Layer-Resolved Grain Morphology and Recrystallized Crystal Evolution in FSP-Assisted Wire Arc Additive Manufacturing of Aluminum Alloy 4043

Abstract

Wire arc additive manufacturing of aluminum generates coarse, anisotropic solidification microstructures that limit mechanical performance, and interlayer friction stir processing (FSP) is increasingly applied to refine them. This study reports the layer-resolved grain morphology and the recrystallized crystal evolution in MIG + FSP-fabricated aluminum alloy 4043 walls, pairing the FSP spindle torque recorded from the CNC controller with multi-descriptor grain morphology in a coupling that, to the authors’ knowledge, has not been previously reported in the WAAM + FSP literature. Methodologically, two four-bead, three-layer walls were co-fabricated under identical deposition conditions on a HAAS VF-3 CNC platform, one by MIG deposition alone and one by the complete MIG + FSP route; the FSP spindle torque was measured at three positions per layer (118 ± 6 N·m at 600 RPM for L1, and 19.1 ± 1.0 and 26.6 ± 1.3 N·m at 1200 RPM for L2 and L3), and quantitative image analysis of 10,091 grains provided the layer-resolved mean grain area, equivalent diameter, aspect ratio, perimeter-to-area ratio, and circularity. The results show that the mean grain area increased from 8.55 μm² (L1) to 12.96 μm² (L3) while the aspect ratio decreased monotonically (1.389 to 1.323), indicating progressive grain equiaxiality with build height; the P/A ratio followed a non-monotonic layer dependence (2.54 to 2.11 to 2.50 μm⁻¹), with the L2 minimum consistent with reduced boundary line density under the combined thermal influence of two adjacent FSP events. The MIG + FSP route produced grain areas 29–48× smaller per layer than the MIG wall and a 45.8% higher hardness (75.8 ± 7.7 versus 52.0 ± 1.3 HV; n = 6; p = 0.0027). In conclusion, the L3 torque exceeds the L2 torque at equal 1200 RPM, qualitatively consistent with the d^−p term in the grain-size-explicit creep framework $\stackrel{.}{\gamma }$ = C·(τⁿ/d^p)·exp(−Q/RT), although temperature, strain rate, and grain size cannot be fully decoupled from the present three-layer dataset. The morphology and the distributional evidence are consistent with dynamic recrystallization (DRX); discrimination between continuous and discontinuous DRX requires EBSD.

1. Introduction

The wire arc additive manufacturing (WAAM) process has attracted remarkable attention because of its cost-effectiveness and the ability to fabricate large aluminum components through layer-by-layer arc deposition. However, the process inherently generates coarse, dendritic, and anisotropic microstructures resulting from repetitive melting and solidification cycles [1,2]. The poor thermal control that is characteristic of the arc-based deposition promotes progressive interlayer heat accumulation, which drives Ostwald ripening and dendritic grain alignment along the prevailing thermal gradient. In a prior study, the mean grain areas in four-bead, three-layer AA6061/ER4043 walls reached 292–420 μm² per layer [3]. These coarse, irregular grains can reduce the grain boundary density and act as stress concentration sites, which can degrade the fatigue resistance and reduce the isotropy of the mechanical response [4,5].

Among the severe plastic deformation (SPD) techniques, friction stir processing (FSP) is widely recognized for its ability to disrupt the dendritic solidification structures through intense shear deformation and frictional heating below the solidus temperature, which promotes dynamic recrystallization (DRX) and produces fine, equiaxed grain structures [6,7,8]. When integrated into the WAAM workflow as interlayer FSP, the combined MIG + FSP process can simultaneously achieve the geometric freedom of additive manufacturing and the microstructural refinement of solid-state deformation [9]. The grain refinement capability of the interlayer FSP in WAAM has been demonstrated [10,11,12], and FSP spindle torque monitoring has been employed for in-process diagnostics in FSP and FSW contexts outside the WAAM context. However, to the authors’ knowledge, the FSP spindle torque available from the CNC controller has not been paired with layer-resolved grain size measurements in any published WAAM + FSP study. This pairing is the experimental foundation of the microstructure-based creep equation $\stackrel{.}{\gamma }$ = C·(τⁿ/d^p)·exp(−Q/RT) [9], which, in principle, can be inverted for predicting the grain size d from the measured torque. The constitutive framework is derived in [9], and the parameter ranges in the present manuscript are drawn from the primary creep literature; the individual parameters (C, n, p, Q) are not fitted from the present dataset.

Beyond the torque–grain size coupling, the published WAAM + FSP studies have characterized grain morphology primarily through the mean grain area, leaving the richer descriptor space (the aspect ratio, the perimeter-to-area ratio, the circularity, and the distribution shape) unexploited [10,11,12,13,14]. The perimeter-to-area ratio (P/A) is particularly informative as a proxy for the grain boundary energy density and the stored deformation energy [15,16], and its non-monotonic evolution across the layers encodes the asymmetric thermomechanical history of the MIG + FSP build stack in ways that the mean grain size alone cannot reveal. Two questions remain unresolved in this field. First, the recrystallization mechanism operative during FSP of aluminum alloys is contested: different studies have attributed grain refinement under nominally similar Al–Si compositions and processing windows to continuous dynamic recrystallization through progressive sub-grain rotation, to discontinuous dynamic recrystallization through boundary bulging, and to recovery-dominated and geometric mechanisms, and no consensus exists for the layer-resolved WAAM + FSP case [17,18,19,20]. Second, although in-process force and torque signals have been used for diagnostics in FSP and FSW, whether such signals carry sufficient microstructural information to support closed-loop control of layer-resolved grain size in hybrid additive manufacturing has not been established. The current state of the field can therefore be summarized as follows: the mean grain area dominates existing WAAM + FSP characterization, while multi-descriptor morphology and the in-process torque–grain size linkage remain largely unexploited, and the controlling recrystallization mechanism remains open pending crystallographic analysis. The objectives of this study are (i) to measure the FSP spindle torque per layer and pair it with the layer-resolved grain size; (ii) to provide a multi-descriptor grain morphology analysis across 10,091 grains; (iii) to compare the layer-resolved MIG and MIG + FSP microstructures quantitatively with representative optical micrographs; and (iv) to interpret the torque–grain size relationship within the grain-size-explicit constitutive framework.

2. Materials and Methods

2.1. Materials

The substrate material used in this study was an AA6061-T6 aluminum alloy (McMaster-Carr, Elmhurst, IL, USA), machined to dimensions of 152 × 102 × 12.7 mm. The filler material was ER4043 aluminum wire (Al–4.5–6.0 wt.% Si, 0.9 mm diameter, Blue Demon, Welding Material Sales, Inc., Geneva, IL, USA). The nominal chemical compositions of the substrate and the filler wire are given in

Table 1. Nominal chemical compositions of substrate (AA6061) and filler (ER4043) in wt.%.

Material	Mg	Si	Cu	Cr	Fe	Mn	Zn	Al
AA6061	0.8–1.2	0.40–0.80	0.15–0.40	0.04–0.35	≤0.70	≤0.15	≤0.25	Bal.
ER4043	≤0.05	4.5–6.0	≤0.30	n.s.	≤0.80	≤0.05	≤0.10	Bal.

. The FSP tool was made of H13 hot-work tool steel, with a shoulder diameter of 18 mm, a pin diameter of 5.4 mm, and a plunge depth of 0.2 mm.

2.2. Wall Fabrication

In this study, two four-bead, three-layer walls were co-fabricated in a single experimental campaign under identical MIG deposition conditions on a HAAS VF-3 CNC machining center (Oxnard, CA, USA); one wall was fabricated by MIG deposition only, and the other wall was fabricated by the MIG + FSP process. The MIG parameters used for both walls were as follows: U = 18 V, I = 120 A, v = 330 mm/min, with argon (Airgas, Inc., Radnor, PA, USA) shielding at 27 CFH. Interpass cooling to 38–40 °C, which was monitored by an infrared pyrometer, was applied between the successive layers for both walls [3]. For the MIG + FSP wall, the FSP was applied after each layer cooled to the interpass temperature, with a plunge depth of 0.2 mm, a traverse speed of 50 mm/min, and a spindle speed of 600 RPM (Layer 1) and 1200 RPM (Layers 2 and 3). The spindle speed increase for the upper layers compensates for the increased heat dissipation with the build height [3,9]. The processing parameters are summarized in

Table 2. Processing parameters for MIG and MIG + FSP wall fabrication.

Parameter	MIG Stage (Both Walls)	FSP Stage (MIG + FSP Only)
Voltage/Current	18 V/120 A
Travel speed	330 mm/min	50 mm/min
Spindle speed	Wire feed (current-controlled)	L1: 600 RPM; L2–3: 1200 RPM
Plunge depth		0.2 mm
Interpass temperature	Cooled to 38–40 °C (IR pyrometer)	After cooling to 38–40 °C (IR pyrometer)
Layers/beads per layer	3 layers, 4 beads, 50% overlap	3 FSP passes (one per layer)

. The process flow is shown schematically in Figure 1, the as-built MIG + FSP wall is shown in Figure 2, and the three FSP torque-sampling positions (P1, P2, P3) per layer are indicated in Figure 3.

2.3. FSP Torque Measurement

The FSP spindle torque was recorded from the CNC controller at three predefined positions per layer—110–115 mm from start (L1), 80–85 mm (L2), and 50–55 mm (L3)—each sampled over a 5 mm traverse window (n = 3 per layer). The measurement positions shift inward with the build height because the effective bead length decreased with the build height due to minor geometric tapering that was observed in the fabricated wall; all sampling windows fall within the steady-state processing region, away from the entry transients and end-of-bead effects. The mean spindle power (kW) and the mean torque (N·m) were extracted from the controller data log. The reported values represent the steady-state averages over the 5 mm sampling window; the torque variability within each window is less than ±5% of the mean value, based on the CNC log resolution.

2.4. Microstructural Characterization

The specimens were prepared following standard metallographic procedures, including hot mounting, sequential grinding (240–1200 grit), diamond polishing (6 and 1 μm), final colloidal silica polishing (0.05 μm), and etching with Keller’s reagent. For the microstructural analysis, nine optical micrographs per layer (three fields at three positions) were acquired by employing a VK-9700 laser scanning microscope (Keyence Corporation, Osaka, Japan) and processed with MIPAR Image Analysis software version 4.2 (MIPAR Image Analysis, Worthington, OH, USA) for grain boundary segmentation (full details of the image-analysis and segmentation pipeline are given in Supplementary Note S2), yielding N = 10,091 grains (MIG + FSP wall) and N = 9744 grains (MIG wall). After the removal of the duplicate rows (identical grain entries arising from overlapping micrograph field boundaries) from the MIG + FSP CSV datasets, per-layer populations of N = 2109 (L1), 2166 (L2), and 5816 (L3) were used for the statistical analysis. The grain descriptors that were computed include area A (μm²), perimeter P (μm), equivalent diameter $\stackrel{-}{D}$_eq = 2√(A/π) per ASTM E112 [22]. A full list of symbols and abbreviations is provided in Table S1. Two representations of equivalent diameter are reported: D(A) = 2√(A/π), the diameter of a grain with mean area (

Table 4. Layer-resolved grain area and equivalent diameter: MIG vs. MIG + FSP walls.

Layer	MIG Area (μm2)	MIG $\stackrel{\mathbf{-}}{\mathit{D}}$eq (μm)	MIG + FSP Area (μm2)	MIG + FSP D(A) (μm)	Refin. (×)	FSP Torque (N·m)
L1	292.10	19.29	8.55	3.30	34.2×	118 ± 6
L2	419.95	23.12	8.82	3.35	47.6×	19.1 ± 1.0
L3	375.84	21.88	12.96	4.06	29.0×	26.6 ± 1.3

, used in the torque–grain size analysis), and $\stackrel{-}{D}$_eq = mean(2√(A_i/π)), the arithmetic mean of the individual grain diameters. The two conventions are not interchangeable. For a right-skewed grain-size distribution, the arithmetic mean of the individual diameters is smaller than the diameter computed from the mean area, that is $\stackrel{-}{D}$_eq < D(A), because the square-root function is concave (Jensen’s inequality). The remaining per-grain descriptors are defined as follows: the aspect ratio AR is the ratio of the major to the minor axis of the best-fit ellipse; the circularity is C = 4πA/P², which equals unity for a perfect circle and decreases for irregular or elongated grains; and the perimeter-to-area ratio P/A (μm⁻¹) quantifies the boundary line length per unit grain area. For the representative optical micrographs presented in the Results, CLAHE (contrast-limited adaptive histogram equalization) with subsequent unsharp masking was applied to improve grain boundary visibility, following standard practice in quantitative metallography [23].

2.5. Constitutive Framework

The grain-size-explicit creep equation, derived from the classical hot deformation literature [24,25,26,27] and implemented for the unified additive–deformation manufacturing process in [9], provides the physical basis for the torque–grain size coupling:

$$\stackrel{.}{\gamma }=C·({\tau }^n/d^p)·\exp (-Q/RT)$$

(1)

For Al and Al–Si alloys under FSP-relevant conditions (T ≈ 300–400 °C, $\stackrel{.}{\epsilon }$ ≈ 1–100 s⁻¹), the dominant rate-controlling mechanism is climb-controlled power-law dislocation creep: n ≈ 4–7 [24,25,27], p ≈ 2–3 for diffusion-assisted dislocation creep, and Q ≈ 130–145 kJ·mol⁻¹ for Al–Si alloys [27,28]. The parameter ranges and their sources are summarized in Table S3. The strain rate sensitivity m = 1/n ≈ 0.14–0.25 yields (m + 3) ≈ 3.14–3.25, and C is a pre-exponential constant whose units absorb dimensional consistency. In the FSP geometry, the Fields–Backofen torsion correction [29] is applied as a first approximation for converting the measured spindle torque T_q to a surface shear stress, τ_s = T_q·(m + 3)/(2π·R_s³), with shoulder radius R_s = 9.0 mm. The original relation was derived for solid-cylinder hot torsion, whereas the FSP strain field is three-dimensional with a shoulder-induced compressive component; the conversion is used here for order-of-magnitude estimation rather than absolute calibration. Inversion of Equation (1) yields d as an explicit function of T_q, T, and rotation speed, which provides a basis for the post-pass prediction of grain size from CNC controller data. The symbols in Equation (1) are defined as follows: $\stackrel{.}{\gamma }$ is the shear strain rate (s⁻¹); C is a pre-exponential constant whose units absorb dimensional consistency; τ is the shear stress (MPa); n is the stress exponent (dimensionless); d is the grain size (μm); p is the grain-size exponent (dimensionless); Q is the activation energy of the rate-controlling process (kJ·mol⁻¹); R is the universal gas constant (8.314 J·mol⁻¹·K⁻¹); and T is the absolute temperature (K). In the Fields–Backofen relation, T_q is the measured spindle torque (N·m), m is the strain rate sensitivity (m = 1/n), and R_s is the shoulder radius (9.0 mm). Because the parameters C, n, p, and Q are drawn from the primary creep literature rather than fitted to the present dataset, the sensitivity of the inverted grain size to each parameter is assessed qualitatively from the inverted form d ∝ (τⁿ/$\stackrel{.}{\gamma }$·exp(−Q/RT))^(1/p). The predicted grain size is most sensitive to the grain size exponent p, which appears as the inversion root, followed by the stress exponent n and the activation energy Q through the exponential temperature term. A quantitative sensitivity surface and a calibrated inversion require an experimental matrix that spans multiple rotational speeds and in situ temperatures, which is identified as future work. The Fields–Backofen relation assumes a one-dimensional torsional surface shear; in the FSP geometry, the shoulder contributes a compressive normal component, and the strain rate varies radially as $\stackrel{.}{\epsilon }$ ∝ ωr. A scaling comparison of the pure torsion shear stress with an estimate that includes the shoulder-normal contribution indicates that the present conversion underestimates the effective stress by a factor of order 1.5 to 2, which is reported here as the order-of-magnitude bound on τ_s.

2.6. Sampling and Repeatability

Two walls were fabricated in a single co-processing campaign under identical MIG deposition conditions: one MIG-only wall and one MIG + FSP wall. The FSP spindle torque was sampled at three windows per layer (n = 3 per layer) within the steady-state processing region, away from entry transients and end-of-bead effects, with intra-window variability below ±5% of the mean. For microstructural characterization, nine to fifteen optical micrograph fields per layer were segmented, yielding per-layer grain populations of N = 2109 (L1), 2166 (L2), and 5816 (L3) after removal of duplicate entries arising from overlapping micrograph field boundaries. The large per-layer grain counts provide the statistical basis for the distributional analysis and quantify repeatability at the level of grain population statistics. The present design therefore establishes repeatability through grain population and torque window statistics rather than through wall-to-wall replication; the absence of replicate walls is stated explicitly in the Limitations section.

2.7. Statistical Analysis

The conformance of the per-layer grain area populations to a log-normal distribution was tested using the one-sample Kolmogorov–Smirnov (KS) test applied to the natural logarithm of grain area. The KS test was selected because it is a distribution-free test based on the maximum absolute difference between the empirical and the fitted cumulative distribution functions (the D statistic), which yields a single scale-independent measure of departure from log-normality that is robust for large, right-skewed populations. The Anderson–Darling test, by contrast, is a weighted quadratic statistic that places greater weight on the distribution tails. This tail sensitivity is the reason the KS test was used here to address the overall conformance of the bulk population to log-normality, while the Anderson–Darling test, together with quantile–quantile plots and finite-mixture analysis, is reserved as the appropriate follow-up for resolving the L3 grain-growth tail and the suspected L1 bimodality, which require crystallographic context to interpret. Because the KS test gains statistical power with increasing sample size, the per-layer D statistic is reported alongside the p-value so that the practical magnitude of any departure can be assessed independently of the count imbalance among layers (L1, L2, and L3). The significance of the hardness difference between the two walls was assessed using the Mann–Whitney U test, a non-parametric test appropriate for the small per-wall sample size. Full statistical details are provided in Supplementary Note S3.

2.8. Mechanical Characterization

Vickers microhardness was measured under a 200 gf load (HV0.2) with n = 6 indents per wall on the co-fabricated walls. The indent-to-grain-size ratio was considered in interpreting the scatter: the Vickers indent at 200 gf is approximately 50 μm in diagonal, which is comparable to the MIG grain diameter (D(A) ≈ 19–23 μm) but much larger than the MIG + FSP grain diameter (D(A) ≈ 3.3–4.1 μm), so each indent in the MIG + FSP wall averages over many grains. The global per-wall sampling documents a process-level hardness comparison and does not resolve layer-by-layer hardness variation; layer-resolved hardness mapping is identified as a goal of future work.

3. Results

3.1. FSP Torque per Layer

Table 3. FSP spindle torque per layer in the MIG + FSP wall (n = 3 per layer).

Layer	Spindle Speed (RPM)	Spindle Power (kW)	Torque (N·m)	Measurement Location
L1	600	7.4	118 ± 6	110–115 mm from start
L2	1200	2.4	19.1 ± 1.0	80–85 mm from start
L3	1200	3.3	26.6 ± 1.3	50–55 mm from start

presents the measured FSP torque per layer. Layer 1 (600 RPM) required a substantially higher torque (118 N·m) than Layers 2 and 3 (1200 RPM). Within the 1200 RPM group, Layer 3 showed approximately 39% higher torque than Layer 2 (26.6 vs. 19.1 N·m). The L2-to-L3 intra-RPM contrast is interpreted as directional evidence for the inverse grain-size dependence in Equation (1) rather than as a quantitative test of the creep framework; a fuller decoupling of temperature, strain rate, and grain size contributions is discussed in Section 4.1 and Section 4.5. Figure 4 presents the FSP torque plotted against the layer-resolved equivalent grain diameter D(A).

3.2. MIG vs. MIG + FSP: Layer-Resolved Grain Area

Table 4 presents the layer-resolved grain area and equivalent diameter D(A) for both walls. The MIG wall showed a progressive coarsening from L1 to L2 (+43.8%) and a partial reduction at L3 (−10.5%), which can be attributed to the asymmetric interpass heat accumulation along the build direction. In contrast, the MIG + FSP wall maintained near-uniform grain areas in L1 and L2 (8.55 and 8.82 μm²), with a 51.6% increase at L3 (12.96 μm²) caused by cumulative thermal exposure without a subsequent FSP pass above it. The grain area refinement ratios ranged from 29.0× (L3) to 47.6× (L2), in line with the 10–50× range previously reported in WAAM + FSP studies of Al alloys [10,11,12]. Figure 5 shows the layer-resolved mean grain area for both walls.

Both walls were fabricated in the same experiment; MIG microstructural data was previously reported in [3]. The MIG wall grain segmentation (N = 9744 grains) was performed for grain area quantification; multi-descriptor morphological analysis (aspect ratio, circularity, P/A ratio) was applied exclusively to the MIG + FSP wall, where the DRX-produced equiaxed grain populations are the relevant comparison targets. The MIG + FSP per-layer grain counts (L1: 2109; L2: 2166; L3: 5816) reflect additional micrograph fields captured at L3 (15 fields vs. 9 for L1 and L2) to ensure adequate spatial coverage of the wider stir zone at the top of the wall, where the processed region expands with cumulative build height. All micrographs were acquired at the same magnification (50×); the higher L3 count should be considered when interpreting KS test sensitivity across layers. MIG + FSP data are derived from quantitative image analysis (post-deduplication). Refinement = MIG area/MIG + FSP area.

3.3. MIG + FSP Multi-Descriptor Layer-Resolved Grain Morphology

For the quantification of the microstructural observations,

Table 5. Layer-resolved grain morphology statistics for the MIG + FSP wall.

Layer	N	A (μm2)	σA (μm2)	$\stackrel{\mathbf{-}}{\mathit{D}}$eq (μm)	AR	Circ. $\stackrel{\mathbf{-}}{\mathit{C}}$	Mean (P/A) Ratio (μm−1)	$\stackrel{\mathbf{-}}{\mathit{P}}$ (μm)
L1	2109	8.55	7.90	2.971	1.389	0.579	2.541	13.22
L2	2166	8.82	8.13	3.059	1.353	0.569	2.105	13.34
L3	5816	12.96	15.65	3.427	1.323	0.612	2.502	14.47

A = mean grain area; σA = standard deviation of grain area; $\stackrel{-}{D}$eq = arithmetic mean equivalent grain diameter, where Deq = 2√(A/π); AR = mean aspect ratio; Circ. = mean circularity, where C = 4πA/P²; Mean P/A Ratio = mean perimeter-to-area ratio; $\stackrel{-}{P}$ = mean grain perimeter. $\stackrel{-}{D}$eq differs from 2√(A/π) in Table 4 because the former represents the arithmetic mean of individual equivalent grain diameters, whereas the latter is calculated directly from the mean grain area. The difference arises from Jensen’s inequality for right-skewed distributions. Ψ = normalized boundary complexity index, defined as (P/A)√A (μm^−0.5); Ψ values are 7.43 for L1, 6.25 for L2, and 9.01 for L3.

summarizes the layer-by-layer morphological statistics of the MIG + FSP wall. The consolidated statistics are also tabulated in Table S2. The mean grain area increased from 8.55 μm² in L1 to 12.96 μm² in L3. The standard deviation also increased from 7.90 to 15.65 μm², indicating a broader grain size distribution in the upper layers because of cumulative thermal cycling during deposition. The aspect ratio decreased from 1.389 in L1 to 1.353 in L2 and 1.323 in L3, indicating progressively more equiaxed grains with increasing build height. The circularity followed a non-monotonic trend, decreasing from 0.579 to 0.569 and then increasing to 0.612 in L3. The higher circularity in L3 is consistent with grain boundary smoothing during continued recrystallization under repeated thermal exposure [15,17]. Figure 6 shows the layer-by-layer evolution of (a) aspect ratio, (b) circularity, and (c) P/A ratio in the MIG + FSP wall.

3.4. P/A Ratio: Non-Monotonic Layer Evolution

The mean perimeter-to-area ratio followed a non-monotonic trend across the layers, being 2.541 μm⁻¹ in L1, decreasing to 2.105 μm⁻¹ in L2, then increasing to 2.502 μm⁻¹ in L3. Because the mean grain area is essentially constant between L1 and L2, the approximately 17% P/A reduction at L2 reflects a reduction in boundary line density at fixed grain size. Two interpretations are admissible from the present two-dimensional descriptors and cannot be distinguished without crystallographic data: a genuine reduction in boundary line density through recovery, or a change in the grain-shape distribution at constant mean area. The recovery interpretation is consistent with the thermal history, since L2 is sandwiched between two thermal–FSP events and would therefore undergo more complete boundary recovery than L1 (no thermal cap above) or L3 (single FSP exposure with no subsequent annealing); confirmation would require EBSD-based misorientation analysis. The absolute ⟨P/A⟩ values depend on the segmentation pipeline (Section 2.4); the layer-to-layer relative trend is robust within this pipeline. The asymmetric P/A evolution indicates that the boundary energy density at each build height cannot be inferred from the mean grain size alone.

3.5. Grain Size Distributions: Log-Normal Analysis by Layer

The Kolmogorov–Smirnov results (

Table 6. Log-normal distribution parameters for MIG + FSP grain area per layer.

Layer	ln(A) Mean (μ)	ln(A) Std σ	KS Stat.	KS p-Value	N	Interpretation
L1	1.629	1.258	0.099	<0.001	2109	Non-log-normal: possible bimodal DRX/parent grain mixture near substrate interface
L2	1.796	0.906	0.016	0.601	2166	Log-normal: statistically unimodal population (CV = 92%, reflecting inherent grain size variability in DRX)
L3	1.746	1.505	0.064	<0.001	5816	Non-normal: grain growth tail broadens distribution

; test described in Section 2.7) show that Layer 2 conformed to a log-normal distribution (KS p = 0.601, μ = 1.796, σ = 0.906), consistent with a unimodal grain area population. Layers 1 and 3 deviated significantly (p < 0.001). The L1 deviation is consistent with a bimodal or mixed population, and the L3 deviation is consistent with a grain growth tail. The interpretation rests on the magnitude of the D statistic and the distribution shape rather than on the p-value alone: at N = 5816, the L3 D statistic of 0.064 indicates only a small practical departure from log-normality despite the highly significant p-value, which reflects the increased power of the test at large sample size. The qualitative conclusion that Layer 2 is the sole log-normal layer is therefore robust to the per-layer count imbalance. Statistical conformity to a log-normal form does not by itself identify a recrystallization mechanism, because dynamic recrystallization, normal grain growth, and other steady-state coarsening processes can all produce approximately log-normal distributions in their stationary regimes. The grain-area probability density distributions with log-normal fits are shown in Figure 7, the cumulative distribution functions per layer in Figure 8, and the aspect ratio distributions by layer in Figure 9.

3.6. Perimeter–Area Power Law

The perimeter–area relationship for the 10,091 grains measured in the MIG + FSP wall followed $P=\alpha A^{0.734}$ with an $R^2$ value of 0.92, corresponding to a fractal dimension of $D_f=1.47$. This value falls between the smooth boundary limit ($D_f$ close to 1) and the irregular space-filling limit ($D_f$ close to 2). This result is consistent with partial dynamic recrystallization together with second-phase particle pinning in the ER4043 Al–Si alloy. Fractal analysis based on two-dimensional metallographic sections is affected by factors such as segmentation threshold selection, polishing and etching quality, image resolution, and CLAHE preprocessing. The D_f value was computed globally across all three layers and reflects the population-weighted average of the competing influences of DRX-induced boundary serration and thermal smoothing under the present MIPAR segmentation pipeline. The same parameters were applied to all layers and both walls, so the qualitative reading of the boundary roughness regime is internally consistent; the absolute D_f value should be compared across studies only when the segmentation conventions are matched. The Si particles (4.5–6.0 wt.% Si) contribute through Zener pinning (P_Z = 3 f γ_gb/2 r), which impedes boundary migration and may preserve the serrated boundary morphology. Figure 10 presents the perimeter–area power law on log–log axes.

3.7. Mechanical Properties

The Vickers microhardness results (method described in Section 2.8) are summarized in Figure 11. The MIG wall exhibited a hardness of 52.0 ± 1.3 HV, while the MIG + FSP wall achieved 75.8 ± 7.7 HV, which corresponds to a 45.8% improvement (Mann–Whitney U test, p = 0.0027). The hardness was measured on the co-fabricated walls described in [3]. The higher scatter in the MIG + FSP wall (SD = 7.7 versus 1.3 HV) is consistent with two contributions. The first is the layer-resolved grain size variation (8.55–12.96 μm²), where L3 coarsening reduces the local Hall–Petch contribution, and in part the ratio of indent size to grain size. The second is the Vickers indent at 200 gf, which is approximately 50 μm in diagonal, comparable to the MIG grain diameter (D(A) ≈ 19–23 μm) but much larger than the MIG + FSP grain diameter (D(A) ≈ 3.3–4.1 μm), so the MIG hardness averages over many grains per indent and is inherently smoother. The global sampling (n = 6 per wall) does not resolve the layer-by-layer hardness variation, and the present sampling density is insufficient for testing the Hall–Petch slope across layers. The present data therefore document a process-level hardness improvement attributable to grain refinement and second-phase dispersion rather than a per-layer test of the Hall–Petch relationship. Layer-resolved hardness mapping with n ≥ 6 indents per layer at standardized sub-bead positions, complemented by nanoindentation arrays for resolving the per-grain Hall–Petch slope and by tensile testing for quantifying the per-layer yield strength contribution, is identified as a priority for future work. Figure 11 presents the Vickers microhardness results for both walls.

3.8. Representative Optical Micrographs

The optical micrographs support the quantitative data presented in Table 4 and Table 5 and provide direct visual evidence of the microstructural differences between the two walls. All the images were processed with CLAHE and unsharp masking to improve the grain boundary contrast for publication; no structural information was altered [23].

Figure 12 shows representative 20× micrographs of the MIG wall. All three layers exhibit coarse, elongated solidification-dendritic grain structures, consistent with the large mean grain areas measured (292–420 μm²). Grain coarsening between L1 and L2 due to interlayer heat accumulation, followed by the smaller grain size in L3 because of fewer reheating cycles in the final layer, can be observed from the change in grain boundary spacing across the panels.

Figure 13 shows the MIG + FSP wall at 50×. Fine, equiaxed dynamically recrystallized (DRX) grains replace the dendritic structure across all three layers, with sharp boundary definition and equiaxed morphology in every panel. The L3 grain coarsening (12.96 vs. 8.55–8.82 μm²) is visible as a measurable increase in the grain boundary spacing in panel (c) relative to panels (a) and (b), which is consistent with the higher FSP torque at L3 (26.6 vs. 19.1 N·m at equal RPM).

Figure 14 places MIG and MIG + FSP walls side by side at matching layer positions. The 29–48× grain area refinement ratios are apparent from the difference in the grain boundary density between the rows.

Figure 15 shows the FSP stir zone at three magnifications. Panel (a) at 150× (scale bar = 10 μm) reveals a fine sub-grain structure from severe FSP shear deformation, providing high-magnification evidence of DRX-induced refinement at the sub-grain scale. Panel (b) at 50× shows the equiaxed DRX grain population in the stir zone interior, with sharp, well-resolved grain boundaries consistent with a fully developed recrystallized microstructure. Panel (c) at 20× shows the stir zone–thermomechanically affected zone (SZ–TMAZ) boundary: equiaxed DRX grains (SZ, left) transition into elongated, partially deformed grains (TMAZ, right) across a spatially narrow gradient, indicating that DRX is confined to the FSP stir zone and does not extend into the adjacent base material.

3.9. Scanning Electron Microscopy: Grain Structure and Second-Phase Particles

The SEM examination of the MIG + FSP Layer 1 specimen at three magnifications (Figure 16) provides a high-resolution confirmation of the DRX grain structure that was identified in the optical micrograph analysis. At ×250 (scale bar = 100 μm, Figure 16a), the grain boundary network spans the full field with a uniform equiaxed topology, which is consistent with the optical micrograph grain area data (8.55 μm²). At ×500 (scale bar = 10 μm, Figure 16b), individual grains of 8–15 μm are fully resolved with clean polygonal boundaries and sharp triple junctions. At ×1000 (scale bar = 10 μm, Figure 16c), the grain boundary character is resolved: the boundaries appear as bright continuous lines against the darker grain interior, with faint low-angle sub-boundary traces visible within some grains. The triple-junction geometry visible in Figure 16c is consistent with near-equilibrium triple junction configurations and indicates thermally equilibrated DRX grain boundaries rather than recovered sub-grain structures. Systematic measurement of dihedral angles at 20–30 triple junctions with reported statistical distribution would further strengthen this equilibration claim in future work. As a provisional morphological indicator of the operative recrystallization mode, the grain boundary character in the Layer 1 SEM panels (Figure 16a,c) was examined qualitatively. The boundary network is predominantly smooth and continuous, with locally serrated and lobate segments concentrated at triple junctions and adjacent to the brighter second-phase clusters. This majority-smooth character with localized serration is consistent with a recovery- and sub-grain-rotation-dominated (continuous DRX) process accompanied by localized boundary-bulging contributions. The indicator is reported qualitatively rather than as a quantified serrated-boundary fraction, because reliable quantification of the serrated fraction requires the misorientation context that only EBSD boundary character mapping provides; definitive discrimination between continuous and discontinuous DRX is accordingly identified as a goal of future work.

SEM examination of the MIG + FSP Layer 2 specimen at high magnification (Figure 17) reveals a second consequence of FSP beyond grain refinement: fragmentation and redistribution of the eutectic silicon second phase. The ER4043 filler wire (4.5–6.0 wt.% Si) solidifies with an interconnected Al–Si eutectic network. Figure 17a (×5000, scale bar = 1 μm) shows discrete Si particles dispersed in the α-Al matrix in two morphologies: elongated rods (~0.5–1.5 μm) and equiaxed particles (~0.2–0.5 μm). Figure 17b (×18,000, scale bar = 1 μm) resolves the individual particle morphologies; the elongated forms are remnant Si rod fragments, while the equiaxed particles correspond to FSP-spheroidized Si. This Si refinement contributes Orowan strengthening, supplementary to the Hall–Petch contribution. Physically, the fragmented and spheroidized eutectic Si particles act as non-shearable obstacles that a gliding dislocation must bow around rather than cut through; this bowing requires an additional shear stress that scales inversely with the spacing between particles, and the increment is estimated from the Orowan–Ashby expression below. The particle volume fraction used in the estimate, f ≈ 8–12%, is the nominal eutectic Si content expected for the ER4043 composition (4.5–6.0 wt.% Si) rather than a fraction measured by image analysis; it is therefore carried through as an assumed range, and the resulting strengthening increment is reported as first-order and sensitive to f and to the choice of spacing definition by approximately a factor of two. Using L2 particle data ($\stackrel{-}{d}$_p ≈ 0.5 μm, f ≈ 8–12%), the random-distribution centre-to-centre spacing is λ_cc ≈ $\stackrel{-}{d}$_p·√(π/4f) ≈ 1.4 μm, and the edge-to-edge mean free path is λ_ee ≈ 0.9 μm, both within the 0.9–1.5 μm range measured from Figure 17. The Orowan–Ashby form τ_OR = (0.4 G b/(π √(1 − ν)))·ln($\stackrel{-}{d}$_p/b)/λ_ee, with G = 26 GPa, b = 0.286 nm, ν = 0.34, and λ_ee = 0.9 μm, yields τ_OR ≈ 9 MPa and a tensile flow-stress increment Δσ_OR = M·τ_OR ≈ 28 MPa (M ≈ 3.06 for FCC aluminum). The estimate is first-order; the assumed volume fraction and the choice of λ definition affect the result by approximately a factor of two.

4. Discussion

4.1. Torque–Grain Size Coupling: Physical Mechanism

Within the 1200 RPM group (L2 and L3), the layer with the larger grains required higher FSP torque, at 26.6 N·m at L3 (D(A) = 4.06 μm) vs. 19.1 N·m at L2 (D(A) = 3.35 μm). The FSP imposes equivalent shear strains of approximately γ ≈ 4–80 within the stir zone, depending on the radial position relative to the pin. At equal rotation speed and comparable interface temperature, larger grains reduce the boundary density, which increases the dislocation mean free path and delays dynamic recovery, leading to higher flow stress and torque as encoded by the d^−p term in Equation (1). This directional agreement should not be read as a unique attribution. Without independent decoupling of temperature and strain rate, the same higher-torque-at-L3 trend could also arise from a lower L3 processing temperature, which raises the flow stress through the exp(−Q/RT) term, or from a higher local strain rate sensitivity. The d^−p interpretation is therefore presented as directional support for the grain-size dependence rather than as a quantitative confirmation of the creep framework. The measured torque envelope is consistent in both trend and order of magnitude with prior instrumented FSP and FSW studies of aluminum alloys. The pronounced decrease in spindle torque from 600 RPM (118 N·m) to 1200 RPM (19–27 N·m) reproduces the established inverse dependence of torque on rotational speed and its direct dependence on traverse speed and downforce reported for FSP of cast Al–Si alloys [30] and for FSW of wrought aluminum [31]. Once the present shoulder geometry (18 mm diameter) and the comparatively low traverse speed (50 mm/min) are taken into account, the absolute magnitudes fall within the range reported by these tool-instrumented studies, which supports the use of the controller torque as a physically meaningful process signal. The L1 → L2 torque drop (118 → 19.1 N·m) at doubled spindle speed is, by contrast, dominated by thermally activated softening through exp(−Q/RT), because the grain sizes at L1 and L2 are nearly identical (D(A) = 3.30 vs. 3.35 μm). The RPM step from 600 to 1200 simultaneously raises the frictional heating rate (∝ ω²), the strain rate (∝ ω), and modifies the DRX kinetics, so the three-layer dataset can only support the d^−p trend directionally; constraining the constitutive parameters (C, n, p, Q) would require an experimental matrix that spans multiple RPM and temperature conditions. Figure 18 summarizes the process–microstructure linkage in four panels: (a) FSP torque per layer; (b) mean grain area MIG vs. MIG + FSP; (c) aspect ratio evolution; and (d) P/A ratio evolution.

4.2. P/A Ratio as a Layer-Resolved Stored Energy Index

The non-monotonic ⟨P/A⟩ evolution (2.541 → 2.105 → 2.502 μm⁻¹) captures thermomechanical history information that the mean grain size and the aspect ratio alone cannot reveal. For a random planar section through an isotropic boundary network, S_V = (4/π) L_A [16], so the per-grain ⟨P/A⟩ in Table 5 is proportional to L_A. A reduction in ⟨P/A⟩ at constant grain size admits two interpretations that the present two-dimensional descriptors cannot distinguish: a genuine reduction in boundary line density through recovery or a change in the grain-shape distribution at fixed mean area. The recovery interpretation is consistent with the thermal history, in that L2 experiences double thermal exposure from the L2 FSP pass and the subsequent L3 deposition, whereas L3 experiences a single FSP pass without subsequent annealing. Discriminating recovery from a shape-distribution change requires EBSD-based misorientation distributions (low-angle versus high-angle boundary fractions, GOS, and KAM), which are identified as future work; the present evidence is therefore reported as consistent with reduced boundary line density rather than as proof of recovery. The tension between L3 showing higher circularity (0.612 vs. 0.569 at L2) and higher ⟨P/A⟩ (2.502 vs. 2.105) reflects a scale dependence: circularity captures the macro-scale grain outline, whereas ⟨P/A⟩ is sensitive to finer-scale serrations retained from DRX. The normalized boundary complexity index Ψ = (P/A)√A removes the absolute grain-scale dependence and therefore isolates boundary irregularity from grain size. Its per-layer values (7.43 for L1, 6.25 for L2, and 9.01 for L3) follow the same non-monotonic profile as ⟨P/A⟩, with the L2 minimum and the L3 maximum, which indicates that the non-monotonic trend reflects a genuine change in boundary irregularity rather than a pure scale effect. This scale-independent confirmation addresses the same question that a layer-resolved fractal-dimension analysis would; the per-layer fractal exponents are identified as future work, while the global perimeter–area power-law fit reported in Section 3.6 remains the registered descriptor for the present dataset. The absolute ⟨P/A⟩ and Ψ values depend on the segmentation pipeline; the relative trends are robust within the present pipeline.

4.3. Grain Size Distribution and DRX Mechanism

The L2 unimodal log-normal population (KS p = 0.601) matches a single dominant recrystallization regime; the same morphological signatures, however, are compatible with continuous DRX (CDRX) through sub-grain rotation, discontinuous DRX (DDRX) through boundary bulging, and recovery-dominated refinement followed by progressive boundary migration [20]. The non-log-normal distributions in L1 (possibly bimodal or mixed) and L3 (growth tail) point to incomplete or mixed regimes. The Zener–Hollomon parameter Z = $\stackrel{.}{\epsilon }$·exp(Q/RT), with Q ≈ 130–145 kJ/mol, yields layer-specific order-of-magnitude estimates of Z_L1 ≈ 10¹⁰–10¹³ s⁻¹ and Z_L2/L3 ≈ 10⁹–10¹⁰ s⁻¹ when combined with the spindle power data (7.4 vs. 2.4–3.3 kW) and thermal profiles from the literature [32,33]. The layer-specific estimates are tabulated in Table S4. These estimates are indicative rather than definitive without in situ temperature measurement. At the lower Z values estimated for L2/L3, CDRX via sub-grain rotation is the expected dominant mechanism for Al alloys [18,19], in line with the unimodal log-normal distribution at L2. At L1, the lower substrate temperature may shift Z upward and could activate a DDRX contribution, which would account for the observed deviation from log-normality. The present optical and SEM evidence tracks DRX-type morphology but cannot rigorously distinguish CDRX, DDRX, recovery-dominated refinement, or pure sub-grain rotation without crystallographic data. Definitive mechanism identification requires EBSD with misorientation distributions, low-angle/high-angle boundary character maps, GOS, and KAM analysis; the present work therefore reports the morphological and statistical signatures of DRX without assigning a specific recrystallization mechanism.

4.4. Process–Microstructure–Property Chain

The process–microstructure–property chain for the present work can be traced as follows. The FSP step replaces the MIG melting regime (peak 870–998 °C [3]) with a solid-state regime estimated below approximately 400 °C, and this thermal substitution drives a 29–48× reduction in grain area, a monotonic decrease in aspect ratio (1.389 → 1.323), and a non-monotonic ⟨P/A⟩ evolution that tracks the build position-dependent boundary energy. The SZ–TMAZ panel in Figure 15c indicates that this refinement is spatially confined to the FSP stir zone. The 34× refinement at L1 produces a 45.8% hardness increase through combined Hall–Petch and Orowan strengthening, and the L3 grain coarsening is detectable directly from the CNC torque log post-pass (26.6 vs. 19.1 N·m at equal 1200 RPM) without post-process metallography.

4.5. Limitations

Several limitations of the present study should be acknowledged. The DRX mechanism cannot be rigorously distinguished from recovery-dominated refinement or pure sub-grain rotation on the basis of grain size distribution and SEM morphology alone; EBSD with LAB/HAB character maps, GOS, and KAM analysis would be required for definitive attribution. In a related sense, the FSP interface temperature was not measured in situ in this work, and the temperature estimates used in the constitutive framework discussion are drawn from prior work [3] and from literature values for similar FSP conditions. The constitutive parameters (C, n, p, Q) in Equation (1) were not individually fitted; the equation serves here as a physical framework, and parameter determination would require an experimental matrix spanning multiple RPM and temperature conditions. On the mechanical side, the Vickers microhardness sampling (n = 6 per wall) does not resolve the layer-by-layer hardness gradient, and nanoindentation arrays together with tensile testing would be needed to constrain the per-layer Hall–Petch slope and the per-layer yield strength contribution. The Fields–Backofen torsion correction assumes a uniform torsional geometry, whereas the FSP strain field is three-dimensional with radial non-uniformity ($\stackrel{.}{\epsilon }$ ∝ ωr) and a shoulder-induced compressive component, so the conversion is approximate rather than an absolute calibration of τ_s. Finally, the perimeter-derived descriptors (⟨P/A⟩, circularity, D_f, and Ψ) depend on the segmentation pipeline of Section 2.4; absolute values should not be compared directly to studies using different segmentation parameters, though the layer-to-layer relative trends are robust within the present pipeline. These limitations bound rather than invalidate the principal findings of this study.

5. Conclusions

This study established the layer-resolved grain morphology and the recrystallized-crystal evolution of MIG + FSP-fabricated aluminum alloy 4043 walls, with the FSP spindle torque from the CNC controller paired with multi-descriptor grain morphology. The specific quantitative findings are summarized below.

• The FSP spindle torque varied systematically across the three build layers: 118 ± 6 N·m at 600 RPM (L1), and 19.1 ± 1.0 and 26.6 ± 1.3 N·m at 1200 RPM (L2 and L3). Within the 1200 RPM group, Layer 3 required approximately 39% higher torque than Layer 2, which is qualitatively consistent with the larger grains at L3 (D(A) = 4.06 vs. 3.35 μm) providing less grain boundary resistance, in line with the d^−p term in Equation (1). The L1 → L2 contrast at the RPM step couples temperature, strain rate, and grain-size effects and is interpreted directionally rather than quantitatively.
• The MIG + FSP wall achieved grain areas of 8.55, 8.82, and 12.96 μm² (L1–L3), representing 34×, 48×, and 29× refinement relative to the MIG wall (292–420 μm²). The aspect ratio decreased monotonically (1.389 → 1.323), which is consistent with the enhanced DRX under the higher heat input and strain rate of the 1200 RPM upper-layer processing.
• The P/A ratio exhibited a non-monotonic layer dependence (2.541 → 2.105 → 2.502 μm⁻¹). Layer 2 showed the lowest boundary energy density from the double thermal exposure; Layer 3 retained a higher complexity from a single FSP pass without annealing. This asymmetry encodes the build position-dependent stored energy that is not captured by the mean grain size.
• The Layer 2 grain area distribution conformed to a log-normal form (KS p = 0.601, μ = 1.796, σ = 0.906), which is consistent with a unimodal grain area population. Log-normal conformance alone does not uniquely identify the recrystallization mechanism. Layers 1 and 3 deviated significantly (p < 0.001), which may be attributed to a bimodal or mixed population near the substrate interface and to a grain growth tail, respectively. The morphological and distributional evidence is consistent with DRX; discrimination between continuous and discontinuous DRX requires EBSD-based mechanism analysis, identified as a priority for future work.
• The optical micrographs (20×–150×) and SEM (×250–×1000) support the equiaxed grain structure of the MIG + FSP wall: there are clean polygonal boundaries, triple junctions whose geometry is consistent with near-equilibrium dihedral configurations (pending dihedral-angle measurement), and faint sub-boundary traces in grain interiors. The SEM at ×5000–×18,000 reveals FSP-driven fragmentation of the eutectic Si network into discrete elongated rods and equiaxed particles (approximately 0.2–1.5 μm), which contributes Orowan strengthening supplementary to Hall–Petch. A first-order strengthening partition consistently using D(A) for both walls (Table 4) yields three contributions: Δσ_HP ≈ 22.6 MPa (k_HP ≈ 0.07 MPa·m^(1/2) [34], d_FSP = 3.30 μm, d_MIG = 19.29 μm at L1); Δσ_OR ≈ 28 MPa from Orowan bypass of the refined Si particles; and Δσ_ρ = αMGb√ρ ≈ 30–40 MPa from retained dislocation density (ρ ≈ 10¹⁴ m⁻², α ≈ 0.3, M = 3.06, G = 26 GPa, b = 0.286 nm). The summed increment (approximately 80 MPa) is consistent with the hardness-derived flow stress difference (Δσ_hardness ≈ 78 MPa via H ≈ 3σ Tabor conversion). The full strengthening partition is detailed in Supplementary Note S1. The partition is first-order; the per-layer Hall–Petch slope is not tested with the present n = 6 sampling.
• Overall, the FSP spindle torque carries enough microstructural information to track the layer-to-layer grain size differences in the MIG + FSP process, which points to a practical route toward post-pass microstructure monitoring without post-process metallography. A closed-loop extension will require experiments at intermediate RPM values and in situ subsurface temperature measurement.

The principal contributions of this work are threefold. First, it provides the first pairing of CNC-controller FSP spindle torque with layer-resolved, multi-descriptor grain morphology in the WAAM + FSP process. Second, it demonstrates that the perimeter-to-area ratio and the normalized boundary complexity index Ψ resolve the build position-dependent stored boundary energy that the mean grain size alone cannot capture. Third, it presents a quantitative first-order strengthening partition that links grain refinement and second-phase dispersion to the measured hardness gain. The practical application follows directly from the first contribution: because the L3 grain coarsening is detectable from the controller torque log (26.6 versus 19.1 N·m at equal 1200 RPM) without post-process metallography, the spindle-torque signal provides a route toward in-process, post-pass microstructure monitoring and, ultimately, closed-loop microstructural control of hybrid additive manufacturing.