Sneha Prabha Narra

Abstract X-ray Micro Computed Tomography (X-µCT) is increasingly regarded as the gold standard for inspecting additively manufactured components used in fatigue-critical applications. However, segmentation of X-µCT data remains inconsistent across users and applications. Additionally, it is unclear if voxel-wise metrics of segmentation quality, such as the Dice coefficient or Intersection over Union (IoU), are relevant to fatigue performance. In this work, we evaluated global binary thresholding, adaptive thresholding, hysteresis thresholding, and a 2.5D U-Net on X-µCT scans of Powder Bed Fusion – Laser Beam manufactured Ti-6Al-4V rotating bending fatigue specimens from the NIST AMBench 2025 challenge (AMB2025-03-FL) to quantify segmentation-induced measurement bias and assess its impact on predicting the fatigue-initiating pore. To identify the fatigue-initiating pore, the Murakami $$\sqrt{A}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msqrt> <mml:mi>A</mml:mi> </mml:msqrt> </mml:math> parameter was modified using beam theory to account for the stress gradient due to bending. This modification enables localization of the most critical pore, independent of the applied stress. Using the modified $$\sqrt{A}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msqrt> <mml:mi>A</mml:mi> </mml:msqrt> </mml:math> metric, the global binary thresholding, adaptive thresholding, and hysteresis thresholding predicted the fatigue-initiating pore correctly for three out of the four test cases. The 2.5D U-Net was able to predict the fatigue-initiating pore in all four cases despite having a lower Dice coefficient and IoU values when compared to the other segmentation models. In fatigue-critical applications, these limitations are most consequential for small, near-surface pores, where X-ray reflection artifacts cause threshold-based methods to underestimate pore size. Such pores can occur even in high-density PBF-LB parts. Consequently, voxel-wise metrics such as Dice coefficient or IoU do not indicate whether a segmentation approach can identify the true fatigue-initiating pore, as they weight all pixels equally, highlighting the need for fatigue-aware, feature-based methods for segmentation of X-µCT data.

Dataset for NASA Award No. 80NSSC25PA713

Dataset for NASA Award No. 80NSSC25PA713

Metal additive manufacturing (AM) enables rapid, on-demand production of parts ranging from prototypes to mission-critical components. However, achieving high-strength metallic components often relies on post-processing heat treatments for many alloys, adding time and cost. For instance, in Alloy 718, a common high-strength alloy used in laser powder bed fusion (L-PBF) AM, strengthening primarily relies on precipitation hardening during an aging heat treatment. Importantly, heating during the deposition process can also elevate temperatures into the precipitation hardening range, causing in-situ aging. Thus, this work leverages the elevated temperatures during fabrication to enable controlled in-situ aging that increases as-fabricated hardness and improves uniformity. For the first time, this is realized through full-part thermal history control during L-PBF fabrication of Alloy 718. The method embeds an experimentally fitted material hardening model in an axisymmetric lumped-layer thermal simulation to predict in-situ part hardness. The resulting thermal and hardness dynamics model is then used in conjunction with a trajectory optimization algorithm to determine time-varying laser power and baseplate temperature profiles. These optimized process conditions target a uniform hardness of 450 HV in an inverted cone geometry by intentionally inducing in-situ precipitation hardening. The planned trajectory increased the mean hardness and improved uniformity from 374 ± 41 HV to 439 ± 29 HV without a separate post-process aging heat treatment. These results are repeatable within 8% and establish a path to integrate microstructural aging control directly into the deposition step to achieve high-strength metallic components without an additional post-process heat treatment step. • Demonstrated high-throughput time–temperature-hardness characterization methodology. • Applied a modified Avrami equation to model continuous cooling transformations. • Predicted part hardness during printing using thermal history simulations. • Determined optimal power and baseplate temperature to maximize hardness.

This dataset contains our segmentation results and models for segmenting X-ray Computed Tomography Scans of Laser-Powder Bed Fusion Additively Manufactured Rotating Bending Fatigue Specimens as part of the NIST AMBench 2025-03 Challenge. The initial dataset containing the raw scans and the segmentation method that was provided is available at https://doi.org/10.18434/mds2-3734.

This dataset contains our segmentation results and models for segmenting X-ray Computed Tomography Scans of Laser-Powder Bed Fusion Additively Manufactured Rotating Bending Fatigue Specimens as part of the NIST AMBench 2025-03 Challenge. The initial dataset containing the raw scans and the segmentation method that was provided is available at https://doi.org/10.18434/mds2-3734.

Advanced process monitoring and model validation are essential for improving weld quality in both welding and welding-based additive manufacturing processes. Specifically, temperature is a key quantity of interest for understanding defect formation and microstructural evolution, which significantly impact mechanical properties. However, achieving accurate in-situ temperature imaging is challenging due to emissivity variations across the dynamic melt pool. To address this, we implemented a two-color imaging technique using a single commercial color camera to reduce temperature readings’ sensitivity to emissivity variations. High dynamic range images during melting were captured at various exposure times, and spatial and temporal filters were applied to minimize interference from the plasma arc emissions. The resulting temperature fields within the melt pool were then utilized to estimate cooling rates, which were further correlated to ex-situ hardness measurements. The strong correlation observed between cooling rates ranging from 20 to 600 K/s and hardness ranging between 250 to 400 HV demonstrated the potential of our easy-to-use two-color thermal imaging setup for preliminary evaluation of mechanical properties in a non-destructive manner. Beyond its significance for predicting mechanical properties, this technique provides a validated temperature measurement approach that can enhance the accuracy of physics-based models, such as those used to predict defect formation mechanisms, like porosity.

This dataset reports the characterization and data processing methodology of 45 individual AISI 316L single melt tracks, fabricated by powder blown laser beam directed energy deposition (DED-LB) metal additive manufacturing. The melt tracks were deposited across a parametric combination of process parameters: powder size distributions, carrier gas flow rates, and laser spot diameter-laser power sets. The measured melt track properties include the average melt track width, height, cross-sectional area, and the powder catchment efficiency. Optical profilometry was used to extract the melt track dimensions and to calculate the powder catchment efficiency. In addition, the corresponding particle stream spatial distributions and particle velocity distributions were measured across the deposition flow parameters by processing high-speed image data. The median particle Stokes number for each flow condition was reported for comparability with other discrete coaxial nozzle systems with particle-laden flows. This dataset can aid in the validation of computational simulations of particle-laden flows from three-jet nozzle systems and the validation of DED-LB models which predict the melt track properties from known process parameters.

Metal additive manufacturing (AM) opens the possibility for spatial control of as-fabricated microstructure and properties. However, since the solid state diffusional transformations that drive microstructure outcomes are governed by nonlinear ODEs in terms of temperature, which is itself governed by PDEs over the entire part domain, solving for the system inputs needed to achieve desired microstructure distributions has proven difficult. In this work, we present a trajectory optimization approach for spatial control of microstructure in metal AM, which we demonstrate by controlling the hardness of a low-alloy steel in electron beam powder bed fusion (EB-PBF). To this end, we present models for thermal and microstructural dynamics. Next, we use experimental data to identify the parameters of the microstructure transformation dynamics. We then pose spatial microstructure control as a finite-horizon optimal control problem. The optimal power field trajectory is computed using an augmented Lagrangian differential dynamic programming (AL-DDP) method with GPU acceleration. The resulting time-varying power fields are then realized on an EB-PBF machine through an approximation scheme. Measurements of the resultant hardness shows that the optimized power field trajectory is able to closely produce the desired hardness distribution.

X-ray micro computed tomography (X- μ CT) of bulk powder bed fusion - laser beam (PBF-LB) Ti-6Al-4V samples shows that, within the optimal process window – where lack-of-fusion and keyhole porosity are minimized – higher laser power reduces the number density of powder-entrapped pores when hatch spacing, layer thickness, and laser spot size remain fixed. To gain insight into this observation, the X- μ CT measurements of powder-entrapped pores are combined with a computational model to simulate pore trajectories in the PBF-LB melt pool. More than 100,000 independent pore trajectories are simulated at two different combinations of laser power and scanning velocity, where the forces acting on the pores are quantified using melt pool temperatures, pressures, and fluid flow velocities from multi-physics simulations. The model is then used to predict the pore size distributions in bulk samples fabricated within the optimal process window at 150 W, 700 mm/s and 370 W, 1200 mm/s. At both laser power settings, the total number density of pores predicted by the model is within one order of magnitude of the experimental values. The model suggests that the differences in the pore size distributions measured with X- μ CT are caused by differences in melt pool overlap (i.e., remelting). Using the model, a process map is constructed to predict porosity as a function of hatch spacing and layer thickness, suggesting that the number density of powder-entrapped pores can vary by two orders of magnitude within the optimal process window. This result suggests that the elimination of powder-entrapped pores poses an obstacle to increasing build rates by increasing the hatch spacing and layer thickness. While previous investigations of pore evolution during PBF-LB focused on experimental approaches, this work will enable the development of model-driven processing strategies to promote pore elimination.

Time-dependent partial differential equations (PDEs) for classic physical systems are established based on the conservation of mass, momentum, and energy, which are ubiquitous in scientific and engineering applications. These PDEs are strictly local-dependent according to the principle of locality in physics, which means that the evolution at a point is only influenced by the neighborhood around it whose size is determined by the length of timestep multiplied with the speed of characteristic information traveling in the system. However, deep learning architecture cannot strictly enforce the local-dependency as it inevitably increases the scope of information to make local predictions as the number of layers increases. Under limited training data, the extra irrelevant information results in sluggish convergence and compromised generalizability. This paper aims to solve this problem by proposing a data decomposition method to strictly limit the scope of information for neural operators making local predictions, which is called data decomposition enforcing local-dependency (DDELD). The numerical experiments over multiple physical phenomena show that DDELD significantly accelerates training convergence and reduces test errors of benchmark models on large-scale engineering simulations.

Fatigue performance in Powder Bed Fusion – Laser Beam is influenced by the largest pore size within the stressed volume, which correlates with fatigue life in porosity-driven failures. However, single value estimates for the largest pore size are insufficient to capture the experimentally observed scatter in fatigue properties. To address this gap, in this work, we incorporate uncertainty quantification into extreme value statistics to estimate the largest pore size distribution in a given volume of material by capturing uncertainty in the number of pores present and the distribution parameter estimates. We then applied this statistical framework to compare the porosity equivalence between two geometries: a 4-point bend fatigue specimen and an axial fatigue specimen in the gauge section. Both geometries were manufactured with the same process conditions using Ti-6Al-4V, followed by porosity characterization via X-ray Micro CT. The results show that the largest pore size distribution of the 4-point bend specimen is insufficient to accurately capture the largest pore size observed in the axial fatigue specimen, despite similar dimensions. Our findings highlight the need for rigorous statistical analysis to quantify the differences between porosity distributions.

Powder catchment and melt track height are foundational for build planning in powder blown laser beam directed energy deposition. However, the interconnected relationships of the catchment efficiency with laser parameters, powder size distribution, and carrier gas flow rate make build planning across machines and feedstock challenging without trial-and-error verification. The primary geometry-based catchment model from laser cladding assumes that this relationship is captured through knowledge of the particle stream and laser spot diameter. First, this work evaluated the applicability of the geometry-based catchment model across a range of AISI 316L powder sizes, carrier gas flow rates, and laser spot diameters. By measuring particle stream diameters from high-speed imaging, the geometry-based catchment model predicted catchment with a root mean squared error of 11.5% for single melt tracks. Second, recognizing the burden of deploying high-speed imaging, this work utilized Stokes number for rapid catchment prediction in place of the particle stream diameter. This approach predicted the catchment with a root mean squared error of 11.8%. Finally, the predicted catchment and laser spot diameter was used to predict the average melt track height with a root mean squared error below 75 μ m. Thus, end-users can apply this Stokes number-based approach to accelerate build planning when using AISI 316L feedstock. • Catchment efficiency was measured across powder, carrier gas, and laser parameters. • Ratio of laser spot to particle stream width predicted catchment with RMSE of 11.5%. • The Stokes number enabled catchment prediction without the use of high-speed imaging. • The Stokes number-based model predicted the melt track height with RMSE below 75 μ m.

Abstract Additive manufacturing enables the fabrication of multi-lattice structures, an advanced design approach featuring heterogeneous lattices at the mesoscale which are arranged to achieve a diverse and purposeful distribution of material properties at the macroscale. Compared to uniform lattice structures, multi-lattice structures permit greater design freedom and a larger design space, which makes it possible to achieve superior structure performance. However, the expanded design space introduces a substantial increase in the complexity that must be managed in order to achieve a multi-lattice structure solution. However, there is a lack of design automation approaches that can tractably create multi-lattice structures. This article introduces an innovative multi-scale topology optimization (TO) framework, called multi-lattice topology optimization with variational autoencoder (MulaTOVA), that is capable of concurrently addressing macro- and mesoscale design requirements. Neural networks (NNs) are employed in this framework to jointly represent the structural topology at the macroscale and the lattice heterogeneity at the mesoscale, enabling simultaneous optimization through the updating of the NNs’ weights. The connectivity between lattices is implicitly constrained by constraining the NNs, while the diversity of the lattices is guaranteed through a generative lattice model which is trained over a large lattice dataset. The performances of various NN types are compared, and Fourier neural operators (FNOs) demonstrated the best flexibility in balancing lattice diversity and local connectivity. Furthermore, our results show that the multi-lattice TO structures achieve a higher stiffness-to-weight ratio than solid TO structures.

Abstract The large temporal and spatial variations in temperature that can occur in layer-wise metal additive manufacturing (AM) lead to thermal excursions, resulting in property variations and defects. These variations cannot always be fully mitigated by simple static parameter search. To address this challenge, we propose a general approach based on modeling AM processes on the part-scale in state-space and framing AM process planning as a numerical optimal control problem. We demonstrate this approach on the problem of minimizing thermal variation in a given layer in the electron beam powder bed fusion AM process, and are able to compute globally optimal dynamic process plans. These optimized process plans are then evaluated in simulation, achieving an 87% and 86% reduction in cumulative variance compared to random spot melting and a uniform power field respectively, and are further validated in experiment. This one-shot feedforward planning approach expands the capabilities of AM technology by minimizing the need for iterative experiments and simulations to achieve process optimization. Further, this work opens the possibility for the application of optimal control theory to part-scale optimization and control in AM.

Previous work on fatigue prediction in Powder Bed Fusion - Laser Beam has shown that the estimate of the largest pore size within the stressed volume is correlated with the resulting fatigue behavior in porosity-driven failures. However, single value estimates for the largest pore size are insufficient to capture the experimentally observed scatter in fatigue properties. To address this gap, in this work, we incorporate uncertainty quantification into extreme value statistics to estimate the largest pore size distribution in a given volume of material by capturing uncertainty in the number of pores present and the upper tail parameters. We then applied this statistical framework to compare the porosity equivalence between two geometries: a 4-point bend fatigue specimen and an axial fatigue specimen in the gauge section. Both geometries were manufactured with the same process conditions using Ti-6Al-4V, followed by porosity characterization via X-ray Micro CT. The results show that the largest pore size distribution of the 4-point bend specimen is insufficient to accurately capture the largest pore size observed in the axial fatigue specimen, despite similar dimensions. Based on our findings, we provide insight into the design of witness coupons that exhibit part-to-coupon porosity equivalence for fatigue.

Metal additive manufacturing (AM) opens the possibility for spatial control of as-fabricated microstructure and properties. However, since the solid state diffusional transformations that drive microstructure outcomes are governed by nonlinear ODEs in terms of temperature, which is itself governed by PDEs over the entire part domain, solving for the system inputs needed to achieve desired microstructure distributions has proven difficult. In this work, we present a trajectory optimization approach for spatial control of microstructure in metal AM, which we demonstrate by controlling the hardness of a low-alloy steel in electron beam powder bed fusion (EB-PBF). To this end, we present models for thermal and microstructural dynamics. Next, we use experimental data to identify the parameters of the microstructure transformation dynamics. We then pose spatial microstructure control as a finite-horizon optimal control problem. The optimal power field trajectory is computed using an augmented Lagrangian differential dynamic programming (AL-DDP) method with GPU acceleration. The resulting time-varying power fields are then realized on an EB-PBF machine through an approximation scheme. Measurements of the resultant hardness shows that the optimized power field trajectory is able to closely produce the desired hardness distribution.

Understanding the temperature history over a part during additive manufacturing (AM) is important for optimizing the process and ensuring product quality, as temperature impacts melt pool geometry, defect formation, and microstructure evolution. While in-process temperature monitoring holds promise for evaluating the part quality, existing thermal sensors used in AM provide only partial measurements of the temperature distribution over the part. In this work, we introduce an innovative approach for reconstructing the complete temperature profile using partial data. We formulate this challenge as an inpainting problem, a canonical task in machine learning which entails recovering missing information across a spatial domain. We present a data-driven model based on graph convolutional neural networks. To train the inpainting model, we employ a finite element simulation to generate a diverse dataset of temperature histories for various part geometries. Cross-validation indicates that the inpainting model accurately reconstructs the spatial distribution of part temperature with strong generalizability across various geometries. Further application to experimental data using infrared camera measurements shows that the model accuracy could be improved by augmenting the training data with simulation data that shares process parameters and geometry with the experimental part. By presenting a solution to the temperature inpainting problem, our approach not only improves the assessment of part quality using partial measurements but also paves the way for the creation of a temperature digital twin of the part using thermal sensors. • Full part temperature history during AM process can be reconstructed by ML models. • A hybrid framework including in-situ sensors, physics-based and ML models is built. • The framework is evaluated in a wire arc additive manufacturing process.

Powder bed fusion-laser beam (PBF-LB) parts experience a significant decline in fatigue performance when process-induced defects are present. In this work, a decline in 4-point bend fatigue life was observed in PBF-LB Ti–6Al–4V coupons fabricated at constant power with increasing scanning velocity and which underwent subsequent stress relief and surface machining. Specifically, the presence of pores that resemble lack-of-fusion (LoF) and a decline in fatigue life were observed at scanning velocities lower than that expected from prior published work. It was hypothesized that this unexpected presence of LoF pores resulted from melt pool geometry variability that was not considered in prior work when the LoF criterion was implemented. Further, these pores can be small in size and infrequent in their occurrence when the melt pool geometry variability is not severe. Such sparse pores are challenging to characterize using conventional 2D characterization methods. This work leverages tall and narrow coupon geometry and high-resolution X-ray micro computed tomography (X- μ CT) to capture LoF porosity. The results show that a modified melt pool overlap-based LoF criterion considering melt pool geometry variability captures the unexpected occurrence of LoF pores observed in X- μ CT. In addition, the LoF percent metric displays a strongly negative correlation with fatigue performance. The insights from this work provide guidance on characterizing melt pool geometry variability across scan lines to systematically evaluate processing parameters that generate LoF pores, which, in turn, could lower fatigue performance.

Abstract Additive manufacturing (AM) technologies are often capable of fabricating geometries that are more complex than traditional manufacturing methods. A notable innovation enabled by AM is the fabrication of multi-lattice structures, an advanced design concept featuring an array of heterogeneous lattices in the mesoscale that are arranged to achieve a diverse distribution of material properties at the macroscale. Compared to uniform lattice structures, multi-lattice structures permit greater design freedom and a larger design space, which makes it possible to achieve superior structure performance. However, the expanded design space introduces a substantial increase in the complexity of multi-lattice structure design. There is still lack of an optimization framework that can maximize the physical properties of the macro-structures through fully exploiting lattice diversity while ensuring lattice connectivity. To solve these challenges, this paper introduces a multi-scale topology optimization (TO) framework for multi-lattice structures which simultaneously optimizes the structure topology at macroscale and the lattice heterogeneity at mesoscale. The distribution of the pseudo-densities and lattice parameters are represented by neural networks (NNs) whose weights and biases are the design variables. The spatial gradients of NN over the physical domain reflect the dissimilarity of adjacent lattices. So, the connection between the lattices can be implicitly constrained by restricting the spatial gradients of NNs. The diversity of the lattices is guaranteed through a generative lattice model which is trained over a large lattice dataset and is embedded into the optimization framework. The performances of various NN types are compared, and we found that Fourier Neural Operators (FNOs) have the best flexibility in balancing the lattice diversity and local connectivity. In the design problems of structural compliance minimization under complex loading conditions, our results show that the multi-lattice TO structures achieve a higher stiffness-to-weight ratio than normal TO structures.

Additive manufacturing (AM) techniques hold promise but face significant challenges in process planning and optimization. The large temporal and spatial variations in temperature that can occur in layer-wise AM lead to thermal excursions, resulting in property variations and defects. These variations cannot always be fully mitigated by simple static parameter search. To address this challenge, we propose a general approach based on modeling AM processes on the part-scale in state-space and framing AM process planning as a numerical optimal control problem. We demonstrate this approach on the problem of minimizing thermal variation in a given layer in the electron beam powder bed fusion (EB-PBF) AM process, and are able to compute globally optimal dynamic process plans. These optimized process plans are then evaluated in simulation, achieving an 87% and 86% reduction in cumulative variance compared to random spot melting and a uniform power field respectively, and are further validated in experiment. This one-shot feedforward planning approach expands the capabilities of AM technology by minimizing the need for experimentation and iteration to achieve process optimization. Further, this work opens the possibility for the application of optimal control theory to part-scale optimization and control in AM.

Processing defects remain the primary cause for fatigue failure of laser beam powder bed fusion (PBF-LB) produced components. Accordingly, process mapping methodologies have been extensively developed to identify optimal processing parameters to avoid defects. For structure-critical applications, it is necessary to validate the defect-based process maps through fatigue testing. We quantify the defect structure (porosity) process map for PBF-LB Ti-6Al-4V based on defect populations and fatigue properties. The defect populations were measured in samples fabricated at constant power and small increments in scanning velocity using X-ray micro-computed tomography and 2D metallography and analyzed using a number density approach. Furthermore, 4-point bend fatigue testing was used to establish stress-cycles to failure properties. Our results reveal distinct defect populations in both keyhole and lack-of-fusion defect regimes, with continuous variation in defect density. The number density-based defect size quantity strongly correlates with process parameters, peak stress, and initiating defect size, offering a quantitative approach to establish process-defect-fatigue relationships. We conclude that the process window exists just as clearly for fatigue as it does for defects, although more sensitive to variability in defects. Consequently, within this fatigue-based process window, one can expect to consistently produce dense components with superior fatigue properties.

We present a machine learning approach that expedites structure–property analysis in materials, bypassing traditional feature extraction and exploratory data analysis techniques. This objective is accomplished by employing a variational autoencoder (VAE) structure that is modified to include a regressor network for property prediction (VAE-Regression). This modification allows for direct linkage of imaged features and quantitative part properties within the VAE latent space. We first demonstrate our approach using 2D optical micrographs and corresponding four-point bend fatigue life data from laser beam powder bed fusion additively manufactured Ti-6Al-4V coupons. The VAE-Regression model extracts spatial features , predicts fatigue life, and identifies features of porosity defect governing fatigue behavior such as pore clusters, pores near sample edges, and jagged pore morphologies. These features corroborate fatigue literature on physics-based modeling and experimentation. We then demonstrate the versatility of our methodology using binder jet additively manufactured WC-Co coupons, where porosity and microstructural discontinuities are known to lower the three-point bend transverse rupture strength , but the interaction between the WC and Co are yet to be completely understood. We attempted to understand these interactions using our VAE-Regression architecture. Within our dataset, we show that coarser WC grains surrounded by larger Co pools indicate lower strength, while finer WC grains with smaller Co pools indicate higher strength. This machine learning approach using image-based data will likely prove to be critical in understanding and identifying structure–property relationships in new materials and manufacturing processes.

Time-dependent partial differential equations (PDEs) for classic physical systems are established based on the conservation of mass, momentum, and energy, which are ubiquitous in scientific and engineering applications. These PDEs are strictly local-dependent according to the principle of locality in physics, which means that the evolution at a point is only influenced by the neighborhood around it whose size is determined by the length of timestep multiplied with the speed of characteristic information traveling in the system. However, deep learning architecture cannot strictly enforce the local-dependency as it inevitably increases the scope of information to make local predictions as the number of layers increases. Under limited training data, the extra irrelevant information results in sluggish convergence and compromised generalizability. This paper aims to solve this problem by proposing a data decomposition method to strictly limit the scope of information for neural operators making local predictions, which is called data decomposition enforcing local-dependency (DDELD). The numerical experiments over multiple physical phenomena show that DDELD significantly accelerates training convergence and reduces test errors of benchmark models on large-scale engineering simulations.

Abstract High-fidelity, data-driven models that can quickly simulate thermal behavior during additive manufacturing (AM) are crucial for improving the performance of AM technologies in multiple areas, such as part design, process planning, monitoring, and control. However, complexities of part geometries make it challenging for current models to maintain high accuracy across a wide range of geometries. In addition, many models report a low mean-square error (MSE) across the entire domain of a part. However, in each time-step, most areas of the domain do not experience significant changes in temperature, except for the regions near recent depositions. Therefore, the MSE-based fidelity measurement of the models may be overestimated. This article presents a data-driven model that uses the Fourier neural operator to capture the local temperature evolution during the AM process. Besides MSE, the model is also evaluated using the R2 metric, which places great weight on the regions where the temperature changes significantly than MSE. The model was trained and tested on numerical simulations based on the discontinuous Galerkin finite element method for the direct energy deposition AM process. The results shows that the model maintains 0.983−0.999 R2 over geometries not included in the training data, which is higher than convolutional neural networks and graph convolutional neural networks we implemented, the two widely used architectures in data-driven predictive modeling.

The purpose of this "Lessons Learned" paper is to investigate how former graduate student leaders can employ their experiences to achieve and excel in service requirements as junior tenure-track faculty members. Research skills, and increasingly teaching ability, have been core to the graduate student curriculum, and match the majority of faculty tenure requirements. However, preparation for the service requirement is often overlooked at both the graduate student and faculty level. While a small part of the overall tenure package, there is an unspoken presumption that faculty members will be able to serve effectively and efficiently. In STEM curricula, the development of interpersonal skills is often overlooked. While this may not be an impediment in research communications, faculty may have a difficult time adapting to highly social university, local community, or governmental service organizations. The authors reflect on how their time as graduate student leaders, in student government, student organizations, and campus committees, influenced their ability to maximize impact while efficiently balancing time spent. The authors' service portfolios span a range of fieldsas student organization advisors, committee members, or advisory board membersin diverse types of institutions (from research universities to undergraduate teaching colleges) and have each balanced their personal and professional goals with their commitments. While not all junior faculty may have comparable graduate student leadership backgrounds, the authors provide broadly applicable suggestions, from one junior faculty member to another, discussing ways to maximize prior experiences to excel in the tenure service requirement category. This "Lessons Learned" paper should be presented as a lightning talk.

My field is materials processing, and research focuses on greenhouse gas emissions reduction, elimination, and drawdown. Current projects aim to reduce vehicle body weight, lower solar cell manufacturing energy use and cost with improved safety, reduce or eliminate aviation greenhouse gas impact, power ships and trains with zero emissions, and improve grid stability as we drive toward 100% renewables. The primary tool for achieving these goals is mathematical modeling of metal processes, particularly electrochemical processes, validated by key experiments. I currently teach Materials Processing, Analytical Methods, and Statics. All of my classes use tests with two sittings, a practice which appears to improve learning outcomes via peer learning between the two sittings, as described by a paper at ASEE 2022. And drawing from 50 years of project based learning scholarship at WPI, most of my classes include a team project, though I haven't yet figured out how to scale this to classes larger than 50 students.

Previous work on additively-manufactured oxide dispersion strengthened alloys focused on experimental approaches, resulting in larger dispersoid sizes and lower number densities than can be achieved with conventional powder metallurgy. To improve the as-fabricated microstructure, this work integrates experiments with a thermodynamic and kinetic modeling framework to probe the limits of the dispersoid sizes and number densities that can be achieved with powder bed fusion-laser beam. Bulk samples of a Ni–20Cr ＋ 1 wt% Y2O3 alloy are fabricated using a range of laser power and scanning velocity combinations. Scanning transmission electron microscopy characterization is performed to quantify the dispersoid size distributions across the processing space. The smallest mean dispersoid diameter (29 nm) is observed at 300 W and 1200 mm/s, with a number density of 1.0 × 1020 m−3. The largest mean diameter (72 nm) is observed at 200 W and 200 mm/s, with a number density of 1.5 × 1019 m−3. Scanning electron microscopy suggests that a considerable fraction of the oxide added to the feedstock is lost during processing, due to oxide agglomeration and the ejection of oxide-rich spatter from the melt pool. After accounting for these losses, the model predictions for the dispersoid diameter and number density align with the experimental trends. The results suggest that the mechanism that limits the final number density is collision coarsening of dispersoids in the melt pool. The modeling framework is leveraged to propose processing strategies to limit dispersoid size and increase number density.

Previous work on additively-manufactured oxide dispersion strengthened alloys focused on experimental approaches, resulting in larger dispersoid sizes and lower number densities than can be achieved with conventional powder metallurgy. To improve the as-fabricated microstructure, this work integrates experiments with a thermodynamic and kinetic modeling framework to probe the limits of the dispersoid sizes and number densities that can be achieved with powder bed fusion-laser beam. Bulk samples of a Ni-20Cr $+$ 1 wt.% Y$_2$O$_3$ alloy are fabricated using a range of laser power and scanning velocity combinations. Scanning transmission electron microscopy characterization is performed to quantify the dispersoid size distributions across the processing space. The smallest mean dispersoid diameter (29 nm) is observed at 300 W and 1200 mm/s, with a number density of 1.0$\times$10$^{20}$ m$^{-3}$. The largest mean diameter (72 nm) is observed at 200 W and 200 mm/s, with a number density of 1.5$\times$10$^{19}$ m$^{-3}$. Scanning electron microscopy suggests that a considerable fraction of the oxide added to the feedstock is lost during processing, due to oxide agglomeration and the ejection of oxide-rich spatter from the melt pool. After accounting for these losses, the model predictions for the dispersoid diameter and number density align with the experimental trends. The results suggest that the mechanism that limits the final number density is collision coarsening of dispersoids in the melt pool. The modeling framework is leveraged to propose processing strategies to limit dispersoid size and increase number density.

Abstract High-fidelity, data-driven models that can quickly simulate thermal behavior during additive manufacturing (AM) are crucial for improving the performance of AM technologies in multiple areas, such as part design, process planning, monitoring, and control. However, the complexities of part geometries make it challenging for current models to maintain high accuracy across a wide range of geometries. Additionally, many models report a low mean square error (MSE) across the entire domain (part). However, in each time step, most areas of the domain do not experience significant changes in temperature, except for the heat-affected zones near recent depositions. Therefore, the MSE-based fidelity measurement of the models may be overestimated. This paper presents a data-driven model that uses Fourier Neural Operator to capture the local temperature evolution during the additive manufacturing process. In addition, the authors propose to evaluate the model using the R2 metric, which provides a relative measure of the model’s performance compared to using mean temperature as a prediction. The model was tested on numerical simulations based on the Discontinuous Galerkin Finite Element Method for the Direct Energy Deposition process, and the results demonstrate that the model achieves high fidelity as measured by R2 and maintains generalizability to geometries that were not included in the training process.

High-fidelity, data-driven models that can quickly simulate thermal behavior during additive manufacturing (AM) are crucial for improving the performance of AM technologies in multiple areas, such as part design, process planning, monitoring, and control. However, the complexities of part geometries make it challenging for current models to maintain high accuracy across a wide range of geometries. Additionally, many models report a low mean square error (MSE) across the entire domain (part). However, in each time step, most areas of the domain do not experience significant changes in temperature, except for the heat-affected zones near recent depositions. Therefore, the MSE-based fidelity measurement of the models may be overestimated. This paper presents a data-driven model that uses Fourier Neural Operator to capture the local temperature evolution during the additive manufacturing process. In addition, the authors propose to evaluate the model using the $R^2$ metric, which provides a relative measure of the model's performance compared to using mean temperature as a prediction. The model was tested on numerical simulations based on the Discontinuous Galerkin Finite Element Method for the Direct Energy Deposition process, and the results demonstrate that the model achieves high fidelity as measured by $R^2$ and maintains generalizability to geometries that were not included in the training process.

Abstract Fatigue failure is a critical performance metric for additively manufactured (AM) metal parts, especially those intended for safety-critical structural applications (i.e., applications where part failure causes system failure and injury to users). This article discusses some of the common defects that occur in laser powder bed fusion (L-PBF) components, mitigation strategies, and their impact on fatigue failure. It summarizes the fatigue properties of three commonly studied structural alloys, namely aluminum alloy, titanium alloy, and nickel-base superalloy.