed by R<sup>2</sup> in table) and best adherence to the values of the exponent of the strength-density relation theoretically predicted by Gibson and Ashby [12] underlines that it is in fact this macroscopic structure which primarily controls a metal foam's mechanical performance. In this respect, the results of the present study confirm the hierarchy of influencing factors put forward in the introduction based on authors like Mosler et al., Martin et al. or Mu et al. [8–10].

Furthermore, it is apparent that additional improvements in grain refining efficiency are required to bring grain sizes down to levels which clearly fall below the cross-sectional dimensions of the main structural elements of an aluminum foam of the type considered here. This relates to cell walls, but predominantly to the struts as the primary structural members; their size at least should ideally be undercut by one order of magnitude. At present, with typical membrane thicknesses of approximately 80 μm and struts approaching two–three times this value [19], this has not yet been achieved. The expected benefit of this would be a shift of cell wall failure mechanisms to regions in which the quasi-isotropic behavior of polycrystal dominates. As of now, the ratio between cell wall thickness and typical grain size implies that single crystal plasticity effects, and namely anisotropy of grains with unfavorable orientation relative to the loading direction, may act as additional weak spots within the foam's structure. Moreover, as the studies of microstructural features of the powder compacts have shown, it must be assumed that any beneficial effect of grain size reduction is partly obscured by the observed coarsening effect of TiB2 and B on the expression of the eutectic phase. In this context, it must be considered unfortunate that the insufficient expansion characteristics of the Sr-containing materials did not allow stable production of sufficient numbers of samples for mechanical testing and specifically Weibull evaluation. Despite these facts, the Weibull evaluation does show a beneficial effect of grain refinement in the increased values of the Weibull modulus m in samples treated, a characteristic that indicates a reduction in scatter of the respective properties.

#### **5. Conclusions**

Though in principle successfully applicable, the results obtained show that grain refinement of the type employed here cannot lead to grain sizes that match or undercut the dimensions of typical cell walls. Thus, it remains questionable whether this approach alone can have significant influence on the mechanical performance of the foams. A different picture is observed for modification treatments affecting the eutectic phase. Here, both the influence of the modifying agents as well as secondary effects of the refining treatment can shift the size of the respective microstructural features to levels

below typical cell wall thickness levels. This is of major relevance in view of the initial argument that specifically the interface area between Si and Al phases within the eutectic can act as weak spot at which cell wall failure is initiated. However, transferring this effect to foams requires a base material, i.e., matrix material powder, which already contains the modifying agent in homogeneous distribution and thus eliminates the need to achieve just this by diffusion during the foaming process; the present study has shown that the combination of time and temperature typical for foam expansion cannot guarantee this. Moreover, Sr additions introduced for eutectic modification turned out to adversely affect expansion characteristics—hence, Sr-modified samples could not be included in the evaluation of the foams' mechanical performance.

Comparison of density-dependent strength and stiffness shows modified foams as weaker compared to both their untreated counterparts. A possible explanation may be that on the whole, more ductile matrix combined with an irregular structure, possibility of ductile deformation promotes premature macroscopic yield when compared to the fracture-based failure associated with the initial assumption that Si phases might be preferred fracture planes in cell walls.

Thus, the investigations clearly show that the influence of microstructural features is limited when it comes to average strength levels determined for samples identical in terms of composition—except for the respective modifying agents—and processing conditions. The scatter of these average strength values, however, is noticeably reduced by grain refinement. For designing with a material, the consequences of this achievement are similar to an increase in strength, since reduced scatter means that higher stress levels can be accepted at a given level of safety.

Furthermore, it is of great importance for metal foam production that the equivalence of two different concepts to achieve the intended microstructural changes could in principle be demonstrated; both an increase in cooling rate and the use of grain refining agents yield similar results. The fact that the effectivity of the former approach seems more pronounced does not devaluate the latter, since in foam production, local cooling rates are limited specifically where large parts are concerned due to effects such as the reduced thermal conductivity within the foam. Thus, grain refinement based on TiB2/TiAl3 additions may prove a valuable alternative.

Comparisons based on non-foamed powder compacts have shown that a limited increase in foaming time and final temperature will positively affect the efficacy of treatments; specifically for treatments based on solution of active substances, this finding is less than surprising. This is an additional argument for combining modification and grain refinement with the use of thermally treated foaming agents, which naturally induce a shift of processing conditions in the required direction. Moreover, such a combination, though not tested in the present study, must seem attractive based on the chance to combine improvements in micro- and pore structure.

Future investigations should have a closer look at conditions of adding grain refining substances, especially in terms of increasing levels of concentration of TiB2/TiAl3-containing Al powders, since using large proportions of such specialty powders for precursor material production would have unwanted economic implications. However, basing refinement on such an emphasized master alloy approach may cause local variations in the availability of nucleation sites, especially when assuming that the nucleant paradigm correctly describes the process, as in this case, non-soluble constituents, which cannot widely be redistributed during foam expansion, take the stress in initiating nucleation. The question of the ideal level of addition also needs to be answered, since in the current study maximum values were set by the powders available. Moreover, a further study of size effects should be envisaged to clarify to what degree microstructural modification will affect results of the kind seen by Blazy et al. [79].

Finally, the ongoing search for alternative grain refiners and modifying agents should be taken into account; recent developments in this area have been summarized by Easton et al. as well as Liu [58,60]. These, and other studies such as that of Xu et al., have highlighted potential new approaches in grain refinement, e.g., scandium addition [80]. Beyond a switch to new refining and modifying agents, combinations of both processes may also be considered [81], the more so since the present study has shown that Sr- and B-based modification on the one hand and TiB2-based grain refinement have opposed effects on foam expansion, Thus, combinations might partially cancel out negative side effects of Sr- and B-based treatments. In general, eutectic modification should in itself not be neglected—specifically if solutions can be identified with less detrimental effect on foam expansion. In the end, it is not only grain size that may reach the dimensions of cell walls, but also the size of eutectic silicon structures, which may just as well act as preferred fracture planes.

In conclusion, we believe the present study has shown that grain refinement of aluminum foam has both promises and challenges. Of the latter, not all have been successfully addressed yet. There is room for improvement, and we believe that our results provide a sound basis for such future research, for which we have suggested several promising paths.

**Author Contributions:** Conceptualization, U.M. (Ulrike Mosler), D.L., D.H., J.W. and U.M. (Ulrich Martin); methodology, U.M. (Ulrike Mosler), D.H. and D.L.; validation, D.H., U.M. (Ulrike Mosler) and D.L.; formal analysis, D.H., U.M. (Ulrike Mosler) and D.L.; investigation, D.H., U.M. (Ulrike Mosler) and D.L.; data curation, D.H., U.M. (Ulrike Mosler) and D.L.; writing—original draft preparation, D.L.; writing—review and editing, D.H., U.M. (Ulrike Mosler), U.M. (Ulrich Martin), J.W. and D.L.; visualization, D.H. and D.L.; supervision, U.M. (Ulrike Mosler), D.L. and U.M. (Ulrich Martin).

**Funding:** Part of this research was funded by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) within SPP 1075, "Cellular Metallic Materials".

**Acknowledgments:** CT measurement facilities were made accessible by B. Illerhaus, Bundesanstalt für Materialprüfung (BAM), Berlin, and J. Gudat, ProCon GmbH, Hannover, whose help is greatly appreciated. Illerhaus also gave support in the evaluation of CT data. Furthermore, Ray Cook, Alpoco, provided Al powders and contributed to the discussions on grain refinement and modification in general.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Automated Continuous Production Line of Parts Made of Metallic Foams**

#### **Isabel Duarte 1,\*, Matej Vesenjak 2and Manuel J. Vide <sup>3</sup>**


Received: 20 April 2019; Accepted: 6 May 2019; Published: 8 May 2019

**Abstract:** The paper presents an automated continuous production line (7 m × 1.5 m × 1 m) of high-quality metallic foams using a powder metallurgical method. This continuous production line was used to obtain metal foam parts and/or components by heating the foamable precursor material at melting temperatures close to the temperature of the metallic matrix and cooling the formed liquid metallic foam (in liquid state), which then results in a solid closed-cell metallic foam. This automated continuous production line is composed of a continuous foaming furnace, a cooling sector and a robotic system. This installation has enabled a technological breakthrough with many improvements solving some technical problems and eliminating the risks and dangers related to the safety of workers due to the high temperatures involved in this process. The whole process becomes automatic without any need for human intervention.

**Keywords:** aluminum alloy foams; powder metallurgy; continuous production; mechanical properties

#### **1. Introduction**

Metallic foams are one of the most interesting lightweight materials that have a high potential to be used in large-scale in various areas of engineering. They are multifunctional, lightweight, recyclable and non-flammable with reduced environmental impact. The fact that they may be formed by open-cells and/or closed-cells, makes it possible for them to be widely used in the near future [1]. The open-cell metallic foams are mainly used as functional materials for chemical and biomedical applications (e.g., treatment of water, heat exchangers) [2]. The closed-cell foams are used as structural materials for commercial, industrial and military applications (e.g., energy absorbing structures, ballistic protection) [1,3–5]. The latter exhibit an excellent structural efficiency and excellent capacity to absorb (impact and sound) energy, together with noise and vibration damping in structures such as houses, cars, trains and other devices and equipment [6]. These special advantages are a combination of properties derived from their porous cellular structures along with the properties of the metal they are made of. These foams are considered as one of the future materials that must be lightweight, durable, ecofriendly and economical. When compared to the other foams made of polymer and ceramic, these foams have more potential. Polymer foams [7] are flexible, but cannot support high temperatures, while metallic ones are extremely tough and can plastically deform and possess a high capacity of energy absorption. Also, metallic foams are easily recyclable unlike some polymeric foams, which cannot be recycled or, if they are, it is quite expensive. Moreover, polymeric foams increase fire hazards, releasing smoke and toxic gases, which does not happen with the non-flammable metallic foams. Ceramic foams [8] are relatively stiff, endure high thermal shocks and are more stable in harsh

environments compared with metallic and polymer foams. However, ceramic foams have a brittle behavior that limits their uses, while metallic foams have a ductile behavior that is more appropriate for engineering purposes. Despite the advantages of metallic foams, they have not been used in a large scale yet. The main reason is the fact that the current manufacturing processes to prepare the closed-cell foams do not allow a meticulous control of the cellular structures in terms of their pores' shape and size. The control of cellular structures of these foams will make it possible to predict the mechanical, thermal and electrical behavior.

Two main strategies have been adopted to achieve cellular metals with regular structures. The first strategy consists of improving the most viable existing manufacturing processes, as the case of the powder metallurgy that allows the preparation of the ductile metallic foams to be used as structural materials. Powder metallurgy method consists of heating a foamable precursor material to a temperature close to the melting temperature of the metallic matrix in which this foamable precursor material is prepared by the hot compaction of a mixture of the metal and blowing agent powders [9]. The improvement was achieved by using pre-treated blowing agent powder (e.g., titanium hydride) instead of untreated powder, ensuring that the initial decomposition temperature is close to the melting temperature of the metallic matrix [10]. The second strategy is to develop new classes of the cellular metals with closed pores using methods that allow the control of the size and shape of their pores. Advanced pore morphology (APM) metallic foam elements [11], syntactic foams [12] and hybrid foams [13] are some of these examples. The APM metallic foam elements are simple spheres of metal closed-cell foams, which are obtained by the powder metallurgy method using the smallest pieces of the precursor material (e.g., 1–2 mm in size) [11]. The small sample sizes ensure that surface tension forces during the melting process are relatively large compared to the hydrostatic pressure, forming near spherical shapes with an easily reproducible unit cell [14]. Syntactic foams are prepared by infiltrating a metal melt into the interstice's spaces of a packed hollow spheres made of metal or ceramic or porous particles [15,16]. Hybrid foams are prepared by impregnating an open-cell metal foam with a polymer [17]. Despite these developments, the closed-cell metallic foams prepared by the traditional powder metallurgy method have several advantages in comparison with other new cellular metals. For example, the syntactic foams have high density (>2 g/cm3) when compared to the typical closed-cell metal foams (<1 g/cm3) that limits their application, especially for lightweight constructions [6]. Also, in the case of hybrid foams the use of polymers increases the risk of fire hazards. Furthermore, the powder metallurgy method allows to fabricate parts with complex geometries, to fill the hollow structures with a formed liquid metal foam during its formation [18] and to easily incorporate metallic inserts (e.g., screws) [19] into the metal foam during its formation, promoting a metallic bonding. For example, aluminum alloy closed-cell foams have been extensively tested as core [20] and filler materials [21,22] of sandwich structures and thin-walled tubes, mainly to be used as energy absorbing structures for vehicles in which the joining between the solid thin-walled structure and the closed-cell aluminum alloy foam is made during the foam formation. However, since the metallic foam component is prepared using a batch furnace, the quality of the components depends on several factors including manufacturing parameters (e.g., temperature and time) and the experience of the workers. In order to have a strict control over the process, without the need of human intervention in dangerous tasks, an automated continuous production line was developed to obtain parts and/or components of metallic foams. This article presents this production line, illustrating several case-studies as well as the potentiality of this newly developed installation.

#### **2. Materials and Methods**

Two foamable precursor materials with 160 × 20 mm (Figure 1a) and 20 mm × 5 mm (Figure 1b) in cross-section, made of AlSi7 alloy with 0.5 wt. % of titanium hydride were prepared by a combination of cold isostatic pressing and hot extrusion as described in [23]. Table 1 presents characteristics of the used powders. The powder mixture of aluminum alloy and titanium hydride were first compacted to cylindrical slugs with 70–80% of theoretical density by cold isostatic pressing (CIP). After this, the cylindrical slug was pre-heated to 360 ◦C and extruded to rectangular bars of 160 × 20 mm and 20 mm × 5 mm in cross-section through a horizontal 25 MN direct extrusion machine. Closed molds made of S235JR carbon steel (0.17% C, 1.40% Mn, 0.045% P, 0.045% S, 0.009% N) were made using steel plates and thin-walled tubes. Square (outer width: 25 mm; wall thickness: 2 mm) and cylindrical (outer diameter: 30 mm; wall thickness: 1.5 mm) thin-walled tubes made of AA 6060 T66 having an outer and inner diameter of 30 mm and 26 mm, respectively, were cut to the required length (e.g., 150 mm). Carbon steel bars (diameter: 4 mm) made of S235JRG2C (0.20% C, 1.40% Mn, 0.045% P, 0.045% S) were used as reinforcements to prepare the in-situ reinforced foams.

**Figure 1.** Foamable precursor materials made of aluminum alloys and titanium hydride with 160 × 20 mm (**a**) and 20 mm × 5 mm (**b**) in cross-section.


**Table 1.** Properties of the powders.

The compressive and three-point bending properties presented in this work were measured using a 50 kN servo-hydraulic dynamic INSTRON 8801 testing machine (Instron, Norwood, MA, USA) under quasi-static loading conditions (crosshead loading rate: 0.1 mm/s).

#### **3. Results**

The development of this automated continuous production line was based on the thorough knowledge acquired in the fabrication of metallic foams using laboratory batch furnaces [9,23]. In the traditional process, a foamable precursor material is placed into the cavity of a closed mold. The worker places the mold containing the precursor into a pre-heated furnace at high temperatures. This is a dangerous task for the worker due to exposure to extremely high temperatures (700–800 ◦C for the case of aluminum alloys). The total weight of the mold with the foamable precursor material should also be appropriate for the easy handling by the worker.

After that, the heating of the foamable precursor material leads to a formation of the liquid metallic foam due to the thermal decomposition of the blowing agent and the melting of the metal. The precursor expands, filling the cavity of the mold. During this step, the worker should supervise the exterior of the mold to follow the filling process of the mold. When the first melt leaves the mold, the worker must quickly yet carefully remove the mold containing the formed liquid metallic foam. In addition to this, the formed liquid foam must be solidified immediately to prevent the collapse mechanisms (drainage and coalescence) and must be done in a very steady manner to avoid any turbulent movements that could damage the foam [24]. This is another dangerous task in which the worker is, once again, exposed to extremely high temperatures. To perform the dangerous tasks, the workers should use personnel protective equipment, such as heat protection clothing, gloves and safety glasses. In the traditional procedure, the cooling is not controlled since it is the worker who

cools the surrounding of the mold with a hose of compressed air until room temperature is obtained. Results show that formation of defects and structural imperfections within solid foams are created due to the fact, that the solidification is not uniform [25]. All of these aspects could increase the number of rejected metallic foam pieces. Automated continuous production line was developed to eliminate the dangerous tasks performed by the workers, to ensure that the quality of the foam is not compromised by the experience of the worker and to guarantee a controlled solidification. The main components of the continues production line are the continuous furnace, cooling sector and robotic system that will be described below.

#### *3.1. Continuous Foaming Furnace*

Figure 2 shows a scheme (Figure 2a) and photos (Figure 2b–d) of the continuous foaming furnace developed to fabricate 3D components of metallic foams. This continuous furnace has five zones, which are the feed zone of the foamable precursor material into the molds (open or closed) or into the thin-walled structures, followed by the three heating zones and the exit zone. The loading and movement of the foamable precursor material within the furnace is achieved by means of a conveyor belt, as shown in Figure 2d. The length of the feed zone is approximately 0.7 m (Figure 2b), while the length of the exit zone is approximately 0.3 m (Figure 2c).

**Figure 2.** Scheme (**a**) and photos of the continuous foaming furnace belt, showing the feed (**b**) and exit (**c**) sectors and the conveyor belt (**d**).

The furnace has three heating zones respective to the three foaming stages until the maximum expansion value is reached [26]. The first foaming stage accounts for temperatures below the solidus temperature of the metallic matrix, where the expansion is very small. In the second foaming stage, a rapid increase of expansion is observed due to the metal being in liquid state and due to the releasing of the gas into the metal from by the thermal decomposition of the blowing agent, simultaneously. The third foaming stage also represent a rapid increase in expansion as the temperature increases until the maximum expansion is reached. After that, gas is no longer released, and the formed liquid foam starts to collapse due to the coalescence and drainage mechanisms. Thus, the formed liquid metallic foam should be removed from the furnace once it is close to the maximum expansion to avoid its collapse. The three heating zones are of the same dimensions (size: 0.3 m in height and 0.4 m in width) and could be controlled individually according to the thermal foaming cycle.

All of the heating zones could be operated from room temperature up to 1273 K. This furnace is heated by means of electrical resistances, which were installed at the top of the furnace and under the conveyor belt. The conveyor belt is made of stainless-steel chains to support the weight of the foamable precursor material and molds or hollow structures, as well as the thermal fatigue during the thermal foaming cycle.

Two panel doors were installed in the furnace to decrease the temperature variation throughout the heating zones, as well as to reduce the turbulence airflows effect. One of them is located at the entrance of the first heating zone (Figure 2b) and the other is located at the end of the third heating zone of the furnace (Figure 2c). Both panel doors have windows for observation of the furnace's interior. Each heating zone (Figure 2b,c) has also got a window located on the sidewall of the front furnace for visualizing and controlling the foaming process. The movement of the foamable precursor material within the furnace is associated with the movement of the conveyor belt created by a set of engines and transmission systems. The maximum linear speed of the conveyor belt is approx. 0.5 m/s. The temperature of the three heating zones and speed rate of the conveyor belt should be adjusted according to the characteristics of the foamable precursor material and the geometry and dimensions of the component to be fabricated. Figure 3 shows two examples of rejected foam parts (Figure 3a,b) and the block foam prepared with adjusted parameters (Figure 3c). The use of a high speed of the conveyor belt and/or low temperatures of the heating zones will not allow enough time for the precursor material to expand and filling the cavity of the mold, as shown in Figure 3a. On the other hand, a very low speed rate of the conveyor belt promotes the collapse of the foam since the formed liquid metallic foam remains at high temperatures for a long time, leading to a high loss of the liquid metal, as can be seen in Figure 3b. For each foam part, the parameters of the furnace should be adjusted to obtain a high-quality foam as shown in Figure 3c.

**Figure 3.** Rejected foam parts using a steel mold (200 <sup>×</sup> <sup>80</sup> <sup>×</sup> 50 mm3) due to no filling the cavity of the mold (**a**) and collapsed foam (**b**) and foam block with properly adjusted parameters (**c**).

#### *3.2. Cooling Sector and Robotic System*

A cooling sector was designed and built to control the solidification of the liquid metallic foam, decreasing the number of rejected foam parts. In fact, uncontrolled solidification of the liquid metallic foam causes defects and structural imperfections, decreasing the mechanical properties of the resulting solid metal foams. Figure 4 shows rejected parts and nearly perfect foam parts, which were prepared with uncontrolled (Figure 4a) and controlled (Figure 4b) cooling, respectively.

This cooling sector was installed immediately after the continuous foaming furnace. The robotic system was also built and installed in the cooling sector to extract the mold or the hollow structures filled with the liquid metallic foam from the furnace at high temperatures to the cooling sector, making it an automatic operation without any human intervention. Figure 5 shows a 3D model of the cooling sector (Figure 5a) composed by the robotic system (Figure 5b), the cooling systems and the conveyor belt (Figure 5c) designed by SolidWorks software. Figure 6 shows real photos of the cooling sector and an overview of the final automated continuous production line.

**Figure 4.** Foam block (80 <sup>×</sup> <sup>50</sup> <sup>×</sup> 50 mm3) and a complex foam part (outer diameter: 250 mm) prepared by the uncontrolled cooling (**a**) and controlled cooling (**b**).

**Figure 5.** Scheme (**a**) and photos of the continuous foaming furnace belt, showing the feed (**b**) and exit (**c**) sectors and the conveyor belt.

**Figure 6.** Cooling sector (**a**) and the automated continuous production line developed (**b**) for fabricating parts made of metallic foams.

The cooling sector is a metallic structure equipped with a high-pressure turbine connected to two manifolds drilled and mounted on both sides of a conveyor belt. The angle of the manifolds drilled varies according to the dimensions of the required component. A robotic system was developed to extract the mold containing the formed liquid metallic foam from the furnace at a high temperature to the cooling sector, which makes it possible to operate it without any human intervention required. This robotic system is installed in the cooling sector. The mechanical arm is built from stainless steel AISI 304 and is mounted at the vertical rail made of aluminum and a horizontal rail made of the same material, as shown in Figure 6. The vertical drive is controlled by a pneumatic cylinder, while the horizontal one is controlled by a motor reducer with the possibility of adjusting the speed rates and positions. The movement of this robotic system is associated with the command of the opening and closing of the furnace door at the end of the third heating zone. The mechanical arm system receives a

#### *Metals* **2019**, *9*, 531

trigger signal, starts to move towards the furnace door, while the furnace door also opens. This signal is triggered when the mold touches the suspended metal chain in the third heating zone. Briefly, the sequence steps during this operation are:


This last operation is associated to the velocity of the conveyor belt of the cooling sector, which is different from the first conveyor belt of the continuous furnace. Two conveyor belts were installed to avoid any interference between the foaming process and the solidification of the formed liquid foam to solid foam. The first conveyor belt moves the precursor material from the loading sector through the continuous furnace to the exit sector, where the expansion of the precursor occurs. The second conveyor belt is placed in the cooling sector, where the foam is solidified. Both carrier conveyor belts are made of stainless-steel chains. The steel chains were designed to support temperatures up to 1000 ◦C in continuous operation, to assure a good thermal fatigue resistance, a good corrosion resistance, good dimensional stability and a good resistance to weld to the workpiece surface. The movement of the precursor material (with or without the mold) is associated to the movement of the conveyor belt created by the engine/motors and the drums. This is equipped by drive motors that ensure a continuous motion through the furnace and the rate variations according to the materials to be expanded. The control of the velocity of the conveyor belt and temperatures of each heating zones are programmed according to the required thermal cycle of foaming.

#### *3.3. Operation Mode*

The operation mode of the automated continuous foaming production line is as follows:


#### *Metals* **2019**, *9*, 531

The command and control parameters of the entire installation are performed through one electric console. For example, it controls the thermal foaming cycle associated to each heating zone, the velocities of the conveyor belts, as well as the movement of the doors.

#### *3.4. Case-Studies*

Simple and complex parts made of aluminum alloy foams were fabricated using this automated continuous production line. The closed mold is built according to the dimensions and geometry of the foam part to be fabricated. Figure 7 illustrates an example to fabricate a complex foam part, showing the 3D model designed in SolidWorks software (Figure 7a), the closed mold made of stainless steel constructed for this purpose (Figure 7b) and finally, the components made of aluminum alloy (Figure 7c).

**Figure 7.** 3D model of the complex foam prototype designed in SolidWorks software (**a**). Stainless steel closed-mold (**b**). Complex foam parts (approx. 290 mm in height) (**c**).

The molds could be fabricated using screw system or clamping system, as illustrated in Figure 8a,b respectively. The use of molds with clamping system makes it much easier to extract the resulting solid foam components compared to those with screw system, which require a special tool, reducing the global production time. Figure 9 shows examples of real molds with screw system (Figure 9a) and with clamping system (Figure 9b), which were constructed to prepare aluminum alloy foam parts.

**Figure 8.** Molds with screw system (**a**) and with clamping system (**b**).

**Figure 9.** Real molds with screw system (**a**) and with clamping system (**b**).

Figure 10 shows simple and complex foam parts made of aluminum alloys fabricated using this automated continuous production line with stainless steel molds constructed for each component. The foamable precursor material is selected according to the dimensions and geometry of the respective component. For example, the foam blocks and the complex foam parts were prepared using the foamable precursor material with 160 mm × 20 mm in cross-section (Figure 1a). Simple thin foam panels and sandwich foam panels were prepared using the foamable precursor material with 20 mm × 5 mm in cross-section (Figure 1b). The rectangular block of the foamable precursor material (160 mm × 20 mm in cross-section) is usually cut according to the characteristics of the foam component to be fabricated (volume, geometry and dimensions). However, foam parts could be fabricated using various small precursor pieces (Figure 11a) instead of a single precursor [27], as illustrated in Figure 11.

**Figure 10.** Simple thin foam panels (e.g., 210 <sup>×</sup> <sup>250</sup> <sup>×</sup> 15 mm<sup>3</sup> and 300 <sup>×</sup> <sup>35</sup> <sup>×</sup> 15 mm3) and foam blocks (e.g., 200 <sup>×</sup> <sup>50</sup> <sup>×</sup> 50 mm3 and 80 <sup>×</sup> <sup>50</sup> <sup>×</sup> 50 mm3) (**a**) and complex foam parts (e.g., steering wheel 250 mm in outer diameter) (**b**).

**Figure 11.** Foam parts (rectangular block: 80 <sup>×</sup> <sup>50</sup> <sup>×</sup> 50 mm3; cylindrical foam: 30 mm in diameter and 60 mm in height) (**a**) fabricated using rejected precursor pieces (**b**).

This manufacturing process has several advantages when compared to other existing processes to produce metallic foams. It enables the production of parts and components of complex geometry of good quality in a simple, effective and practical way, as illustrated in Figures 7 and 10b. Another advantage is that it joins the metal foams with other materials during its foam formation, promoting a metallic bonding. This facilitates and simplifies its application, since it eliminates the expensive joining step, leading to highly competitive products. Moreover, the application of the common joining techniques [28–30], such as welding and brazing, damages the cellular structures of the foams. The aluminium alloy foams are usually used as a core or a filler of sandwich panels and thin-walled structures, respectively. Results have shown that it is possible to fabricate in-situ foam filled tubes [18,20] (Figure 12a) and in-situ sandwich foam panels [21] (Figure 12b) using aluminium alloy tubes or aluminium alloy sheets by applying this manufacturing process. The manufacturing conditions could be adjusted to ensure the structural integrity of these sheets or tubes when exposed to high foaming temperatures that are close to the melting temperature of the alloy. Results have also demonstrated that the thermal treatment that is submitted to these tubes or sheets at high temperatures during the foam formation is beneficial to obtain a predictable and stable mechanical behavior of the resulting in-situ foam structures. The ductility of these structures increases, leading to an efficient crashworthiness without formation of cracks and abrupt failure when subjected to compressive and bending loads. Results also show that the global weight of the resulting in-situ foam structures can be controlled by reducing the wall thickness of the tubes or sheets [18]. Moreover, the metallic inserts

could be incorporated into the foams [31], resulting in in-situ reinforced foams, as shown in Figure 12c. This could be an advantage for the industry. For example, fasteners or other standard parts used in vehicles, like screws, bolts and pin rivets could be incorporated, minimizing the discontinuity, slip and fracture under loading utilized in the processes such as screwing.

**Figure 12.** In-situ foam filled tubes (150 mm in length, 30 mm in outer diameter) (**a**), in-situ sandwich foam panels (rectangular panel: 210 <sup>×</sup> 250 <sup>×</sup> 15 mm3-foam core and two aluminum sheets 1.5 mm in thickness; long panel: 300 <sup>×</sup> 35 <sup>×</sup> 15 mm3-foam core; two aluminum sheets 1.5 mm in thickness) (**b**), and simple and in-situ reinforced foams (length: 150 and 200 mm; diameter: 25 mm) made of aluminum alloys (**c**).

The metallic foams prepared by powder metallurgy method have a more ductile behavior when compared to the direct foaming methods [32,33]. In addition to this, the broad spectrum of foam parts with a free design can be easily fabricated with a good surface finish (Figures 10 and 12c) as an outer surface dense (integral) skin around each component is created during its production. Results have demonstrated that the powder metallurgy is a near-net shape manufacturing technology that produces foam components close to the finished size and shape, minimizing the number of secondary finishing processes. Results have shown that this technology allows a reduction in the number of processing steps (no joining step), a significant reduction in the amount of waste material and an ability to easily produce complex and advanced geometries. Results have indicated that AlSi7 foam parts or components could be fabricated at 700 ◦C. Tables 2 and 3 give an overview of the main properties of some parts made of aluminum alloy foams, in which some of them are presented in Figures 10–12, such as the cubic and cylindrical integral-skin foams, in-situ foam filled tubes made of aluminum alloys and in-situ reinforced carbon steel reinforced foams. The compressive and bending strengths of these foams increase with the foam density.


**Table 2.** Compressive properties of the simple and composite structures made of aluminum alloy foams.

\* φ and *h* are the diameter and height of the specimen; \*\* Outer and inner diameters of the thin-walled tube; \*\*\* Carbon steel bars of S235JRG2C (0.20% C, 1.40% Mn, 0.045% P, 0.045% S); \*\*\*\* Foam core.


**Table 3.** Bending properties of the simple and composite structures made of aluminum alloy foams.

\* φ and L are the diameter and the length of the specimen; \*\* Outer and inner diameters of the thin-walled tube; \*\*\* Carbon steel bars of S235JRG2C (0.20 %C, 1.40% Mn, 0.045% P, 0.045% S); \*\*\*\* Foam core.

The compressive strength of the foam components is higher than the bending strength. In fact, the closed-cell aluminum foams are used as crash energy absorbers.

Results have demonstrated that the larger the foam blocks, the higher the cellular architecture variation due to a non-uniform foam growth during the heating of the single precursor material [34,35]. For smaller foam blocks [34], a more regular distribution of the pores through the foam is obtained.

#### **4. Conclusions**

The automated continuous production line composed of a continuous furnace, a cooling sector and a robotic system is approximately 7 m long, 1.5 m high and 1 m wide. This foaming production line allows the production of larger quantities of foam materials when compared to those fabricated by the batch furnace. The new production line has several advantages in comparison to the conventional procedures as follows:


**Author Contributions:** Conceptualization, I.D. and M.J.V.; methodology, I.D. and M.J.V.; validation, I.D., M.J.V. and M.V.; formal analysis, I.D. and M.V.; investigation, I.D. and M.V.; data curation, I.D. and M.V.; writing—original draft preparation, I.D. and M.V.; writing—review and editing, I.D. and M.V.; visualization, I.D.; supervision, I.D.

**Acknowledgments:** Supports given by M. J. Amaral-Equipamentos Industriais Lda (Vale de Cambra), the Portuguese Science Foundation (FCT), UID/EMS/00481/2019-FCT and CENTRO-01–0145-FEDER-022083 (Centro 2020 program-Portugal 2020).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **E**ff**ect of Primary Crystals on Pore Morphology during Semi-Solid Foaming of A2024 Alloys**

#### **Takashi Kuwahara 1,\*, Mizuki Saito 1, Taro Osaka <sup>1</sup> and Shinsuke Suzuki 1,2**


Received: 23 November 2018; Accepted: 9 January 2019; Published: 16 January 2019

**Abstract:** We investigated pore formation in aluminum foams by controlling primary crystal morphology using three master alloys. The first one was a direct chill cast A2024 (Al-Cu-Mg) alloy (DC-cast alloy). The others were A2024 alloys prepared to possess fine spherical primary crystals. The second alloy was made by applying compressive strain through a Strain-Induced Melt-Activated process alloy (SIMA alloy). The third one was a slope-cast A2024 alloy (slope-cast alloy). Each alloy was heated to either 635 ◦C (fraction of solid *fs* = 20%) or 630 ◦C (*fs* = 40%). TiH2 powder was added to the alloys as a foaming agent upon heating them to a semi-solid state and they were stirred while being held in the furnace. Subsequently, A2024 alloy foams were obtained via water-cooling. The primary crystals of the DC-cast alloy were coarse and irregular before foaming. After foaming, the size of the primary crystals remained irregular, but also became spherical. The SIMA and slope-cast alloys possessed fine spherical primary crystals before and after foaming. In addition to average-sized pores (macro-pores), small pores were observed inside the cell walls (micro-pores) of each alloy. The formation of macro-pores did not depend on the formation of the primary crystals. Only in the DC-cast alloy did fine micro-pores exist within the primary crystals. The number of micro-pores in the SIMA and slope-cast alloys was one third of that in the DC-cast alloy.

**Keywords:** semi-solid; aluminum foam; primary crystals; SIMA process; slope casting; pore morphology

#### **1. Introduction**

New, lighter materials are required to reduce the environmental impact and cost of transportation equipment, including automobiles and aircrafts [1,2]. In addition, passenger safety should be maintained or enhanced by these new materials. Aluminum foam has attracted significant attention in recent years for meeting weight reduction and safety requirements. Aluminum foam has many pores and, owing to its ultralight and excellent shock-absorbing properties, could be applied to transportation equipment. However, the compressive properties of aluminum foam must be improved before it can be effectively used for this purpose. The compressive properties of foams are determined by their base metal and structure [3]. To this end, Fukui et al. [4,5] improved the compressive properties of aluminum foam using the super duralumin A2024 (Al-Cu-Mg) alloy, which is a well-known high-strength and lightweight material, as a base metal to improve the strength of the foam.

Foaming via the melt route is a common and efficient fabrication method for aluminum alloy foams [6]. With this method, the melt must be thickened to maintain pore stability. In general, thickening is achieved by adding Ca, which generates oxides that spread throughout the melt [7]. To fabricate the A2024 alloy foams, the oxide of the alloying element, Mg, is used as a thickener instead

of Ca. However, decreases in base metal strength have been observed because the quantity of Mg as an alloying element also decreases.

To solve this problem, Hanafusa et al. [8] fabricated foams in a semi-solid state using primary crystals, instead of oxides, as a thickener. Sekido et al. [9] investigated the thickening effect of primary crystals in a semi-solid state. Our group demonstrated that it is possible to fabricate an A2024 alloy foam without adding a thickener in a semi-solid state. However, the effects of this fabrication on the properties of foam have been researched only experimentally, not systematically.

Also, the primary crystals acting as a thickener in the melt are larger in size than those of the oxides. Moreover, the primary crystals are generally larger than the preferred range of thickener sizes in the melt route [10]. Therefore, foams fabricated in a semi-solid state may have different stabilization mechanisms than those in the melt route. How these large thickeners affect the formation of pores in foams is not evident. However, compared to oxides, the sizes and shapes of primary crystals are easy to control and evaluate. Therefore, it may be possible to improve the foam fabrication process by aggressively controlling the primary crystals.

The objective of this study is to investigate the effects of the large primary crystals present inside the melt on pore formation in aluminum alloy foams in a semi-solid state. The effects of the diameter and shape of the crystals on pore morphology were examined. To observe the changes caused by the differences in the morphology of primary crystals, A2024 alloy foams were fabricated via the processing of an A2024 master alloy by controlling the formation of primary crystals to favor a fine spherical shape. The fabrication processes used were Strain-Induced Melt Activation (SIMA) [11] and slope casting [12]. Subsequently, we evaluated pore formation in the foam and the formation of primary crystals in the cell wall. In addition, we measured the Ti distribution to trace the path of TiH2, which was the blowing agent used.

The innovations of this study are as follows. First, to systematically derive the effects of diameter, shape, and fraction of solid on primary crystals, the size and shape of the primary crystals were controlled by SIMA and slope casting. These controlling processes were difficult in oxide particle thickening. Second, we investigated fabrication through the stabilization of cell walls by controlled primary crystals.

#### **2. Materials and Methods**

#### *2.1. Fabrication of Master Alloys*

We fabricated three kinds of A2024 master alloys. Table 1 shows the composition of the A2024 alloys used in this study. Figure 1 shows a schematic of the fabrication method of the A2024 master alloys. The A2024 direct chill slab was sliced perpendicular to the pull-out direction of the DC casting. The A2024 plane was cut into several pieces. These pieces will be hereafter referred to as the DC-cast alloys. Then, some of the DC-cast alloys were processed via SIMA. The DC-cast alloys initially measured 20 <sup>×</sup> 30 <sup>×</sup> 30 mm<sup>3</sup> and were compressed to a compressive strain *fs* = 10%, as done in the study presented by Sirong et al. [13]. Then, the pieces were annealed at 350 ◦C for 3 h in a muffle furnace. This master alloy will be hereafter referred to as the SIMA alloy. In addition, some of the DC-cast alloys were processed via slope casting. The A2024 pieces were heated at 649 ◦C and poured onto a cooling plate made of Cu. The degree of the slope, θ, and the contact length, *l*, were 60◦ and 200 mm, respectively, as in a previous study [14]. Afterwards, the alloy was cast; this master alloy will be referred to as the slope-cast alloy.


**Table 1.** Chemical composition of the A2024 alloys.

**Figure 1.** Schematic of the preparation of the master alloys.

#### *2.2. Fabrication of A2024 Alloy Foam in a Semi-Solid State*

Figure 2 shows the fabrication method of the A2024 alloy foam in a semi-solid state. First, 100 g of the DC-cast alloy, SIMA alloy, or slope-cast alloy was placed in a SUS304 crucible coated with Al2O3. Each alloy was heated to and held at 635 ◦C ± 0.3 ◦C (fraction of solid *fs* = 20%) [15] or 630 ± 0.3 ◦C (*fs* = 40%) [15] in an electric furnace, as shown in Figure 2a. The measured temperature was inside the error range. The temperature was measured using a K-type thermocouple covered with an Al2O3 protective tube. After the temperature stabilized, 1 mass% TiH2 was added to the slurry after the thermocouple was taken out. Then, the slurry was stirred using an impeller coated with boron nitride at 900 rpm for 100 s, as shown in Figure 2b. After stirring, the slurry was held for 200 s and the crucible was removed from the furnace, as shown in Figure 2c. Subsequently, the foam was solidified and cooled using 3 L/min of water as shown in Figure 2d, and the A2024 alloy foam was obtained. Six foams were obtained in total, one for each combination of master alloy and fraction of solid. Additionally, to analyze the differences in the primary crystals before stirring and after foaming, metal samples held in a semi-solid state were obtained for each foam.

**Figure 2.** Fabrication method of the A2024 foam in a semi-solid state; (**a**) heating; (**b**) stirring; (**c**) foaming; (**d**) water cooling.

#### *2.3. Evaluation of Pores and Primary Crystal Formation*

To derive the porosity of the foam, we measured the volume of the foam via Archimedes' method using the mass of the foam and a spring scale. From Equation (1), we calculated the porosity *p* of the foam using the density of the A2024 alloy foam ρ*P*, which in turn was calculated using the volume and mass of the foams and the density of the A2024 alloy, ρ*NP* = 2.77 Mg/m3. The pore morphology (average pore diameter *dp* and average pore circularity *ep*) of a cross-section from the middle of a sample was measured using the image-analysis software WinROOFTM version 6.1 (Mitani Corporation, Fukui, Japan). Equations (2) and (3) describe the pore formations, where *S* is the area of the pores and *L* is their perimeter. To improve the accuracy of our measurements, pores smaller than *d* = 0.2 mm were not considered. The diameter of the primary crystals *d*α and their circularity *e*α before and after the foaming of each alloy were also measured using WinROOFTM and calculated via Equations (2) and (3). Every measurement was taken once for every foam.

$$p = (1 - \rho\_p / \rho\_{NP}) \times 100\% \tag{1}$$

$$d\_p = (4 \text{S}/\pi)^{1/2} \tag{2}$$

$$
\varepsilon\_a = 4 \pi \text{S/L}^2 \tag{3}
$$

#### *2.4. Analysis of the Pure Ti Distribution*

We fabricated non-porous metal samples to estimate the distribution of the foaming agent and the foaming sites. These non-porous metal samples were fabricated using a similar procedure to the fabrication method for the A2024 foams, except that pure Ti was used instead of TiH2. A Ti distribution analysis was performed via qualitative mapping using an electron probe micro-analyzer (EPMA, JEOL, JXA-8230, Tokyo, Japan; measurement elements: Ti, Cu; pressurization voltage: 15 kV; irradiation current value: 3 <sup>×</sup> 10−<sup>8</sup> A; irradiation interval: 10 <sup>μ</sup>m; acquisition time: 20 ms/point; measurement range: 5 mm × 3.75 mm). Next, mapping images of Ti and Cu were processed using the following method. First, the Ti mapping image was made monochrome and binarized using WinROOFTM and Ti was colored green. Then, the parts that were not Ti were made transparent using the processing software paint.net Version 4.0.6. (dotPDN LLC, Washington, WA, USA). Finally, the Ti image and the monochrome Cu image were merged. One merged image was obtained for each DC-cast alloy and the SIMA alloy with a fraction of solid *fs* = 40%.

#### **3. Results and Discussion**

#### *3.1. Pore Formation on the A2024 Alloy Foams*

Figure 3 shows the microstructures of the (a) DC-cast, (b) SIMA, and (c) slope-cast A2024 alloys at *fs* = 40%. Figure 4 shows the average values of (a) the primary crystal diameter *d*α and (b) the circularity *e*α of the A2024 alloy foams. For Figure 4, approximately 100 primary crystals were measured for each sample, except for the after-foaming data of the DC-cast alloy, for which 27 primary crystals were measured. Dendritic primary crystals were present before melting and during holding at a semi-solid state in the DC-cast alloy, as shown in Figure 3a. Because some dendritic primary crystals grow while others do not, the diameters *d*α of the primary crystals in the DC-cast alloy have high error values. In the SIMA alloy, shown in Figure 3b, dendritic primary crystals were present before melting, while fine and spherical primary crystals were formed during holding at a semi-solid state. In the slope-cast alloy, shown in Figure 3c, fine and spherical primary crystals were present before melting and during holding at a semi-solid state. Figures 3a and 4 show that the primary crystals were coarse and uneven in the DC-cast alloy before forming and remained coarse but spherical afterwards. The primary crystals in the SIMA and slope-cast alloys maintained their fine and spherical shape, as shown in Figure 3b,c and Figure 4. Therefore, the primary crystals of the DC-cast alloy spheroidized after foaming, whereas those of the SIMA and slope-cast alloys maintained their fine and spherical shape.

**Figure 3.** Images of the A2024 alloys (*fs* = 40%); (**a**) direct chill cast (DC-cast); (**b**) Strain-Induced Melt-Activated (SIMA); and (**c**) slope-cast.

**Figure 4.** Average values of the primary crystals of the A2024 alloy foams fabricated via foaming in a semi-solid slurry (*fs* = 40%); (**a**) diameter *d*α; (**b**) circularity *e*α. The error bars show the standard deviations.

#### *3.2. Macro-Pores of the A2024 Alloys Fabricated in a Semi-Solid State*

Figure 5 shows cross-sections of the A2024 alloy foams of the DC-cast, SIMA, and slope-cast alloys with an *fS* of 20% and 40%. The pores were distributed homogeneously as shown in Figure 5. Although a non-porous part is conventionally found in the bottom part [4], the foams in this study did not show a clear difference of pore distribution at different heights. Therefore, the A2024 alloy foams of each alloy can be fabricated at *fs* values of 20% and 40%. All of the foams except for the slope-cast alloy foam fabricated at *fs* = 40% reached a porosity of 60%. Figure 6 shows the average pore (a) diameter *dp* and (b) circularity *ep* of the A2024 alloy foams with porosities of approximately 60%. In Figure 6, 1000–1900 pores were measured for each sample. The pores of the alloy foams fabricated at *fs* = 20% were finer and more spherical than those of the foams fabricated at *fs* = 40% (Figure 5; Figure 6). However, the differences between the primary crystals of the DC-cast and SIMA alloys had little effect on macro-pore morphology. Although the error value of the primary crystal size of the DC-cast was higher than that of the SIMA alloys, this difference did not affect macro-pore morphology. This result is due to the difference between the sizes of the primary crystals and the macro-pores. In each alloy, most primary crystals had a diameter *d*<sup>α</sup> of less than 300 μm, whereas the diameter of the macro-pores *dp* was larger than 1000 μm. Therefore, differences in the formation of finer primary crystals had little effect on the formation of macro-pores in terms of macro-pore diameter. Moreover, as shown in Figure 6, the diameter of the macro-pores *dp* for a fraction of solid of 40% had high error values. These error values occur because a high fraction of solid makes it difficult for TiH2 to be distributed uniformly. A schematic drawing of this mechanism is shown in Figure 7.

**Figure 5.** Cross-sections of foams: (**a**) direct chill cast (DC-cast); (**b**) Strain-Induced Melt-Activated (SIMA); and (**c**) slope-cast. *p*: porosity.

**Figure 6.** Average values of the pores of the A2024 alloy foams with a porosity of approximately 60%, fabricated via foaming in a semi-solid slurry: (**a**) pore diameter *dp*, and (**b**) pore circularity *ep*. The error bars show the standard deviations. DC-cast: direct chill cast, SIMA: Strain-Induced Melt-Activated.

**Figure 7.** Schematic drawing of the TiH2 distributed in the melt and pores after foaming. (**a**) *fs* = 40%; (**b**) *fs* = 20%.

#### *3.3. Micro-Pores of the A2024 Alloys Fabricated in a Semi-Solid State*

Figure 8 shows the location of the micro-pores in the A2024 alloy foams (*fs* = 40%) for the (a) DC-cast, (b) SIMA, and (c) slope-cast alloys. Figure 9 shows the number of micro-pores per unit area in all the A2024 alloy foams. The *fS* values of the foams were 20% and 40%. In this study, micro-pores are defined as pores in the cell wall with a diameter of less than 1000 μm, which is approximately the thickness of the cell walls. In addition, to improve image processing, pores with a diameter of less than 40 μm were ignored. Mukherjee et al. defined micro-pores as pores less than 350 μm in diameter [16]. Moreover, Ohgaki et al. reported that cracks initiate from pores with a diameter of 30 μm to 350 μm

when an aluminum foam is compressed [17]. However, as shown in Figure 10, most of the micro-pores in this study were inside this range. Therefore, the micro-pores measured in this study corresponded to those in previous studies. In the DC-cast alloy, micro-pores were present inside the primary crystals, as shown in Figure 8a, whereas micro-pores were present between the primary crystals in the SIMA and slope-cast alloys, as shown in Figure 8b. As indicated in Figure 9, the number of micro-pores in the SIMA and slope-cast alloys was approximately 1/3 of that in the DC-cast alloy. As the number of micro-pores increased, the strength of the base metal decreased. However, micro-pores also suppressed densification under compression, as reported by Toda et al. [18].

In addition, Figure 5, Figure 6, Figure 9, and Figure 10 show that there was no major difference in the results of each master alloy except for the number of micro-pores. In contrast, the fraction of solid seems to be the major factor that determines each property in aluminum foam. Furthermore, aluminum foam was fabricated three times for each fraction of solid. Because this foam seems to be reproducible, it was considered unnecessary to repeat the experiment.

**Figure 8.** Locations of micro-pores (*fs* = 40%). (**a**) direct chill cast (DC-cast) alloy; (**b**) Strain-Induced Melt-Activated (SIMA) alloy; and (**c**) slope-cast alloy.

**Figure 9.** Number of micro-pores per unit area, *Np*, in the A2024 alloy foams for the direct chill cast (DC-cast), Strain-Induced Melt-Activated (SIMA), and slope-cast alloys.

#### *3.4. E*ff*ect of Primary Crystals on Micro-Pore Formation*

As described in Section 3.2., the primary crystals were fine and spherical before and after stirring the SIMA and slope-cast alloys. Figure 11 shows the mappings for Ti and Cu obtained with the EPMA for the non-porous metal samples described in Section 2.4. In Figure 11, the darker parts with less Cu are considered to be primary crystals, which are mostly composed of Al. Moreover, Ti is not completely dispersed because the primary crystals exist as solids. For a better observation, Figure 12 shows the merged image made from the Ti and Cu images shown in Figure 11, as mentioned in Section 2.4. In the DC-cast alloy, the eutectic structure and Ti particles co-exist in one primary crystal, as shown in

Figure 12a. This eutectic structure was considered to have been liquid before solidification. In contrast, the liquid phase and the Ti particle exist between the primary crystals in the SIMA alloy, as shown in Figure 12b. From these results, we deduced a reason for the high quantity of micro-pores inside the DC-cast alloy. Figure 13 shows a schematic drawing of the mechanism behind micro-pore formation. As explained in Section 3.1, the uneven primary crystals in the DC-cast alloy spheroidize after foaming. Based on a previous study by Chen et al. [19], primary crystals have been shown to bend via stirring and spheroidize. Therefore, the liquid phase and TiH2 were likely trapped in spheroidized primary crystals, similar to pure Ti, as shown in Figure 12a; for this reason, micro-pores could be found inside the primary crystals in the DC-cast alloy, as shown in Figure 13a.

**Figure 10.** Average diameter of the micro-pores of the A2024 alloy foams fabricated via foaming in a semi-solid slurry for the direct chill cast (DC-cast), Strain-Induced Melt-Activated (SIMA), and slope-cast alloys.

**Figure 11.** Mapping images of Ti and Cu obtained using the electron probe micro-analyzer (EPMA) (*fs* = 40%). (**a**) direct chill cast (DC-cast) alloy; (**b**) Strain-Induced Melt-Activated (SIMA) alloy.

However, when the primary crystals were spherical before foaming, TiH2 rarely got trapped in the primary crystals while stirring, similar to pure Ti as shown in Figure 12b; this would happen in the SIMA and slope-cast alloys. Therefore, micro-pores were present between the primary crystals, as shown in Figure 13b. In addition, micro-pores between the primary crystals easily coalesce with other micro- or macro-pores because they are not completely wrapped by primary crystals. Consequently, the number of micro-pores in the SIMA and slope-cast alloys was 1/3 of that in the DC-cast alloy.

**Figure 12.** Merged images of Ti and Cu obtained using the electron probe micro-analyzer (EPMA) (*fs* = 40%). (**a**) direct chill cast (DC-cast) alloy; (**b**) Strain-Induced Melt-Activated (SIMA) alloy.

**Figure 13.** Schematic of the mechanism behind micro-pore formation in the (**a**) direct chill cast (DC-cast) and the (**b**) Strain-Induced Melt-Activated (SIMA) and slope-cast alloys.

#### **4. Conclusions**

A2024 alloy foams were fabricated in a semi-solid state using three different preparation methods for the master alloys. The most important finding of this study is that the morphology of the primary crystals influences the quantity of micro-pores inside the cell walls, but not on the quantity of macro-pores. The results can be summarized as follows.

1. In the DC-cast alloy, coarse, dendritic, and uneven primary crystals were present before stirring. They were subsequently bent and spheroidized via stirring. On the other hand, the primary crystals in the SIMA and slope-cast alloys maintained their fine and spherical shape before and after stirring.


**Author Contributions:** Writing–reviewing and editing, T.K. and S.S.; conceptualization, M.S. and S.S.; methodology, M.S. and S.S.; validation, M.S. and S.S.; formal analysis, T.K., M.S., and S.S.; investigation, M.S.; data curation, M.S. and S.S.; writing–original draft preparation, T.O.; visualization, M.S.; supervision, S.S.; project administration, S.S.

**Funding:** This study was supported by the Grant-in-Aid program of The Light Metal Educational Foundation.

**Acknowledgments:** The authors thank N. Sakaguchi from UACJ Corporation for suppling the A2024 ingots used in this study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
