1 Introduction

Because aluminum alloy has a series of advantages, such as low density, high specific strength, strong corrosion resistance, easy thermal conductivity, good plasticity, and processing performance, it is always the main material for aviation, transportation, electrical, and other industries [1,2,3,4,5]. At present, high-complex thin-walled precision aluminum alloy castings have been used in aircraft structures, such as the internal reinforcement ribs, webs, particle separator front casing, and intake lip [6]. Since the patent of Hull CW using laser beam deposition of metal powder directly forms large metal parts in the 1980s [7], additive manufacturing (AM) has developed rapidly. AM cannot only achieve better lightweight design and improve the efficiency of materials and energy use, but also manufacture complex mesh and honeycomb structures that are difficult to process by traditional manufacturing methods and satisfy the various requirements [8].

However, the AM parts have many problems such as anisotropy and poor surface quality, making them difficult to meet product performance requirements. The nature of the layered forming of AM leads to a staircase effect on the surface of the formed samples [9]. The schematic diagram of the staircase effect is shown in Fig. 1. In Fig. 1, the red dashed line is the outline of the model, and the black stepped solid line is the outline of actual samples. In addition, balling effect [10] and powder adhesion [11] in AM further exacerbate the variation in the micromorphology and reduce the surface uniformity. The surfaces with high roughness are prone to stress concentration, which in turn induces microcracks, greatly reduces the fatigue performance of parts [12]. On the contrary, the low roughness makes the parts have better friction performance, corrosion resistance, and longer fatigue life [13].

Fig. 1
figure 1

Schematic diagram of staircase effect. t is the powder thickness and α is the forming angle of the surface

Full size image

Advanced surface finishing technology is an effective way to improve the surface quality of complex structural parts.  Beaucamp et al. used the shape adaptive grinding (SAG) to finish the free-form surface of the AM titanium alloy. After three times finishing, the final surface roughness reached below Ra10 nm [14]. Pyka et al. used a combination of chemical and electrochemical polishing to polish the porous structure stent and reduced the surface roughness Ra from 6 ~ 12 μm before polishing to 0.2 ~ 1 μm [15]. They argued that in this process, chemical polishing mainly quickly removes the metal powder adhering to the surface, and then electrochemical polishing further reduces the surface roughness [16]. Can Peng et al. carried out abrasive flow machining (AFM) experiments on AlSi10Mg parts additively manufactured. After AFM, compressive residual stress with the value of − 30.5 ± 7.3 MPa is induced, and the surface roughness reduced from the initial Sa 13–14 μm to the final 1.8 μm [17]. Bhaduri et al. used laser polishing (LP) to polish the AM aluminum alloys in a low-pressure environment with auxiliary ceramic heat insulation substrates. The average roughness Sa is reduced by about 80–88% [18].

Although the above-mentioned methods with various advantages have been used to try to solve the problem of subsequent surface polishing of AM parts, these methods also have some problems. The SAG is only suitable for continuous outer surfaces [12] and difficult to machine narrow grooves and inner surfaces. Chemical and electrochemical polishing are faced with the eternal problem of environmental protection. The cost of LP has been high, and the accessibility of LP to get to complex structures or inner surface is poor because of the nature of light propagation in a straight line. Magnetic abrasive finishing (MAF) has the unparalleled advantage of accessibility on complex inner surfaces, but the uniformity of material removal is difficult to control, and there exists a problem of easy over-cutting at the edges. MAF is an effective processing method to improve the quality of workpiece surfaces, especially the irregular curved surfaces [19]. In the MAF process, the magnetic abrasive particles (MAPs) are bound by the magnetic field force to form a flexible magnetic abrasive brush with certain rigid characteristics. MAF is suitable for the polishing and deburring of complex free-form surfaces, internal cavities, and irregular-shaped parts. There have been precedents of scholars using MAF to polish AM components. Yamaguchi et al. through MAF reduced the roughness of 316L stainless steel formed by SLM exceeding Rz 100 to Rz 0.1 μm and further increased the residual compressive stress on the surface by replacing MAPs with − 50/ + 100 mesh iron particles [20]. Guo et al. through a novel rotating-vibrating magnetic abrasive polishing method to finish a kind of complex internal surface which has a double-layered tube structure made by SLM of Inconel 718. The roughness was reduced from about Ra 7 μm to less than 1 μm [21]. Zhang et al. focused on the rapid removal instead of high precision of MAF SLM 316L stainless steel. They selected G50 steel grains, which with an average diameter of 500 μm and sharp edges, as the abrasive to polish the uneven surface produced by SLM. A maximum decrease of 75.7% in surface roughness Ra was achieved [22]. Zeng et al. carried out MAF experimental research on GH4169 (Inconel 718) alloy fuel injector formed by SLM. The roughness of the detected workpiece drops to 0.47 μm from the original 5.8 μm [23], but the MAPs they use are simple mechanical mixing or sintering abrasives, even only the iron matrix without abrasive phase, which have slow removal efficiency and poor polishing effect. In contrast, the MAPs used in this paper, prepared by gas atomization with rapid solidification, have high bonding strength between abrasive phase and matrix and high sphericity [24]. The morphology of MAPs is shown in Fig. 2. There are no Al2O3 particles in the MAPs, as shown in Fig. 2b, which maximizes the proportion of the iron matrix of MAPs, thereby ensuring that the abrasive is subjected to a greater magnetic field force in the magnetic field.

Fig. 2
figure 2

SEM micrographs of Al2O3/Fe-based MAPs prepared by GARS. (a) MAPs outside morphology. (b) MAPs cross-sectional morphology

Full size image

In this paper, we carry out MAF experiments of AlSi10Mg surfaces prepared by SLM with different forming angels. Our study aims to study the difference in MAF process of parts with different forming angels, and to optimize the process parameters through the response surface methodology (RSM). By using white light interferometer (MicroXAM-100, KLA-TENCOR Corporation, USA), 3D profiler (UP-Dule Mode, RETC, USA), metallurgical microscope (Axio Lab A, Carl Zeiss AG, Germany), and Vickers microhardness tester (DHV-1000Z, Shanghai Optical Instrument Fifth Factory Co., Ltd., China), we discussed the changes of morphology and Vickers hardness before and after MAF.

2 Experimental procedure

2.1 Experimental materials and samples forming

SLM® 125HL (SLM Solutions GmbH, Germany) is selected as the AM forming equipment, as shown in Fig. 3. Figure 4 is the SEM micrograph of AlSi10Mg powder used in our SLM forming process. Implement SLM forming according to the parameters shown in Table 1. The parameters in Table 1 are the optimal parameters determined according to previous experiments [25]. The model of the sample is shown in Fig. 5a, and the blue parts are the supports. The samples formed by SLM are shown in Fig. 5b. Among the six workpieces (forming angle α = 90, 100, 110, 120, 130, 140), α = 90 is the control sample without staircase effect, and α = 140 is the exploratory test piece sample to study the removal ability of MAF for supports.

Fig. 3
figure 3

SLM®125HL equipment

Full size image
Fig. 4
figure 4

SEM micrograph of the AlSi10Mg powder

Full size image
Table 1 SLM AlSi10Mg forming parameters
Full size table
Fig. 5
figure 5

3D model (a) and SLM forming sample (b)

Full size image

At room temperature (27 ℃), by using Archimedes drainage method (Eq. 1) [26], the average density of the sample is measured to be ρsamp = 2.683 ± 0.06 g/cm3. Use the hardness tester to measure the hardness of the samples and the white light interferometer to measure the surface roughness. Considering that the initial roughness and the difference of roughness between different samples are large, which both causing the larger hardness measurement error or even difficulty to see the indentation, therefore, NO.100 sandpaper is used to reduce the surface roughness of the sample to about 2 μm before measuring the hardness. The applied load is 4.903 N, and the dwell time is 15 s. These data are all the average values of five measurements. The initial surface roughness (Ra) of each sample right side is shown in Fig. 6. α = 140 is unable to characterize its true roughness because of the residual support on the surface, which is not shown in Fig. 6.

Fig. 6
figure 6

Initial surface roughness (Ra) of SLM forming samples

Full size image
$${\rho }_{samp}=\left({\rho }_{alco}-{\rho }_{air}\right)\bullet \frac{{m}_{a}}{{m}_{air}-{m}_{alco}}+{\rho }_{air}$$
(1)

where ρalco is the density of absolute ethanol, and ρair is the density of air. At room temperature 27 ℃, ρalco = 0.78352 g/cm3 and ρair = 0.001177 g/cm3. ma is the mass of the sample in the air, and malco is the mass of the sample in absolute ethanol.

2.2 The MAF system

The MAF system applied in the experiments is refitted from the XK7136C CNC milling machine, as shown in Fig. 7. Figure 7a is the physical picture of XK7136C CNC milling machine. Figure 7b is a photo taken during the MAF processing. Figure 7c is a schematic diagram of the modification section of the MAF system. Copper connecting rod is installed on the tool holder to connect the spindle with the magnetic pole. The workpiece is fixed on the workbench by a fixture, and the magnetic pole, after the MAPs is absorbed, is connected to the spindle through the copper connecting rod with two bolts, and then rotates with the spindle. Then, the trajectory control is carried out through the numerical control program to achieve the MAF process. Figure 7e and f are the physical map of the magnetic pole after the abrasive is adsorbed.

Fig. 7
figure 7

The MAF system. (a) Physical diagram of MAF device. (b) MAF processing site. (c) Schematic diagram of the modification section. (e) The magnetic pole. (f) Top view of magnetic abrasive brush

Full size image

2.3 Response surface methodology and experimental setup

For each SLM forming part with different angles, according to the Box-Behnken design (BBD) principle of RSM [27], the spindle speed (A/rpm), feed speed (B/mm∙min−1), and machining gap (C/mm) are independent variables, and the surface roughness (Ra) is the response value Y (α). A three-factor three-level response surface analysis test is performed. There were 17 groups in each experiment, including 5 central points for repeated tests to estimate the error. The test factors and levels are shown in Table 2. Other MAF processing conditions are shown in Table 3. After determining the optimal processing parameters through RSM, the finer MAPs (100–150 mesh) were selected to perform MAF of the samples again. The finer MAPs have the small particle size, meaning a small proportion of the iron matrix, leading to a small magnetic field force. The scratches on the surface of the workpiece caused by finer MAPs are shallower compared with 80–100 mesh abrasive, which can achieve a better polishing effect.

Table 2 Factors and levels of RSM experiments
Full size table
Table 3 MAF process parameters
Full size table

3 Experimental results and discussions

3.1 Experimental results

After the first time MAF, the response value Y (α) of each workpiece at each level is shown in Table 4. The sample with α = 140° is grinded with 600-mesh sandpaper to remove the support before MAF. From the results in the Table 4, the optimal roughness of each sample has little difference, all fluctuating around Ra 320 nm (bold data in Table 4). And for different samples, the best parameters for the MAF effect are A = 1700 n/min, B = 15 mm/min, and C = 2 mm, but the time consumed of MAF is different. The differences in MAF time consumption are shown in Fig. 8. Based on the initial roughness of 90° samples (about Ra 4 μm), we artificially divide the processing time into two parts. The negative semi-axis of the time axis is the time of MAF for the initial roughness to reach about 4 μm, that is, the − T. The positive semi-axis represents the time from 4 μm to the final roughness which no longer decreases, that is, the + T. For the − T of Fig. 8, the differences in time consumption are largely due to the difference in initial surface roughness. For the + T, the time consumed increases as the forming angle increases. With the continuous progress of MAF, the material removal capacity of the brush becomes worse [19], and as the forming angle increases, the influence of the step effect becomes more obvious, result in the rasing in material removal. These factors both lead to an increase in the time consumption of MAF, as shown in Fig. 9. The hardness before and after MAF of each sample with different forming angles is shown in Fig. 10. The hardness of each sample before MAF is different, which is caused by the difference in surface roughness and the difference in micromorphology. After MAF, because the surface roughness of each sample tends to be similar since the surface microscopic morphology is similar, the roughness difference is significantly reduced. This change is what we want to see during the MAF, which shows that MAF can significantly reduce the hardness difference caused by inconsistent forming angles.

Table 4 RSM setup of MAF and the result of Y (α)
Full size table
Fig. 8
figure 8

Time consumed for MAF

Full size image
Fig. 9
figure 9

Schematic diagram of the difference in material removal caused by staircase effect of SLM. (a) α = 140°. (b) α = 100°

Full size image
Fig. 10
figure 10

Hardness change before/after MAF

Full size image

3.2 ANOVA results of RSM

The ANOVA results of the six RSM models are shown in Table 6 in the Appendix at the end of this article. The p-values of the models are all much less than 0.0001, which indicates that the models are statistically significant. And the p values of the primary terms A, B, and C and the secondary terms A2 and C2 are less than 0.05, indicating that these terms have a significant impact on the response value. The p values of the interaction terms AB, AC, and BC are all greater than 0.05, indicating that the impact of the interaction response value is not significant; that is, the three factors have no interaction effect on the response value. The model’s lack-of-fit terms are all greater than 0.05, and the general requirement is greater than 0.1. The p-values ​​of the six models lack-of-fit terms are all greater than 0.1, which is beneficial to the model and indicates that there is no lack of fit factors, so these regression equations can be used instead to analyze the experimental results at the real points of the test. The coefficient of variation (C.V.%) is less than 15%, indicating that there is no abnormal data. The difference between the corrected coefficient of determination R2 (adj) and the predicted coefficient of determination is less than 0.2, and both are greater than 0.8, indicating that the coefficients are valid, and the results of calculations only 2.83%, 1.65%, 2.18%, 3.35%, 2.37%, and 3.26% errors cannot explain the actual model. The coefficient of variation of less than 15% and the coefficient of determination of correction close to unity further indicate that the model fits well and can be used for preliminary analysis of the process of the AlSi10Mg samples with different forming angles of MAF [27, 28].

From the relationship of the mean square error of the single term, the selected three factors have an impact on the response value in descending order: machining gap (C), spindle speed (A), and feed speed (B). The source of the force in MAF is the magnetic field force. If the gap is too large, the magnetic field intensity on the surface of the samples will decrease, which will reduce the magnetic restraint of MAPs, and the flexibility of the abrasive brush will be increase, and the magnetic abrasive brush will lose the finishing ability. If the gap between the pole and sample is too small, the gap can contain a few MAPs, the fluidity of the magnetic abrasive brush becomes poor, and the rigidity is too large. Therefore, the finishing ability of the magnetic abrasive brush can only be guaranteed under a suitable gap. In summary, the machining gap has the greatest impact on the response value Y (α).

Next, we diagnose the reliability of the model from the perspective of visualization, that is, through the normal distribution of residuals, residuals vs predicted and residuals vs actual, as shown in Appendix Table 7. The normal distribution of the residuals, that is, the first column of pictures in Table 7, is basically on a straight line. The residuals vs predicted graph, the second column, has points randomly distributed. The above two indicate that the standardized residual has nothing to do with the predicted value and prove that the fitted quadratic regression equation is effective. The picture in the third column of Table 2 is the comparison between the predicted value and the actual value. The actual value obtained from the experiment and the predicted value of the model are basically on the same straight line, which further shows that the obtained regression equation is highly consistent with the actual situation.

As mentioned earlier, the test coefficients of the interaction terms are all greater than 0.05, indicating that the interaction terms have no obvious influence on the response value. It can also be seen from Appendix Table 8 that the contours of the response surface projection do not appear to be closed and elongated ellipse and there is no very steep change in the response surface, further verifying that the interaction term has basically no effect on the response value.

$$Y\left(\alpha \right)={\beta }_{0}+{\beta }_{1}\bullet A+{\beta }_{2}\bullet B+{\beta }_{3}\bullet C+{\beta }_{4}\bullet AB+{\beta }_{5}\bullet AC+{\beta }_{6}\bullet BC+{\beta }_{7}\bullet {A}^{2}+{\beta }_{8}\bullet {B}^{2}+{\beta }_{9}\bullet {C}^{2}$$
(2)

The coefficients of the equations in terms of coded established by RSM are shown in Table 5. The coded equations in the form of Eq. 2 are useful for identifying the relative impact of the factors by comparing the factor coefficients [29]. Among them, the p values of quadratic terms AB, BC, AC, and B2 are all greater than 0.05, which is not significant, so these terms are finally discarded. Finally, Eq. 3 is obtained.

Table 5 RSM model coefficients
Full size table
$$Y\left(\alpha \right)=0.325+0.012 \bullet A+0.008\bullet B-0.023 \bullet C+0.067\bullet {A}^{2}+0.044\bullet {C}^{2}$$
(3)

In summary, the optimal processing parameters of each angle forming sample obtained through the software are not very different. Combined with the actual situation of the processing system, rounded to A = 1700 n/min, B = 15 mm/min, and C = 2.2 mm. After determining the optimal processing parameters, we replaced the finer-grained abrasive for the second MAF of the workpiece and finally reduced the surface roughness to below 100 nm. The specific values and errors are shown in Fig. 11. Compared with the initial roughness (Ra), not only the numerical value is significantly reduced, but the uniformity of roughness value is also significantly improved. This is the same as the change in hardness, which shows that MAF can significantly reduce the roughness difference caused by inconsistent forming angles.

Fig. 11
figure 11

Surface roughness (Ra/nm) and error bar after fine MAF

Full size image

When processing the same material, the processing parameters have little effect on the forming samples with different forming angles. To reduce the difficulty of programming and reduce the program statements, we can consider using the same parameters to process the forming surfaces with different forming angles. However, the processing time is different, which can be considered when writing the CNC machining program.

3.3 Morphology analysis before and after MAF

Figure 12 shows the morphology of initial, rough MAF, and fine MAF. Because the initial surface was too rough to take pictures by the white light interferometer, the RTEC 3D profiler was used to observe the morphology of initial surface. The roughness after MAF is significantly reduced, so the white light interferometer with higher precision is used to analyze the morphology. From the 3D micromorphology, it can be seen more intuitively that the surface roughness has been greatly reduced, and the surface uniformity has also been significantly improved. Figure 13 shows the finely ground surface of the SLM sample observed through the metallurgical microscope. Figure 14 include photos of the surface of the initial, rough, and fine MAF. After fine MAF, the improvement of the mirror effect is very significant. What needs additional explanation here is the picture on the left side of the second row of Fig. 12f. This picture shows the surface of the MAF without removing the support. The support has not been completely removed. This is because the flexible magnetic abrasive brush is adaptive; it is difficult to effectively remove particularly sharp residual supports. Figure 14(f2.2) is the MAF after grinded with 600-mesh sandpaper. Compared with Fig. 14(f2.1), it has a foggy mirror effect. Therefore, for surfaces with supports, to improve the quality and efficiency of processing, other methods should be used in advance to remove the remaining supports.

Fig. 12
figure 12

3D morphology of the SLM samples (from top to bottom, X1, X2, and X3 correspond to the sample surface of initial, rough MAF, and fine MAF, respectively). (a) α = 90°, (b) α = 100°, (c) α = 110°, (d) α = 120°, (e) α = 130°, (f) α = 140°

Full size image
Fig. 13
figure 13

Morphologies of the samples observed by the metallurgical microscope. (a) α = 90°, (b) α = 100°, (c) α = 110°, (d) α = 120°, (e) α = 130°, (f) α = 140°

Full size image
Fig. 14
figure 14

Physical picture of Mirror effect (from top to bottom, X1, X2, and X3 correspond to the sample surface of fine finishing, rough finishing, and initial, respectively. In (f), that is α = 140°, the picture on the right (f2.2) of the rough finishing surface is MAF after removing the support with sandpaper; the left (f2.1) is directly MAF.). (a) α = 90°, (b) α = 100°, (c) α = 110°, (d) α = 120°, (e) α = 130°, (f) α = 140°

Full size image

4 Conclusions

  1. 1.

    The results in this study show that MAF can effectively improve the surface quality of the AM sample, the optimal surface parameters of the sample with different forming angles are not much different, but different processing time need to be set for different forming angles to ensure processing surface uniformity.

  2. 2.

    MAF can significantly reduce the differences in micromorphology, roughness, hardness, and other aspects caused by different forming angles before processing and effectively improve the surface uniformity of the AM samples.

  3. 3.

    It has been verified by experiments that MAF has limited support removal ability in SLM forming. It is necessary to adjust the position of the forming part to reduce the contact area between the support and the forming part or to remove the surface support through pretreatment.