SmartScan 2.0: Teaching Lasers to Think — How Coupled Thermomechanical Optimization Cuts Residual Stress by 69% in LPBF

University of Michigan engineers upgrade their intelligent scanning framework from thermal-only to coupled thermoelastic — delivering the largest residual stress reductions yet demonstrated for scan sequence optimization in metal 3D printing

Published: August 2026

Source: Additive Manufacturing, Volume 120 (2026), Article 105133

Keywords: LPBF, scan strategy, scan sequence optimization, thermo-mechanical model, residual stress, distortion minimization, SmartScan

Laser powder bed fusion (LPBF) has transformed metal manufacturing. We can now print geometries that would be impossible to cast, forge, or machine — lattices, topology-optimized brackets, conformal cooling channels. But ask any production engineer what keeps them up at night, and you will hear the same answer: residual stress and part distortion.

The root cause is thermal. LPBF subjects material to extreme, localised heating followed by rapid cooling — generating steep temperature gradients that drive differential thermal expansion and contraction. The result is a build-up of residual stress that can warp parts, crack during printing, or fail prematurely in service. Post-build heat treatment helps, but it adds time and cost and does not fully reverse as-built defects like cracking.

The additive manufacturing community has invested enormous effort in optimising process parameters: laser power, scan speed, hatch spacing, layer thickness. But one variable has received surprisingly little systematic attention: scan sequence — the order in which the laser visits each island, vector, or geometric feature within a layer.

Most LPBF scan sequences today are generated by simple rules: random island order, successive chessboard, or 'least heat influence' — a geometry-based heuristic that picks the next island farthest from previously scanned regions. These methods offer measurable improvements over purely random scanning, but they share a fundamental limitation: they are blind to physics. They use geometric distance as a proxy for thermal history, ignoring the actual heat transfer, thermal accumulation, and mechanical constraint conditions that govern stress development.

In 2022, researchers at the University of Michigan introduced SmartScan 1.0 — a fundamentally different approach. Rather than relying on geometric heuristics, SmartScan 1.0 used a physics-based thermal model and control-theoretic optimisation to select the scan sequence that maximises temperature uniformity across each layer. The results were striking: up to 92% reduction in temperature inhomogeneity, 86% reduction in residual stress, and 24% reduction in maximum deformation compared to state-of-the-art heuristic sequences. SmartScan 1.0 was subsequently implemented as a commercial plug-in to industrial slicing software, successfully preventing catastrophic build failure on a complex 3D part.

But SmartScan 1.0 had a blind spot of its own: it optimised only for thermal uniformity, ignoring the mechanical physics that directly produce residual stress and deformation. Its 'optimal' scan sequence was identical regardless of whether a part was clamped on two edges or three — a clear indication that something important was missing.

In their latest work, published in Additive Manufacturing, Chuan He and Chinedum Okwudire introduce SmartScan 2.0 — a major architectural upgrade that replaces the thermal-only objective with a sequentially coupled linear thermoelastic model. For the first time in scan sequence optimisation, the temperature field AND the displacement field are solved together, layer by layer, using a finite difference thermal solver coupled to a finite element mechanical solver.

The objective function has been fundamentally redefined. Instead of minimising temperature non-uniformity (SmartScan 1.0) or a heuristic combination of local temperature gradients weighted by local compliance (SmartScan 2.0 Pre), SmartScan 2.0 minimises a global elastic deformation metric — the L2 norm of the part-wide displacement vector caused by thermal strain. The logic is straightforward: residual stress and permanent distortion are consequences of plastic deformation, which occurs when thermally induced elastic stresses exceed the material's yield strength. By directly minimising the elastic deformation that drives yielding, SmartScan 2.0 attacks the problem at its mechanical root.

The SmartScan Family: Three Generations Compared

Table 1: The three-generation evolution of SmartScan. SmartScan 2.0 uniquely combines coupled thermomechanics with boundary-condition awareness while maintaining real-time computational performance.

A key technical achievement of SmartScan 2.0 is its nondimensionalized formulation — an elegant application of similarity theory that makes the optimisation effectively size-independent. By expressing the governing equations in dimensionless form and matching the key similarity parameters (dimensionless groups), a scaled-down computational model can reproduce the dominant thermomechanical dynamics of a much larger physical system.

The practical impact is dramatic. A high-fidelity physical model with 187,500 elements required prohibitive computation times. Simply coarsening the mesh to 12,100 elements made the problem tractable — but degraded optimisation quality by 41.3% in the mean elastic deformation metric. The nondimensional model, also using approximately 12,000 elements, slashed computation time by 98% while limiting quality degradation to just 11.1%. The message is clear: smart scaling beats brute-force mesh reduction every time.

The team validated SmartScan 2.0 through two complementary experimental campaigns, both conducted on an open-architecture PANDA 11 LPBF system using AISI 316L stainless steel.

Case Study 1: Laser Marking of 2D Metal Plates

Square plates (50 × 50 × 0.6 mm) partitioned into 100 islands were laser-scanned under two different mechanical constraint conditions: (a) left and right edges fully fixed, and (b) left and bottom edges fully fixed. The plates remained clamped during scanning, after which out-of-plane deformation was measured via 3D scanning — both while clamped and after unclamping.

Under the first boundary condition, SmartScan 2.0 reduced maximum out-of-plane deformation by 23.9% vs. SmartScan 1.0 and 30.9% vs. SmartScan 2.0 (Pre), while mean deformation dropped by 49.0% and 31.1%, respectively. Under the second boundary condition, SmartScan 2.0 achieved similar relative improvements — maximum deformation down 24.4% and 16.6%, mean deformation down 35.1% and 37.3%.

Crucially, these measured improvements were directionally consistent with the optimisation predictions: SmartScan 2.0's elastic deformation metric faithfully guided the scan sequence toward lower actual distortion — confirming that the linear thermoelastic proxy, despite its deliberate simplifications (no plasticity, no temperature-dependent properties), captures the essential physics well enough to be an effective optimisation target.

Case Study 2: 3D-Printed Cantilever Beams

Moving from 2D marking to full 3D LPBF printing, three cantilever beams were fabricated — one for each SmartScan variant — using 680 bidirectional hatch vectors at Layer 150. The built parts were detached from their supports and 3D-scanned to quantify post-print deformation. Residual stress was measured via X-ray diffraction (sin²ψ method) on specimens extracted from the solid block portion of each beam.

The headline results: SmartScan 2.0 reduced mean part deflection by 17.4% relative to SmartScan 1.0 and 12.8% relative to SmartScan 2.0 (Pre). Far more impressively, it slashed mean X-direction residual stress by 69.0% compared to SmartScan 1.0, and mean Y-direction residual stress by 60.6% compared to SmartScan 2.0 (Pre). These are not incremental improvements — they represent a step change in the ability of scan strategy alone to mitigate the most persistent quality challenge in LPBF.

Notably, SmartScan 2.0 achieved these gains while reducing total print time by 7.8% relative to SmartScan 1.0 — a rare case where better quality costs less time, not more.

SmartScan 2.0 arrives at a critical moment for the LPBF industry. As build volumes grow and part geometries become more ambitious — larger heat exchangers, thinner aerospace brackets, more complex medical implants — the residual stress problem scales with them. Traditional mitigation strategies (heated build plates, post-process stress relief, support structure optimisation) are reaching their practical limits.

What makes SmartScan 2.0 particularly compelling is its deployability. The per-layer computation time stays below 15 seconds — compatible with the interlayer recoating dwell on commercial machines. SmartScan 1.0 has already been implemented as a plug-in to a commercial slicing software package and licensed to Ulendo Technologies, a University of Michigan spinout. The upgrade path to SmartScan 2.0 is architecturally straightforward: the same control-theoretic optimisation framework is preserved, with the thermal model augmented by the mechanical stiffness assembly and the objective function swapped from thermal uniformity to elastic deformation.

Moreover, the framework is material-agnostic and pattern-agnostic. The paper demonstrates it on both island-based and vector-based scan patterns, and the nondimensionalized formulation is designed to accommodate different thermal and mechanical properties through its similarity parameters.

The authors are transparent about the limitations. The linear thermoelastic model does not capture plasticity, creep, or temperature-dependent material properties — the very nonlinearities that ultimately determine residual stress magnitude. The elastic deformation metric is a proxy, not a direct prediction. Extending the model to incorporate simplified plasticity while preserving real-time computational tractability is the central research challenge ahead.

Other open frontiers include: evaluating the effects of SmartScan 2.0 on microstructure, surface roughness, and porosity (not just residual stress and deformation); validating across a broader range of materials (titanium, aluminium, nickel alloys); and coupling scan sequence optimisation with real-time process parameter tuning — laser power and speed adjusted on the fly — to create a fully closed-loop, multiphysics optimisation framework. The authors also hint at integrating SmartScan with in-situ sensing (thermography, melt-pool monitoring) for adaptive, sensor-driven scan sequence generation.

He and Okwudire have delivered a clean, principled, and experimentally validated advance in LPBF scan sequence optimisation. SmartScan 2.0's core insight — that the objective function should directly target the elastic deformation that precedes plasticity, rather than a thermal proxy — is both physically sound and practically impactful. The 69% residual stress reduction demonstrated on 3D-printed cantilever beams sets a new benchmark for what scan strategy alone can achieve. For an industry where residual stress remains the single largest barrier to reliable, first-time-right LPBF production, that is a contribution of immediate and lasting significance.

Reference: He, C. & Okwudire, C. (2026). SmartScan 2.0: An intelligent scanning approach for reduced residual stress and deformation in LPBF using a coupled linear thermoelastic model. Additive Manufacturing, 120, 105133.