High-fidelity Model Research Articles

Laboratory-Based Micro-X-ray Computed Tomography of Energy Materials at Idaho National Laboratory

AbstractThe Idaho National Laboratory (INL) has implemented laboratory-based micro-X-ray computed tomography in a laboratory equipped for the examination of highly radioactive samples. This capability provides nondestructive three-dimensional volumetric information on samples to inform subsequent traditional destructive examinations as well as real-world inputs for high-fidelity scientific modeling. Samples can be imaged with spatial resolutions ranging from several hundred nm/voxel up to ~ 100 µm/voxel. The best usable spatial resolution achieved to date is 384 nm/voxel with this instrument, while the highest radiological dose rate of a sample imaged is ~ 60 R/h β/γ on contact. Advanced data analysis, including custom tomographic reconstruction and segmentation methods, have also been developed to support this capability. In addition to traditional digital X-ray radiography and tomography, this instrument is also able to visualize in situ tensile and compression testing as well as perform diffraction contrast tomography. This work describes the X-ray computed tomography post-irradiation examination capabilities at INL, as well as detailing a variety of applications this instrument has examined.

Machine learning-based constitutive modelling for material non-linearity: A review

Machine learning (ML) models are widely used across numerous scientific and engineering disciplines due to their exceptional performance, flexibility, prediction quality, and ability to handle highly complex problems if appropriate data are available. One example of such areas which has attracted a lot of attentions in the last couple of years is the integration of data-driven approaches in material modeling. There has been several successful researches in implementing ML-based constitutive models instead of classical phenomenological models for various materials, particularly those with non-linear mechanical behaviors. This review paper aims to systematically investigate the literature on ML-based constitutive models for materials and classify these models based on their suitability for material non-linearity including Non-linear elasticity (hyperelasticity), plasticity, visco-elasticity, and visco-plasticity. Furthermore, we also reviewed and compared the ML-based approaches that have been applied for architectured materials as these groups of materials are designed to represent specific material behaviors that might not exist in classical and conventional material categories. The other goal of this review paper is to provide initial steps in understanding of various ML-based approaches for material modeling, including artificial neural networks (ANN), Gaussian processes, random forests (RF), generated adversarial networks (GANs), support vector machines (SVM), different regression models and physics-informed neural networks (PINN). This paper also outlines different data collection methods, types of data, data processing approaches, the theoretical background of the ML models, advantage and limitations of various models, and potential future research directions. This comprehensive review will provide researchers with the knowledge necessary to develop high-fidelity, robust, adaptable, flexible, and accurate data-driven constitutive models for advanced materials.

AI-Powered Multimodal Modeling of Personalized Hemodynamics in Aortic Stenosis.

Aortic stenosis (AS) is the most common valvular heart disease in developed countries. High-fidelity preclinical models can improve AS management by enabling therapeutic innovation, early diagnosis, and tailored treatment planning. However, their use is currently limited by complex workflows necessitating lengthy expert-driven manual operations. Here, we propose an AI-powered computational framework for accelerated and democratized patient-specific modeling of AS hemodynamics from computed tomography (CT). First, we demonstrate that the automated meshing algorithms can generate task-ready geometries for both computational and benchtop simulations with higher accuracy and 100 times faster than existing approaches. Then, we showthat the approach can be integrated with fluid-structure interaction and soft robotics models to accurately recapitulate a broad spectrum of clinical hemodynamic measurements of diverse AS patients. The efficiency and reliability of these algorithms make them an ideal complementary tool for personalized high-fidelity modeling of AS biomechanics, hemodynamics, and treatmentplanning.

Proof of concept testing of a vascular closure device for use in fetal surgery

Objective Prior clinical findings have demonstrated that maternal laparotomy with trans-amniotic trans-uterine suturing of the fetoscopic port site during in utero myelomeningocele repair reduces the risk of membrane rupture. However, due to laparotomy-associated morbidity, we aimed to explore the feasibility of using a vascular closure device for percutaneous trans-amniotic trans-uterine suturing. Methods This IRB and IACUC-exempt study utilized 2 strategies for proof-of-concept testing of using the Abbott Perclose™ ProStyle™ Device for suture placement; 1. Ultrasound guided application on a high fidelity maternal abdominal uterus model used for fetal procedures and 2. Placement under direct visualization with sheep undergoing cesarean delivery for other research purposes. In the high-fidelity uterus model, the Abbott Perclose™ ProStyle™ device was used to place a transuterine/transamniotic stitch with accompanying video recording of the approach (https://go.screenpal.com/watch/cZfhoDVsYvW password: perclose). Regarding the second approach, at the time of a cesarean section, 12 French Checkflo® cannulas were inserted into the sheep amniotic space via different approaches: (1) Seldinger technique, (2) Seldinger technique insertion of Checkflo® cannula and subsequent use of the Abbott Perclose™ ProStyle™ device to suture the port site after check flow removal, (3) Abbott Perclose™ ProStyle™ device utilized in what is described as a “pre-close” technique, where prior to cannula placement, trans-uterine trans-amniotic stitches are placed followed by the insertion of a 12 French Checkflo® cannula over the same guidewire. Samples of the sutured uterine wall were sent to pathology and H&E staining was performed to assess uterine hole closure and amnion-to-uterus fixation. Results The high-fidelity model confirmed that the Perclose™ ProStyle™ Device was easily visualized by ultrasound and suture deployment was without complication. In the animal model, the Perclose™ ProStyle™ device effectively sutured the amnion to the uterus in both the pre- and post-close approach. The pre-close technique achieved better amnion-to-uterus approximation and more appropriate uterine hole closure. H&E staining revealed that without suturing, amnion separation from the chorion layer occurred, and the uterine hole persisted. The post-close technique showed partial connection between the amnion and chorion, but inadequate uterine hole closure with amnion shift into the defect. Optimal closure, with secure amnion-to-chorion fixation and uterine closure, was achieved through the pre-close technique. Conclusion The Abbott Perclose™ ProStyle™ Device seems to be a feasible device for use of uterine port closure in maternal-fetal surgery, larger animal studies with mid-pregnancy application are needed to further validate or refute these findings.

Decoding complex transport patterns in flow-induced autologous chemotaxis of multicellular systems.

Cell migration via autologous chemotaxis in the presence of interstitial fluid flow is important in cancer metastasis and embryonic development. Despite significant recent progress, our understanding of flow-induced autologous chemotaxis of multicellular systems remains poor. The literature presents inconsistent findings regarding the effectiveness of collective autologous chemotaxis of densely packed cells under interstitial fluid flow. Here, we present a high-fidelity computational model to analyze the migration of multicellular systems performing autologous chemotaxis in the presence of interstitial fluid flow. Our simulations show that the details of the complex transport dynamics of the chemoattractant and fluid flow patterns that occur in the extracellular space, previously overlooked, are essential to understand this cell migration mechanism. We find that, although flow-induced autologous chemotaxis is a robust migration mechanism for individual cells, the cell-cell interactions that occur in multicellular systems render autologous chemotaxis an inefficient mechanism of collective cell migration. Our results offer new perspectives on the potential role of autologous chemotaxis in the tumor microenvironment, where fluid flow is an important modulator of transport.

Functional ultrasound and brain connectivity reveal central nervous system compromise in Trembler-J mice model of Charcot-Marie-Tooth disease

The Charcot-Marie-Tooth-1E (CMT1E) disease is typically described as a peripheral neuropathy in humans, causing decreased nerve conduction, spastic paralysis, and tremor. The Trembler-J (TrJ) mice serve as a high fidelity model of this disease. Here, we use functional ultrasound (fUS) and functional connectivity (FC) to analyze TrJ mice’s brain activity during sensory stimulation and resting state experiments against wild type (WT) mice - the healthy counterpart. fUS is an imaging technique that measures cerebral blood volume (CBV) temporal changes. We study these changes in the primary somatosensory cortex barrel field (S1BF) of both mice populations during periodic vibrissae stimulation, measuring the number of pixels that correlate to the stimulation (i.e., the size of the activation area), the average correlation of these pixels (i.e., the response strength), and the CBV’s rate of change for each stimulation (i.e., the hemodynamic response). Then, we construct a FC matrix for each genotype and experiment by correlating the CBV signals from the eight cortical regions defined by the Paxinos and Franklin atlas. Our results show that TrJ mice have significantly diminished neurovascular responses and altered brain connectivity with respect to WT mice, pointing to central nervous system effects that could shift our understanding of the CMT1E disease.

Sequential adaptive design for emulating costly computer codes

Gaussian processes (GPs) are generally regarded as the gold standard surrogate model for emulating computationally expensive computer-based simulators. However, the problem of training GPs as accurately as possible with a minimum number of model evaluations remains challenging. We address this problem by suggesting a novel adaptive sampling criterion called VIGF (variance of improvement for global fit). The improvement function at any point is a measure of the deviation of the GP emulator from the nearest observed model output. At each iteration of the proposed algorithm, a new run is performed where VIGF is the largest. Then, the new sample is added to the design and the emulator is updated accordingly. A batch version of VIGF is also proposed which can save the user time when parallel computing is available. Additionally, VIGF is extended to the multi-fidelity case where the expensive high-fidelity model is predicted with the assistance of a lower fidelity simulator. This is performed via hierarchical kriging. The applicability of our method is assessed on a bunch of test functions and its performance is compared with several sequential sampling strategies. The results suggest that our method has a superior performance in predicting the benchmark functions in most cases. An implementation of VIGF is available in the dgpsi R package, which can be found on CRAN.

Developing Microsurgery Skills Outside of the Operating Room.

The complex skills required to perform microsurgery are primarily taught in the high-stakes environment of the operating room. However, learners would benefit from developing these abilities in lower-stakes environments beforehand, allowing them to focus on higher-level tasks intraoperatively. This article outlines available resources for developing microsurgical skills outside the operating room and evaluates their alignment with best practices for performance enhancement, thereby identifying ways to improve microsurgical education. A systematic review and web search were performed in April 2024 to identify available microsurgical skills courses. Descriptive data were extracted from these resources, including course objectives, unique features, model used, and curriculum. Literature and web search revealed 7 online video courses addressing microsurgical skill development. These had freely available multimedia content and used low-fidelity models with widely accessible materials. Six offered a curriculum. By contrast, 14 in-person flap and microsurgery courses in the United States were identified. These occurred over 2-5 days, cost thousands of dollars, and used high-fidelity models with guidance from experts. Finally, there were many simulation platforms ranging from low-fidelity synthetic models to cadaveric tissue to high-fidelity live animal models. These also encompassed technology-based practices such as virtual reality. Compared with high-fidelity training, low-fidelity models are more affordable, are reusable, and allow for dedicated educational opportunities that are better aligned with best practices for knowledge and skill acquisition. Consequently, they have the potential to reach a broader range of trainees and accelerate the learning curve, and therefore should be integrated into every microsurgery training program.

Effect of powder and absorptivity assumptions in the high-fidelity modeling of Laser Powder Bed Fusion processes

Laser Powder Bed Fusion (LPBF) technology represents the most promising metal additive manufacturing process to be scaled up to an industrial level for producing full batches of parts within the automotive aerospace and health sectors, among others. Despite this higher technological readiness, imperfections such as warping, porosities or undesired non-equilibrium microstructures can arise unexpectedly due to the complex thermal histories inherent to LPBF. Being able to anticipate to such imperfections is thus of high importance. With this purpose, numerical modeling strategies can be employed to significantly reduce experimental work on the melt pool scale. This allows to quantify the odds and magnitude of these process unwanted effects. More precisely, Computational Fluid Dynamics (CFD) enables high-fidelity models on the laser-matter interaction during LPBF processing. In this work, a modeling approach is presented in which the driving forces for melt formation are included in a CFD workspace, where multiple reflections of the laser on the melt pool Liquid-Gas Interface (LGI) are considered by means of an absorptivity function of the LGI depth. This simplification and the use of adaptive mesh strategies allows the model to be computed within a reasonable time. Results and applicability of the model on the determination of the melting mode (conduction or keyhole) are displayed and compared with experimental measurements of the melt pool width and depth for 316L and Ti64 single tracks produced on a LPBF machine. Likewise, further aspects such as total absorbed power are assessed.

Investigation of thermal and fluid dynamics behaviors of melt pool during laser direct metal deposition on inclined substrates

The laser direct metal deposition (DMD) technique offers an efficient solution for repairing and manufacturing structures with variable curvatures, such as impeller blades. However, how the non-vertical laser irradiation on curved surfaces impacts the thermal and fluid dynamics of the melt pool still remains unclear. In the current research, a high-fidelity multi-physics numerical model is developed to explore the mechanisms of melt pool development and the effects of inclined angle and laser spot diameter on morphological evolution of deposition layers during DMD under non-vertical irradiation on inclined substrates. Results indicate that as the inclination angle increases, deposited track fluctuations become evident, with a critical angle from 40° to 42.5° identified for hump formation. A smaller laser spot diameter increases the likelihood of humps and irregularities, while a larger spot diameter expands the deposition layer and enhances melt pool wettability. This study is expected to provide significant theoretical guidance for selecting and applying process parameters for the fabrication of curved structures by DMD.

Digital Twin Methodology in Food Processing: Basic Concepts and Applications.

The goal of this article is to present a concise review about digital twin (DT) methodology and its application in food processing. We aim to identify the building blocks, current state and bottlenecks, and to discuss future developments of this approach. DT methodology appears as a powerful approach for digital transformation of food production, via integration of modelling and simulation tools, sensors, actuators and communication platforms. This methodology allows developing virtual environments for real-time monitoring and controlling of processes, as well as providing actionable metrics for decision-making, which are not possible to obtain by physical sensors. So far, main applications were focused on refrigerated transport and storage of fresh produces, and thermal processes like cooking and drying. DT methodology can provide useful solutions to food industry towards productivity and sustainability, but requires of multidisciplinary efforts. Wide and effective implementation of this approach will largely depend on developing high-fidelity digital models with real-time simulation capability.

Strain-gradient crystal plasticity model with slip-system level GND tracking: Simulation vs experiment for sequential strain path change in AA6016-T4

Crystal plasticity models that track strain gradients and associated geometrically necessary dislocations (GNDs) typically determine the Nye tensor, mimicking the experimental approach. However, estimating GND densities for each slip-system is then intrinsically ambiguous. This study seeks to build upon current state of the art by quantifying GNDs at the slip-system level in the model using the local strain gradients. The model is exercised by replicating experiments undertaken on AA6016, which are performed under multi-step strain paths; both GND and SSD populations are quantified at various stages using both high resolution EBSD (HREBSD) and XRD. A full 3D volume of the material is extracted using ion beam serial sectioning to enable the creation of a high-fidelity model of the material.The combined modeling and experimental campaigns conclude that: 1) Calculation of the GND content at the slip level gives effectively equivalent GND evolution as the tradition Nye tensor method, but provides the significant advantage of knowing unambiguously the contribution on each slip system. 2) The net hardening predicted by the SGCP model is accurate, including prediction of a rapid increase in hardening (and associated dislocation content) following strain path change; and 3) Comparisons of observed and simulated GND populations reveal that buildup in the real sample is dominated by precipitate distribution rather than by grain boundary (GB) networks; such precipitates are not present in the current model, hence this result was not predicted.

A probabilistic integration of LSTM and Gaussian process regression for uncertainty-aware reservoir water level predictions

ABSTRACT Reservoir-level forecasting, while being crucial for optimal operation, is challenged by complex physical processes and changing climate conditions. Machine learning approaches offer deterministic predictions but often neglect system physics and uncertainty. This article presents a probabilistic data-driven approach combining long short-term memory (LSTM) and Gaussian process regression (GPR) to provide both point forecasts and uncertainty estimates. The hybrid model leverages LSTM’s fitting capabilities with GPR’s robust Bayesian frameworks for uncertainty estimation in non-linear problems, offering accurate predictions without extensive high-fidelity modelling, and avoiding frequent training and parameter optimization. Evaluation with real reservoir data from India shows the model’s superiority over the vanilla LSTM for both univariate and multivariate scenarios. The proposed model achieved a Nash-Sutcliffe efficiency of 0.97 to 0.98, a mean biased error of −0.5634 to −1.0314 for 10-day forecasts, and a continuous ranked probability score of 5.80 and 1.87 for the Bhakra and Pong reservoirs, respectively.

High-fidelity experimental model verification for flow in fractured porous media

Mixed-dimensional mathematical models for flow in fractured media have been prevalent in the modeling community for almost two decades, utilizing the explicit representation of fractures by lower-dimensional manifolds embedded in the surrounding porous media. In this work, for the first time, direct qualitative and quantitative comparisons of mixed-dimensional models are drawn against laboratory experiments. Dedicated displacement experiments of steady-state laminar flow in fractured media are investigated using both high-resolution PET images as well as state-of-the-art numerical simulations.

Research on traffic congestion prediction based on analyzable machine learning

As motor vehicles proliferate around the world, traffic congestion increases, hindering travel, polluting the environment and hampering economic growth. Accurate congestion prediction is the key to optimizing traffic flow. In this study, the interaction between different traffic modes and congestion degree is deeply investigated, and a high-fidelity prediction model is established. The study analyses monthly, weekly, daily and hourly multi-modal traffic and congestion data and identifies congestion patterns. In this paper, Pearson's coefficient is applied to identify the key influencing factors, and congestion is predicted by KNN, logistic regression, decision tree and random forest models. The random forest model reaches 99.88% accuracy after 5-fold cross-validation and grid search optimisation. Further analysis using the Random Forest's feature importance scores revealed that traffic volume was the most important predictor of congestion, while time factors such as time and date played a smaller role. This study not only establishes a robust congestion prediction model, but also emphasizes the importance of traffic volume in congestion prediction. The findings provide valuable insights for traffic management agencies to take proactive measures to alleviate traffic congestion. By harnessing the predictive power of random forest models, these agencies can increase the efficiency of traffic management, improve the overall travel experience, and contribute to sustainable urban development. This study also lays a solid foundation for future research in the field of traffic congestion prediction and management.

Characterizing multi-source heavy metal contaminated sites at the Hun River basin: An integrated deep learning and data assimilation approach

In real-world scenarios involving groundwater contamination, the environmental complexity substantially complicates the tasks of tracing pollution sources and characterizing the features of affected sites. To address these challenges, this study presents an integrated framework that combines deep learning (AR-Net-DA) with data assimilation (ES-MDA). This approach effectively traces pollution sources and characterizes site features using sparse data from complex contamination scenarios. The paper introduces a case study involving multisource heavy metal (manganese) pollution in the Hun River basin, Liaoning Province, China. A high-fidelity model for groundwater flow and solute transport was developed. Subsequently, the innovative convolutional neural network model, AR-Net-DA, was employed to replace traditional process-based groundwater models by dynamically optimizing weights in proximity to various pollution sources. This model was then integrated into the ES-MDA inversion framework to concurrently determine pollution source parameters and the spatial distribution of aquifer permeability fields. The results demonstrate that this coupled inversion framework can accurately pinpoint pollution source locations and their release histories using limited observational data, while also mapping the spatial distribution of hydraulic conductivity fields with enhanced computational efficiency. These findings have significant implications for groundwater resource management, pollution risk control, and the remediation of contaminated sites.

Silicone as a smart solution for simulating soft tissue-an iterative approach to developing a high-fidelity sustainable training model for laparoscopic appendectomy.

Laparoscopic appendectomy (LA) is an effective treatment for the surgical care of appendicitis, with this minimally invasive approach allowing patients to typically spend less time in hospital and promptly return to normal life activities. Residents can acquire the competence and confidence needed in a safe learning environment prior to real patient encounters through simulation-based learning of these techniques. We propose a low cost, sustainable, high fidelity simulation-based training model for LA to compliment regular resident practice of these skills. A team dedicated to developing this surgical simulation training model was established, equipped with the clinical knowledge and model engineering expertise. We used concepts of design-based research (DBR) to iteratively develop this model at key intervals. Our LA training model underwent four stages of model development prior to unified stakeholder consensus that this model was deemed effective and suitable for integration into formative surgical simulation curricula. This model simulates most of the key anatomical structures associated with performing an LA. In order to provide high fidelity haptic feedback, attempts were made to mimic the tensile properties of real tissue using different concentrations of silicone. The model can be utilized with laparoscopic box trainers of various sizes due to its scalability. It cost €9.67 to create, and single use appendix components cost €1.22 to build thereafter. Surgical residents can benefit from the platform that simulation-based education offers to develop the psychomotor skills necessary to perform LA in a safe learning environment. We describe a model for LA, which allows learners to develop their skill proficiency in this area under expert supervision.

Non-intrusive parametric hyper-reduction for nonlinear structural finite element formulations

Model Order Reduction (MOR) is a core technology for the creation of comprehensive executable Digital Twins, since it efficiently reduces the computational burden of high-fidelity models. When dealing with nonlinear structural Finite Element analyses, several Hyper-Reduction (HR) approaches have been developed to reduce the computational cost. Nonetheless, HR approaches are typically intrusive in nature, posing challenges when it comes to integration into existing (commercial) software. Recently, data driven Non-Intrusive MOR methodologies have been proposed. However, these techniques often suffer from overfitting and violate key physics properties, leading to unstable behavior. This work proposes to use Scientific Machine Learning to reintegrate critical stability-preserving physics properties. It introduces a data-driven, physics-augmented, parametric approach that combines Proper Orthogonal Decomposition (POD) with a Partially Input Convex Neural Network (PICNN) architecture. The proposed method effectively reduces the computational burden associated with parametric static nonlinear elastic structural problems while retaining material consistency, hyper-elasticity, and material stability properties in the Reduced Order Model. Numerical validation on several structural models subjected to geometrical and material nonlinearities under static loading conditions demonstrates the effectiveness of the POD-PICNN approach. Additionally, three different sampling strategies have been compared to assess their impact on the method performance. The results emphasize that physics-augmentation is required, as it inherently embeds essential physical constraints into the neural network architecture, ensuring stable and consistent behavior, while highlighting its potential for dynamic and multiphysics applications.

Generative Portrait Shadow Removal

We introduce a high-fidelity portrait shadow removal model that can effectively enhance the image of a portrait by predicting its appearance under disturbing shadows and highlights. Portrait shadow removal is a highly ill-posed problem where multiple plausible solutions can be found based on a single image. For example, disentangling complex environmental lighting from original skin color is a non-trivial problem. While existing works have solved this problem by predicting the appearance residuals that can propagate local shadow distribution, such methods are often incomplete and lead to unnatural predictions, especially for portraits with hard shadows. We overcome the limitations of existing local propagation methods by formulating the removal problem as a generation task where a diffusion model learns to globally rebuild the human appearance from scratch as a condition of an input portrait image. For robust and natural shadow removal, we propose to train the diffusion model with a compositional repurposing framework: a pre-trained text-guided image generation model is first fine-tuned to harmonize the lighting and color of the foreground with a background scene by using a background harmonization dataset; and then the model is further fine-tuned to generate a shadow-free portrait image via a shadow-paired dataset. To overcome the limitation of losing fine details in the latent diffusion model, we propose a guided-upsampling network to restore the original high-frequency details (e.g. , wrinkles and dots) from the input image. To enable our compositional training framework, we construct a high-fidelity and large-scale dataset using a lightstage capturing system and synthetic graphics simulation. Our generative framework effectively removes shadows caused by both self and external occlusions while maintaining original lighting distribution and high-frequency details. Our method also demonstrates robustness to diverse subjects captured in real environments.

Design approach for tilt propellers of UAM/eVTOLs for cruise and hover considering aerodynamic and aeroacoustic characteristics via a multi-fidelity model

The design of quiet and efficient propellers/rotors is driven by the rapid development of Vertical Takeoff and Landing (VTOL) vehicles in urban areas. However, this task is challenging due to the expensive computational cost of propeller aerodynamic and aeroacoustic optimization based on accurate high-fidelity (HF) models. In contrast, many low-fidelity (LF) methods are less costly but also less accurate for complex propeller blade geometries, such as those involving sweep. This work aims to use a multi-fidelity (MF) surrogate model (SM) that combines a small HF data set for accuracy and a larger LF data set to speed up modeling. The MF approach inherits the accuracy of the HF method while retaining the computational efficiency of the LF model. The HF and LF aerodynamic data sets for training the MF model are obtained from a RANS solver and a meshless Large Eddy Simulation (LES) method, respectively. The obtained surface pressure of the blade is then used to calculate the noise signals radiating from the propeller using a Farassat's Formulation 1A (F1A) code. For practical design, we developed an optimization framework that combines these aerodynamic and aeroacoustic solvers, the MF model, design of experiment (DoE), and a multi-objective optimizer. The framework aims to optimize the blade aerodynamic shape in terms of chord, twist, and sweep distributions along the span from a baseline propeller, with objectives of efficiency and noise reduction under thrust constraints for both cruise and hover operating conditions. The obtained optimal designs, in the form of three-dimensional Pareto fronts, increased the aerodynamic efficiency by approximately 2% for cruise and 5% for hover and decreased the overall sound pressure level (OASPL) for hover by about 4 dB from the baseline propeller. The MF surrogate model, which utilizes both HF and LF data, doubled the optimization efficiency compared to the surrogate model based solely on HF data. Furthermore, the model based only on LF data fails to capture the three-dimensional flow effects induced by propeller sweep, which is crucial for reducing propeller noise. The aerodynamic benefits include reduced flow separation near the blade root during hover and more ideal loading distribution during cruise, while the aeroacoustic improvements are demonstrated through the “dephase” effect on noise signals emitted from different parts of the swept blade.