1 Introduction

Additive manufacturing (AM) has matured into a production-capable technology for metallic components with highly optimized complex geometries, especially in domains where performance and mass efficiency are more essential. Among the various metal AM processes, Selective Laser Melting (SLM) has established itself as a leading technique for fabricating complex structural parts in aerospace, automotive, and tooling applications, due to its ability to produce near-net shape geometries with high material strength [1]. The integration of AM into design workflows has accelerated the adoption of structural optimization techniques, such as Topology Optimization (TO), that exploit this geometric freedom in highly complex structures [2]. Despite ongoing advances in process monitoring, support optimization, and thermal modelling, the physical reality of SLM fabrication continues to introduce geometric deviations that are both spatially irregular and mechanically consequential [3, 4].

TO has benefited substantially from the geometric flexibility enabled by SLM and other 3D printing techniques [5,6,7]. Its capacity to produce non-intuitive, organic structures that outperform conventional designs is well documented, especially for load-bearing mechanical components such as brackets, suspension arms, and other heavy components in both automotive and aerospace industries [8,9,10,11,12]. These complex structures, often featuring curved load paths, internal stiffening ribs, and non-uniform cross-sections, are inherently sensitive to dimensional irregularities. While SLM can approximate the optimized geometry with impressive fidelity, its as-built surfaces are subject to distortions driven by thermal gradients, powder adhesion, microstructural contractions, overhang instabilities, and deviations most commonly caused by support structures [13,14,15]. The disconnect between the idealized CAD output from TO algorithms and the realized SLM build, therefore, introduces uncertainty that is not captured in most digital simulations.

In the present work, SLM is not presented as a universally superior manufacturing route to conventional casting or forging for all connecting rods. Its importance here is linked to the topology-optimized nature of the investigated component, which contains curved transitions, non-uniform cross-sections, and lightweight material distribution that are more compatible with additive manufacturing. The objective of this study is therefore not to compare SLM with conventional production routes, but to investigate how SLM-induced as-built geometric deviations affect the structural response of a topology-optimized connecting rod.

A central issue is that most TO studies assume perfect geometric fidelity. The optimized structure is fed directly into a finite element simulation under the idealized premise that the physical part will reproduce the CAD geometry with negligible deviation. This assumption often holds for cast or machined components but becomes questionable for powder-bed fusion systems, especially when thin-walled features, organic branching morphologies, or delicate fillets are present. Small deviations in such localized regions can sometimes redirect stress paths subtly or drastically, which may lead to shifts in peak stress locations and magnitudes [16, 17]. Despite extensive literature on AM-induced distortions and dimensional analysis, the propagation of these distortions into structural performance remains insufficiently explored.

In parallel with advances in SLM and other AM methods, the growing availability of cost-effective 3D scanning technologies has enabled improved geometric validation of as-built AM parts. Advanced metrology platforms, including structured-light scanners, laser scanners, and photogrammetry systems, allow high-resolution acquisition of the manufactured surface. These scan results are typically compared against the nominal CAD model to quantify dimensional deviations using methods such as iterative closest point (ICP) alignment, Hausdorff distance measurements, volumetric comparisons, or full-surface-to-CAD deviation maps [18]. In most reports, deviations are presented as global metrics in the form of errors, volumetric changes, average deviation magnitudes, or as deviation maps highlighting local discrepancies [19, 20].

Although many investigations use 3-D scanning and deviation mapping to validate as-built AM parts, most stop short of incorporating the detailed scan geometry into structural simulations [21,22,23,24,25,26]. Even when scanned geometries are reconstructed into mesh models, they are rarely used directly in structural analysis due to concerns about mesh quality, noise artefacts, watertightness, and the effort required to convert point-cloud data into a FEA-ready format. Instead, researchers tend to employ CAD-based geometry for performance evaluation, treating the scan merely as a verification tool rather than as a direct input to functional simulation.

The current body of work offers extensive insight into geometric deviations in AM parts and an equally mature foundation in the structural performance of topology-optimized components. However, efforts that meaningfully integrate these two domains remain limited. The influence of real, spatially distributed dimensional deviations introduced during SLM on the structural behavior of topology-optimized components under service-like loading conditions remains poorly established. This gap becomes more significant in designs where topology optimization reduces material in non-critical regions to achieve lightweight architectures, consequently lowering safety margins and increasing sensitivity to localized geometric irregularities.

Furthermore, even when deviations are examined, they are often treated as scalar metrics such as maximum error, average deviation, and dimensional tolerance assessments, rather than as spatial fields. Structural performance, particularly stress flow, is inherently governed by local geometry, and thus it cannot be reliably inferred from global deviation metrics. Local swelling near a fillet, thinning of a rib, or rounding of a sharp curvature may produce stress amplifications that do not scale linearly with deviation magnitude [16, 17]. The nonlinear coupling between geometry and stress distribution makes this a problem requiring direct simulation with the accurate as-built geometry.

The absence of studies that directly incorporate the as-built geometry into FE simulation is partly practical and partly conceptual. Point-cloud datasets from scanners often contain noise, occlusions, and reconstruction artefacts that can compromise meshing. Processing the data into a watertight surface and subsequently into a volumetric mesh is non-trivial. Additionally, many studies assume that minor geometric deviations have little influence on stress distributions, thereby reinforcing continued reliance on the idealized CAD representation. For topology-optimized components, especially those with slender members or curvature-sensitive regions, this assumption can be overly optimistic, as even subtle deviations can alter local load paths and increase stress concentrations.

To address this gap, this study develops a deviation-aware verification workflow for a topology-optimized Maraging Steel 300 connecting rod manufactured by SLM. The workflow integrates topology optimization, fabrication, structured-light 3D scanning, scan-to-CAD alignment, geometric deviation quantification, reconstruction of the as-built geometry into a simulation-ready model, and finite element analysis of both the nominal CAD and scan-derived geometries under identical loading conditions.

To clearly define the scope of the present work, this study is positioned as a single-case investigation of one topology-optimized Maraging Steel 300 connecting rod manufactured in one SLM build orientation and evaluated under one representative static load case. Accordingly, the results are intended to demonstrate the value of the deviation-aware scan-to-FEA workflow and the possible structural significance of as-built geometric deviations in a topology-optimized component, rather than to claim broad quantitative generalization across all geometries, materials, build orientations, or loading conditions.

The primary contribution of this work is the quantitative demonstration that measured as-built geometric deviations, although globally small, can materially amplify peak stress and alter stress flow in a topology-optimized component. A secondary contribution is the demonstration that low-cost structured-light scanning, when combined with careful alignment and reconstruction, can generate scan-derived geometry suitable for direct finite element analysis and structural assessment. While this is not positioned as the principal novelty of the study, it serves as an important enabling step for the primary contribution.

By integrating topology optimization, SLM fabrication, 3D scanning, and finite element analysis into a unified workflow, this study establishes a reproducible framework for evaluating performance-based geometric deviations in metal additive manufacturing.

2 Materials and methodology

This study implements an integrated design–manufacture–scan–FEA workflow, as shown in Fig. 1, to evaluate how as-built dimensional deviations influence the structural response of a topology-optimized SLM-manufactured connecting rod. The methodology begins with topology optimization to obtain the nominal CAD geometry, followed by fabrication in Maraging Steel 300 using laser powder bed fusion. The printed component was digitized using structured-light 3D scanning with a low-cost handheld scanner, and the resulting point cloud was aligned with the CAD model for deviation quantification using a Python program. The scan-derived geometry is then reconstructed into a simulation-ready STEP file using Ansys SpaceClaim. Finally, both nominal and as-built models are subjected to identical finite element loading conditions, enabling a direct comparison of stress behavior and the impact of geometric deviations.

Fig. 1
Fig. 1
Full size image

Flowchart of the closed-loop workflow for deviation-aware structural analysis

2.1 Topology optimization

The nominal geometry of the connecting rod was obtained by topology optimization in ANSYS Discovery, using the same workflow as in our previous study on the same component [10]. In that work, multibody dynamics and static analysis were used to identify a representative worst-case compressive load acting on the small end of the rod while the big end was fixed. The resulting design load of 64,500 N was adopted here as the static load case for topology optimization and subsequent finite element analysis [10]. The topology optimization in reference [10] was conducted against a defined set of design criteria, which included reducing the component mass by at least 3%, reducing the maximum von Mises stress and von Mises strain, defining protected zones at both rod ends, maintaining major dimensions, obtaining a cross-section geometry other than the standard I-section, keeping the factor of safety (FOS) above 2.5, and ensuring manufacturability by both additive manufacturing and conventional forging. The optimized topology adopted in the present study satisfies all of these criteria.

In the present study, the connecting rod was optimized for minimum compliance under this load, with the big-end bore fully constrained and the 64,500 N compressive force applied at the small-end interface. Manufacturing constraints were imposed to preserve the big- and small-end bearing regions as non-design zones, while only the shank and transition regions were allowed to evolve during optimization (Fig. 2).

Fig. 2
Fig. 2
Full size image

The a Protected region of the connecting rod. b Applied boundary conditions on the connecting rod

The finally optimized topology has a 5.47% mass reduction compared to the original design from 606.77 to 573.58 g, as shown in Fig. 3.

Fig. 3
Fig. 3
Full size image

Original Connecting rod versus the topology optimized connecting rod

2.2 SLM fabrication in maraging steel 300

The optimized connecting rod was fabricated in Maraging Steel 300 on a Renishaw 500 S Flex laser powder bed fusion machine. Maraging Steel 300 was selected because the connecting rod is a highly loaded structural component, and the present study required a metal with high strength and stiffness that is also suitable for laser powder bed fusion processing. In addition, the use of SLM is consistent with the topology-optimized geometry considered here, since the process enables fabrication of complex freeform features without redesigning the part for conventional casting constraints. Processing parameters followed standard manufacturer recommendations, including a 50 μm layer thickness, and laser power and scan speeds followed according to the manufacturer’s build rate of 13.5 cm3/h. The properties of the material were shown in Table 1.

Table 1 Properties of Maraging Steel 300 (Renishaw’s M300 1.2709) [27]
Full size table

The part was oriented to reduce overhang-induced thermal distortion, with support structures concentrated around the fillet transitions and small-end curvature. Stress-relief heat treatment was applied at 300 °C for three hours before support removal. Sandblasting was performed to clean the surface after removing the support material. The connecting rod was positioned at a 45-degree tilted position, and support structures were provided as shown in Fig. 4.

Fig. 4
Fig. 4
Full size image

Position of the connecting rod in Renishaw’s QuantAM slicer

2.3 3D scanning

The as-built connecting rod was digitized using a Revopoint MetroX blue-light laser scanner, operated through the Revo Scan 5 software, as shown in Fig. 5. To maintain full geometric fidelity, no markers were placed on the component itself. Instead, reference markers were positioned on the surrounding floor surface, providing stable tracking and ensuring that the scanning process did not modify or contaminate the part.

Fig. 5
Fig. 5
Full size image

Scanning connecting rod (left) and digital copy in Revo Scan 5 software (right)

Multiple scan passes were collected from different orientations to capture fine geometric details, including fillets, curvature transitions, and slender topology-optimized members. The raw scans were imported into Revo Scan 5 for pre-processing, where noise reduction, point-cloud merging, and removal of stray or incomplete regions were performed, and the resulting high-resolution STL file was exported for deviation analysis. The Revopoint MetroX system was selected due to its metrology-grade accuracy, multi-mode scanning capability, and high acquisition speed. The key specifications relevant to this study are summarized in Table 2.

Table 2 Specifications of Revopoint MetroX [28]
Full size table

2.4 Alignment and deviation analysis

The deviation analysis between the nominal topology-optimized connecting rod and the as-built scanned geometry was performed using a custom Python workflow. The procedure consists of three sequential steps: (i) Principal Component Analysis (PCA) based coarse pre-alignment, (ii) ICP refinement using nearest-neighbor correspondence, and (iii) geometric deviation computation, including component-wise displacements, Euclidean deviation magnitude, Hausdorff distance, and volumetric comparison. The pseudocode of the python workflow to perform deviation analysis is provided in the Appendix A.1.

2.4.1 PCA-based initial alignment

PCA was used only for coarse rigid-body pre-alignment between the scanned and nominal geometries. For a point set \(V=\{ {v_i}\} _{{i=1}}^{n} \subset {{\mathbb{R}}^3}\)with centroid \(\mathop v\limits^{\prime }\), the covariance matrix is computed as

$$C=~\frac{1}{n}\mathop \sum \limits_{{i=1}}^{n} \left( {{v_i} - \bar {v}} \right)~{\left( {{v_i} - \bar {v}} \right)^T}$$
(1)

Eigen-decomposition of C gives the principal axes and their associated variances, which are used to define the initial orientation of each point cloud. The initial rotation for coarse alignment is then obtained from the principal-axis bases of the reference and scan point sets, while the initial translation is obtained from the centroid difference after rotation. This PCA provides only a coarse alignment and serves as the initialization for the subsequent ICP refinement [29]. C is decomposed as,

$$C~=~Q{\text{\varvec{\Lambda}}}{{\text{Q}}^T}$$
(2)

where columns of Q are the principal axes, Λ is a diagonal matrix containing the eigenvalues of the covariance matrix [30]. The initial rotation is computed as

$$R{~_0}=~{Q_2}Q_{1}^{T}$$
(3)

where Q1, Q2, R0 ∈ R3 × 3, R0 is the initial rotation matrix obtained from PCA alignment, and Q1, Q2 are the Orthogonal matrices whose columns are the principal axes of V1 and V2 [31].

$$t{~_0}=~\bar {V}{~_1}~ - ~R{~_0}\bar {V}{~_2}~$$
(4)

where t0 is the initial translation vector aligning the centroids of both points, and it provides a coarse rigid-body alignment before ICP refinement [32].

2.4.2 ICP refinement

Fine alignment was performed using a point-to-plane ICP algorithm, where point correspondences are established using a nearest neighbor search. For each iteration k, a correspondence set can be expressed as.

$$V_{1}^{{NN}} = ~\left\{ {\upsilon _{i} \in ~V_{1} ~:~\upsilon _{i} ~ = ~arg\mathop {{\text{min}}}\limits_{{u \in V_{1} }} \left\| {u - \upsilon _{i} } \right\|} \right\}_{{j = 1}}^{N}$$
(5)

where \(V_{1}^{{NN}}\) is the set of all nearest neighbors for each point in the scan, and \({\upsilon _i}\) is a single point in either point cloud [33]. It is computed using a KD-tree for efficient O (log N) queries. The optimal rigid transform minimizing

$$E~\left( {R,t} \right)=~\mathop \sum \limits_{{j=1}}^{N} {\left[ {n_{j}^{T}\left( {R{p_j}+t - {q_j}} \right)} \right]^2}~$$
(6)

where E(R, t) is the ICP objective function representing the sum of squared alignment errors, and \({{\text{p}}_{\text{j}}}\) is scan point \({{\text{q}}_{\text{j}}}\) is the corresponding closest point on the reference surface \({{\text{n}}_{\text{j}}}\)is surface normal at \({{\text{q}}_{\text{j}}}\) [34]. Convergence is monitored through the change in mean nearest neighbor error, terminating when

$$\left| {{E_k}~ - ~{E_{k - 1}}} \right|~<\epsilon ~~$$
(7)

where Ek and Ek−1 are the mean nearest-neighbor errors at iterations k and k – 1, respectively, and \(\varepsilon\) is the convergence for termination [35]. Fine alignment was performed using a point-to-plane ICP algorithm. At each iteration, nearest-neighbor correspondences were established from the scan point set to the reference surface using a KD-tree search. For each scan point, the closest reference point and its associated surface normal were used to define the local point-to-plane residual. The rigid transformation increment was estimated by minimizing the sum of squared normal-direction distances between the transformed scan points and the corresponding reference surface. Convergence was monitored using the change in mean point-to-plane residual between successive iterations, and the refinement was terminated when this change fell below the prescribed tolerance \(\varepsilon\).

2.4.3 Deviation vector computation

For each aligned scan point pi, the deviation vector.

$${\Delta _i}~=~{p_i}~ - ~p_{i}^{{NN}}~$$
(8)

Which is later decomposed into component-wise deviation Δxi, Δyi, and Δzi in the x, y, and z directions [36]. The Euclidean deviation magnitude is.

$$\left\| {{\Delta _i}} \right\|~=~\sqrt {{{\left( {\Delta {x_i}} \right)}^2}+{{\left( {\Delta {y_i}} \right)}^2}+{{\left( {\Delta {z_i}} \right)}^2}}$$
(9)

The two-sided Hausdorff distance between meshes A and B is computed as

$$H~\left( {A,B} \right)~ = ~\max ~\left( {\mathop {\max }\limits_{{a \in A}} \mathop {\min }\limits_{{b \in B}} \left\| {a - b} \right\|,~\mathop {\max }\limits_{{b \in B}} \mathop {\min }\limits_{{a \in A}} \left\| {a - b} \right\|~} \right)$$
(10)

where A, B are the surface point sets of the CAD and scan meshes [37].

2.4.4 Volume evaluation

The enclosed volume of each STL mesh is computed using signed tetrahedral decomposition, where total volume is.

$$V~=~\mathop \sum \limits_{{i=1}}^{{{N_{tri}}}} \frac{1}{6}~{v_{i1}} \cdot \left( {{v_{i2}} \times {v_{i3}}} \right)$$
(11)

This formulation provides a metric for material gain or loss between the design intent and the as-built component [38].

2.5 Scan to mesh reconstruction and pre-processing for simulation

The scan-derived mesh was prepared as a watertight STL prior to import into ANSYS SpaceClaim. This is consistent with the capabilities of Revo Design, powered by QUICKSURFACE, which provides mesh-editing tools to generate watertight meshes and supports STL export of processed scan data. The watertight STL was then imported into ANSYS SpaceClaim and converted into a solid body through direct solid reconstruction of the closed triangulated mesh. No NURBS fitting, freeform surface approximation, surface smoothing, or automated re-surfacing operations were applied during this conversion.

In this way, the scan-derived geometry retained the original triangulated shape representation captured from the as-built component. Only planar reference faces were manually created at the small-end and big-end interfaces to enable consistent load and boundary-condition application in the finite element model; these operations were limited to interface definition and did not modify the surrounding scan-derived geometry.

The resulting solid STEP file represents a direct volumetric conversion of the scan-derived mesh and was used for meshing and structural analysis under boundary conditions identical to those applied to the nominal CAD model. This approach ensured a one-to-one comparison between the ideal and as-built geometries while retaining full geometric fidelity of the manufactured component.

2.6 Post-conversion geometric fidelity verification

To verify that the scan-to-STEP conversion did not introduce meaningful approximation error, an additional post-conversion geometric fidelity verification was performed. The original scan representation and the simulation-ready converted geometry were both aligned in the CAD coordinate system, and the converted geometry was then compared directly against the original scan. This procedure was introduced to distinguish true as-built geometric deviation from any discrepancy that might arise during the reconstruction and conversion workflow.

No NURBS fitting, freeform surface approximation, surface smoothing, or automated re-surfacing operations were introduced in this verification route; therefore, any reconstruction-related geometric discrepancy is limited to the solid conversion step itself rather than to additional surface idealization.

2.7 Finite element model setup

Both the nominal topology-optimized CAD model and the scan-derived STEP file were imported into ANSYS Workbench for structural simulation under identical conditions. To ensure consistent discretization quality, an initial finite element mesh was generated using the 1/10th of the minimum feature rule, excluding fillets and chamfers. This provided a uniform starting point for both geometries while capturing slender members and curvature transitions introduced by topology optimization.

The mesh was considered converged continuously until the change in maximum stress fell within ± 5%, providing a consistent and reproducible refinement benchmark. Both models were assigned identical material properties corresponding to Maraging Steel 300.

Boundary conditions were consistently applied; the big end of the connecting rod was fully constrained in all degrees of freedom, and a compressive force of 64,500 N was applied at the small-end interface, as shown in Fig. 6. By maintaining uniform meshing rules and boundary conditions across both models, the simulation framework isolated the influence of as-built geometric deviations on the structural response under load.

Fig. 6
Fig. 6
Full size image

Boundary conditions applied on the CAD model (left) and the scanned model (right)

3 Results

3.1 Deviation analysis between the CAD model vs. the scanned model used in FEA

The ICP-based alignment converged smoothly, with a rapid reduction in registration residual during the early iterations followed by progressively smaller updates as the solution approached stability. The PCA initialization provided a consistent coarse orientation, and the subsequent ICP refinement converged effectively at approximately the 24th iteration. Beyond this point, successive updates produced negligible changes of less than 0.001 mm, indicating numerical convergence of the rigid registration. The monotonic decay of the residual confirms that the scan and CAD geometries were brought into stable correspondence prior to deviation quantification.

Component-wise signed deviation statistics revealed a predominantly symmetric distribution around zero in the Y- and Z-directions, whereas the X-direction showed a slight positive bias of + 0.0898 mm. The slight positive bias observed in the X-direction (+ 0.0898 mm) is likely attributable to the combined effects of the selected 45° build orientation, localized support interaction, and outward material accumulation in curvature-sensitive and support-affected regions. Because the scan and CAD geometries were rigidly aligned through PCA initialization followed by ICP refinement, this directional bias is interpreted not as a residual global misalignment, but as a real geometric asymmetry arising from the SLM fabrication and post-processing route. The fact that the scan-derived model also exhibited a 3.10% volume increase supports the interpretation that this bias is associated with localized swelling or excess material retention rather than a simple positional offset. This indicates that deviations arose primarily from localized geometric offsets on freeform surfaces rather than from any global directional shift, which is also consistent with the side-by-side comparison of the CAD and scanned models shown in Fig. 7. The signed deviation spread of approximately ± 0.23 mm demonstrates that most surface points fall within a narrow band of geometric consistency. In contrast, the heavy-tailed ends of the signed deviation distribution, extending from roughly − 0.83 mm to + 0.83 mm, are associated with isolated outlier regions related to support interaction and local geometric irregularity.

Fig. 7
Fig. 7
Full size image

Side-by-side comparison of the CAD model and the scanned model

The Euclidean deviation magnitude exhibited a maximum of 1.746 mm, as measured by the two-sided Hausdorff distance. These maxima correspond to locations where the as-built geometry deviated significantly from the intended topology-optimized design, most notably around support-contact surfaces, small-end interior curvature, and thin ribs where thermal accumulation is most pronounced during the SLM process. Volumetric comparison further confirmed the geometric divergence between the nominal and as-built geometries. The scan-derived model had a volume of 73,014.09 mm3, which exceeded the nominal CAD volume of 70,822.04 mm3 by 3.10%, indicating a measurable accumulation of excess material in specific regions. This increase is consistent with localized surface thickening along curved and thermally intensive regions. The magnitude of the volume increase aligns with the spatial distribution of deviation hotspots observed in the deviation heatmap. The distribution of signed surface deviations is illustrated in Fig. 8, which shows a unimodal, slightly right-skewed histogram centered near zero. Most surfaces fall within a narrow deviation range of approximately − 0.20 mm to + 0.30 mm, confirming that the bulk of the as-built geometry remains close to the nominal CAD normal direction. The histogram tails extend to approximately − 0.83 mm and + 0.83 mm, capturing the occurrence of larger but relatively infrequent local deviations.

Fig. 8
Fig. 8
Full size image

Signed deviation histogram representing the scanned model over the CAD normal

Complementing this statistical view, Fig. 9 presents a full-surface deviation heatmap, highlighting the spatial heterogeneity of deviation distribution. Regions shaded in red correspond to outward deviations relative to the CAD, while blue regions indicate inward shifts. Most mid-span beam surfaces fall within green–yellow bands, indicating near-zero deviation. In contrast, more pronounced deviations are concentrated around curvature transitions, support structure regions, and both end features of the connecting rod.

Fig. 9
Fig. 9
Full size image

± Heatmap based on the deviation between the CAD model and the Scanned Model

Together, the ICP convergence curve, deviation distribution, and volume discrepancy shows that although the overall geometric fidelity remains within typical SLM tolerances, several localized irregularities represent non-negligible departures from the topology-optimized intent. These geometric anomalies form the basis for evaluating how such deviations influence structural performance in the subsequent finite element analysis.

3.2 Post-conversion fidelity verification

The post-conversion fidelity verification confirmed that the simulation-ready converted geometry preserved the original scan geometry with negligible geometric alteration. The mean absolute deviation between the original scan and the converted geometry was 0.0182 mm, with a median deviation of 0.0167 mm, an RMS deviation of 0.0217 mm, and a two-sided Hausdorff distance of 0.0594 mm. The volume difference between the two representations was only 0.0040 mm3 (0.0000055%).

These values are negligible when compared with the previously reported CAD-versus-as-built maximum deviation of 1.746 mm and the 3.10% volume increase of the as-built geometry. Figure 10 presents the signed deviation histogram for this post-conversion fidelity check, showing that the distribution remains tightly bound around zero with only limited tails.

Fig. 10
Fig. 10
Full size image

Signed deviation histogram between the original scan representation and the simulation-ready converted geometry after alignment in the CAD coordinate system

Figure 11 presents the corresponding full-surface deviation heatmap, confirming that the conversion-related discrepancy is spatially small compared with the main CAD-versus-as-built deviation pattern.

Fig. 11
Fig. 11
Full size image

Full-surface deviation heatmap between the original scan representation and the simulation-ready converted geometry after alignment in the CAD coordinate system

Accordingly, the simulation-ready geometry is considered to preserve the as-built shape with sufficient fidelity for the present FEA, and the stress differences reported in this study are interpreted as arising predominantly from the real manufactured geometry rather than from reconstruction-induced approximation.

3.3 FEA stress comparison

For the CAD geometry, successive mesh refinements increased the peak equivalent von Mises stress from 614.29 MPa to 728.97 MPa, and then to 757.78 MPa, with consecutive increases of 17.08% and 3.88%. In contrast, the node and element counts increased from 273,754 / 158,890 to 526,126 / 333,785 and 931,989 / 621,001, respectively.

For the scanned geometry, three refinement levels raised the peak stress from 706.65 MPa to 744.82 MPa, 850.13 MPa, and ultimately 890.39 MPa, with corresponding changes of + 10.79%, + 10.83%, and + 4.74%, accompanied by an increase in mesh density from 179,411 / 104,920 to 344,470 / 220,332, 609,712 / 409,814, and 914,568 / 635,217 nodes/elements. These mesh-convergence checks ensured that both geometries were evaluated on meshes of comparable quality, so any differences in structural response can be attributed to as-built geometric deviations rather than discretization error (Table 3).

Table 3 Stress-based mesh convergence data of CAD model and scanned model
Full size table

At the final converged mesh, the CAD model exhibited a peak von Mises stress of 757.78 MPa, while the scanned model reached 890.39 MPa, corresponding to a stress increase of approximately 17.50%, as shown in Fig. 12. This difference was obtained under the present static linear-elastic analysis and the single nominal compressive loading case considered in this study; therefore, it should be interpreted as a case-specific comparative result for the present geometry, boundary conditions, and loading configuration. The observed amplification is attributable to localized geometric irregularities captured in the scan-derived model and highlights how as-built manufacturing deviations can influence stress distribution in this topology-optimized component.

Fig. 12
Fig. 12
Full size image

The equivalent stress of the CAD model (left) and the scanned model (right)

4 Discussion

The comparison between the nominal CAD geometry and the scan-derived as-built model indicates that sub-millimeter geometric deviations introduced during SLM can produce structurally meaningful changes in stress distribution in a topology-optimized component. As shown in Fig. 13, the CAD model exhibits smoother and more continuous stress trajectories along the rod shank and transition regions, whereas the as-built geometry shows more irregular and localized stress patterns. These differences are consistent with the presence of localized geometric deviations in curvature-sensitive and support-affected regions.

Fig. 13
Fig. 13
Full size image

The Equivalent Stress path of the a CAD model and b the Scanned Model

In the CAD model (Fig. 13a), the iso-stress contours and principal stress vectors follow coherent arcs that trace the optimized structural corridors. These smooth trajectories indicate that the designed geometry provides a continuous curvature profile, allowing compressive loads to be transferred uniformly from the small-end interface toward the big-end eye. This behavior is characteristic of topology-optimized structures, in which material is precisely arranged to channel stresses through efficient, uninterrupted pathways.

In contrast, the stress field of the scanned model (Fig. 13b) is influenced by geometric irregularities captured in the as-built surface. The deviation heatmap and signed-distance statistics show pronounced dimensional departures in support-contact regions, including the big-end fillet, the inner bore surfaces, and the underside of the shank. These locations are therefore plausible contributors to the altered stress trajectories observed in the scan-derived model. The resulting geometric features, including surface roughness, localized swelling, discontinuous fillet curvature, and shallow depressions, are consistent with mechanisms that can perturb the intended stress flow.

Although the X-direction bias is small in absolute magnitude, its mechanical relevance depends on where it occurs. In topology-optimized geometry, even modest directional asymmetry near fillets, curvature transitions, support-contact zones, or bore-adjacent regions can locally modify curvature continuity and cross-sectional geometry, thereby disturbing the intended stress-flow path. In the present case, such asymmetry is therefore considered a plausible contributing factor to the altered stress trajectories and locally intensified stress regions observed in the scan-derived model [39, 40].

4.1 Influence of support-structure regions on stress flow

Support-affected regions appear to be among the most influential locations in the observed stress-path changes. Deviations of a few tenths of a millimeter can modify local curvature and surface orientation, which may perturb principal stress trajectories in topology-optimized geometry. In the scanned model, these deviations are associated with less coherent iso-stress contours, including jagged or locally intensified bands. This effect is especially visible near the big-end transition, where outward bulging and local geometric irregularity coincide with altered stress-path behavior.

Downward-facing surfaces along the rod shank exhibit similar stress-field disturbances. Thermal distortion and stair-stepping may introduce small geometric undulations that coincide with more irregular principal stress trajectories in the scan-derived model. Rather than following the smoother compression surface observed in the CAD model, the stress trajectories in the as-built component show local detours around surface irregularities. This behavior is consistent with the approximately 17.5% increase in peak von Mises stress observed in the refined FEA results. Relative to the nominal yield strength of 875 MPa, the CAD-based result remains below yield, whereas the scan-derived result slightly exceeds that value in a localized region; within the present linear-elastic framework, this should be interpreted as a potential local yield onset rather than confirmed material yielding.

These results should therefore be interpreted within the bounds of the present linear-elastic analysis. The simulations do not include residual stresses arising from the SLM process, nor do they account for elastoplastic material response or fatigue damage accumulation under cyclic loading. Consequently, although the scan-derived geometry indicates a higher local stress state than the nominal CAD model, the actual onset, extent, and structural consequences of local plasticity in service may be either lower or higher depending on the combined influence of residual stress redistribution, nonlinear material behavior, and loading history.

Although these deviations are small in magnitude, they occur at structurally sensitive regions where topology optimization assumes exact geometric fidelity. Therefore, the as-built stress field diverges from the ideal distribution predicted by the CAD model. This divergence explains why the scan-derived geometry consistently exhibits higher peak stresses, even though its overall dimensional change is relatively minor.

Accordingly, the reported 17.5% stress amplification should not be interpreted as a universally generalizable value. It represents the response of the present scan-derived geometry relative to the nominal CAD model under one representative compressive load case within a linear-elastic framework. Under different loading conditions, altered constraint definitions, residual-stress states, or nonlinear constitutive behavior, both the magnitude and spatial distribution of stress amplification could change.

The obtained peak stress values should also be considered against the mechanical limits of Maraging Steel 300 (Renishaw M300 1.2709), which has a reported yield strength of 875 MPa and ultimate tensile strength of 972 MPa [27]. For the nominal CAD model, the peak von Mises stress of 757.78 MPa lies approximately 13.3% below the yield strength and approximately 22.1% below the UTS, suggesting that the nominal design maintains an elastic safety margin under the applied compressive load within the assumptions of linear-elastic analysis.

For the as-built scan-derived model, the peak von Mises stress of 890.39 MPa marginally exceeds the yield strength by approximately 1.8%, which is consistent with the possible local yield onset discussed above. However, within the present linear-elastic framework, this exceedance should be interpreted as an indicator of potential yielding rather than definitive plastic deformation.

It is acknowledged that a direct comparison with published experimental fatigue or fracture data for SLM-processed Maraging Steel 300 under equivalent loading is not available within the present study; therefore, the structural severity of these stress levels cannot be conclusively established. Future work incorporating such comparisons would allow a more complete assessment of the practical structural consequences of the observed stress amplification.

4.2 Implications for topology-optimized AM components

These findings underscore a key challenge in manufacturing topology-optimized parts using SLM: Topology-optimized geometry is exceptionally sensitive to deviations in curvature continuity. Surface perturbations in high-stress regions, particularly those shaped by support structures, directly undermine the stress-flow mechanisms established by the optimization process. Unlike conventionally machined parts with redundant material or generous safety margins, topology-optimized designs rely on precise geometric fidelity to maintain structural efficiency. Disruptions to these optimized contours lead to disproportionate increases in stress concentration and may compromise fatigue life or long-term durability.

The scan-to-FEM workflow employed here provides a systematic means of quantifying these influences. By comparing the as-built structural response directly with the nominal design, the workflow identifies stress-path disruptions that are not apparent from deviation metrics alone. It also highlights where compensatory design or orientation strategies may be required to mitigate support-induced distortion in future builds.

The present results, therefore, confirm that the choice of SLM to manufacture a topology-optimized component does introduce a heightened susceptibility to stress amplification—not as an inherent deficiency of SLM in isolation, but as a consequence of the reduced geometric redundancy that topology optimization imposes. Any manufacturing process that cannot perfectly reproduce the optimized geometry will produce a qualitatively similar effect; SLM amplifies this risk because its characteristic deviations are spatially concentrated in curvature-sensitive and support-affected regions that are structurally critical in topology-optimized architectures. This susceptibility is precisely what motivates the deviation-aware scan-to-FEA verification workflow proposed in the present study.

4.3 Limitations and future work

While the proposed scan-to-FEA workflow provides clear insight into deviation-induced stress amplification, the present study has several limitations that define its scope and suggest directions for future work:

  • This study focuses on a single topology-optimized connecting rod, manufactured in one SLM build and evaluated under one nominal compressive load case; future work should include multiple builds, load cases, and geometries to assess variability.

  • Only Maraging steel 300, one build orientation, and one support strategy were considered; extending the workflow to different materials, orientations, and process parameters would clarify how general the observed stress amplification is.

  • The FEA is restricted to a static linear-elastic framework and therefore does not capture SLM-induced residual stresses, local elastoplastic deformation, or fatigue damage evolution under cyclic service conditions. Accordingly, the reported stress levels and the inferred risk of local yield should be interpreted as comparative indicators within the present modelling assumptions rather than as a complete prediction of in-service structural reliability. Incorporating residual-stress-aware, nonlinear, and fatigue analyses in future work would provide a more comprehensive assessment of structural performance.

  • The scan-to-FEA pipeline was demonstrated using a single low-cost structured-light scanner and one reconstruction route; comparing different metrology systems and automated reconstruction workflows could quantify their impact on deviation and stress predictions.

  • Deviation information is used only in post-processing; future work could integrate measured or predicted deviations directly into topology optimization (deviation-aware TO and support-structure design) to improve as-built stress flow and reduce amplification.

5 Conclusion

This work demonstrated an end-to-end methodology that connects TO, SLM, high-resolution 3D scanning, deviation quantification, and FEA. By examining the as-built geometry of a topology-optimized connecting rod and comparing it directly with the nominal CAD design, the study revealed how seemingly small manufacturing deviations can produce meaningful shifts in stress distribution. The findings underline an essential reality of metal additive manufacturing: the relationship between geometric fidelity and structural performance is highly nonlinear, particularly for optimized geometries that rely on smooth stress-flow paths, and the key findings include:

  • ICP alignment behavior: The scan-to-CAD registration converged at approximately the 24th iteration, after which successive updates became negligible, confirming stable rigid alignment prior to deviation analysis.

  • Deviation distribution: Signed surface deviations remained concentrated between − 0.20 mm and + 0.30 mm, but outliers up to ± 0.83 mm were observed in thermally sensitive and support-contact regions.

  • Localized deviation hotspots: Support removal zones, including the big end fillets, the bore edges, and the underside shank surfaces, exhibited the most significant geometric departures, consistent with melt pool oversizing and curvature deformation during selective laser melting.

  • Volume increase: The as-built part exhibited a 3.10% increase in volume relative to the CAD model, driven by swelling and surface build-up in high-heat regions.

  • Stress-path distortion: Iso-stress trajectories in the as-built model showed fragmentation, vector rotation, and clustering in regions with geometric irregularities, whereas the CAD model maintained smooth stress-flow corridors.

  • Stress amplification: Under the present static linear-elastic loading condition, the as-built geometry demonstrated an approximately 17.5% increase in peak von Mises stress relative to the CAD model, indicating that measured geometric deviations can materially influence structural response in this specific case study.

  • TO sensitivity: The results suggest that topology-optimized geometries are sensitive to local geometric fidelity, particularly in curvature-critical and support-affected regions, where even modest deviations may produce disproportionately large mechanical effects.

Collectively, these findings highlight the importance of incorporating as-built geometry into the structural verification of topology-optimized SLM components. The scan-to-FEM methodology presented here provides a robust and reproducible means of capturing deviation-driven performance changes, paving the way for deviation-aware optimization, compensation strategies, and improved manufacturability of next-generation lightweight metal components.