High-fidelity Data Research Articles

Instability-wave prediction in hypersonic boundary layers with physics-informed neural operators

Fast and accurate prediction of the nonlinear evolution of instability waves in high-speed boundary layers requires specialized numerical algorithms, and augmenting limited observation in this extreme flow regime is challenging. The deep operator networks (DeepONet) has been shown to be an effective tool for providing accurate and fast physics-informed predictions. DeepONet is trained to map an incoming perturbation to the associated downstream flow field within the nonlinear flow regime. The training is performed using high-fidelity data from direct numerical simulations of the compressible Navier–Stokes equations, when the gas can be approximated as calorically perfect and when chemical non-equilibrium effects must be computed. The performance and requirements of training the DeepONet in each case are evaluated. In addition, we show that informing the training of the DeepONet with the continuity equation improves the accuracy of the results, especially in absence of sufficient training data. Success of the physics-informed DeepONet to predict missing fields depends on the observables. Specifically, prediction of a unique solution depends on the available measurements. These results are a promising step towards applications of neural operator networks to more complex high-speed flow configurations and to data assimilation.

Sterile surgical supply waste identification using asynchronous analysis: Pediatric surgery QI pilot

BackgroundThe operating room (OR) is a major cost and revenue center for a hospital. One of the few modifiable costs in the OR is single-use, sterile surgical supplies (SUSSS). If SUSSS are opened on the scrub table and not used, then they are wasted. High-fidelity SUSSS usage data is important to strategically implement solutions to reduce waste of SUSSS in the OR. OR waste reduction may decrease health systems' carbon footprints and reduce spending. MethodsA convenience sample of general pediatric surgical cases was observed in summer 2021. HIPAA-free images of the surgical scrub table were acquired every 2 s with minimal impact on pediatric OR workflow. These images were asynchronously analyzed to obtain SUSSS usage data for each case. ResultsImage data from three pediatric surgeons performing 41 pediatric surgeries was reviewed. The median cost of unused SUSSS was $13.10 (IQR = $2.73–$47.97) with a range of $0.07 to $489.08 wasted in a single surgery. The mean number of items wasted was 9.3 ± 6.4. The most frequently wasted items were sutures, syringes, towels, paper rulers, and specimen cups. The most expensive sources of waste were laparoscopic trocars, sutures, insufflation needles, drapes, and guidewires. ConclusionsSUSSS that were discarded without being used were successfully identified through the asynchronous analysis of HIPAA-free OR scrub table image data. This may be an opportunity to identify SUSSS waste efficiently without an observer in the OR.

Multifidelity uncertainty quantification with models based on dissimilar parameters

Multifidelity uncertainty quantification (MF UQ) sampling approaches have been shown to significantly reduce the variance of statistical estimators while preserving the bias of the highest-fidelity model, provided that the low-fidelity models are well correlated. However, maintaining a high level of correlation can be challenging, especially when models depend on different input uncertain parameters, which drastically reduces the correlation. Existing MF UQ approaches do not adequately address this issue. In this work, we propose a new sampling strategy that exploits a shared space to improve the correlation among models with dissimilar parameterization. We achieve this by transforming the original coordinates onto an auxiliary manifold using the adaptive basis (AB) method (Tipireddy and Ghanem, 2014). The AB method has two main benefits: (1) it provides an effective tool to identify the low-dimensional manifold on which each model can be represented, and (2) it enables easy transformation of polynomial chaos representations from high- to low-dimensional spaces. This latter feature is used to identify a shared manifold among models without requiring additional evaluations. We present two algorithmic flavors of the new estimator to cover different analysis scenarios, including those with legacy and non-legacy high-fidelity (HF) data. We provide numerical results for analytical examples, a direct field acoustic test, and a finite element model of a nuclear fuel assembly. For all examples, we compare the proposed strategy against both single-fidelity and MF estimators based on the original model parameterization.

Machine-learning data-driven modeling of laminar-turbulent transition in compressor cascade

Laminar-turbulent transition plays a critical role in the aerodynamics of axial-flow compressor cascades. The Reynolds-averaged Navier–Stokes method is so far the most practicable and popular means for numerical simulations of transitional flows to support aerodynamic analysis and the design of compressor cascades. However, the prediction accuracy of the conventional transport equation-based transition models has reached a plateau. In the present work, a machine-learning data-driven transition modeling method that can take full advantage of high-fidelity simulation data is proposed. The turbulence intermittency is calculated algebraically from local flow quantities through a neural network. The proposed method is then applied to construct an algebraic transition model, which is tailored for compressor cascades and coupled with the Spalart–Allmaras turbulence model. The validation results show that the constructed transition model is able to predict flows in compressor cascades with transition in both the Kelvin–Helmholtz instability-induced and bypass modes. Furthermore, the constructed transition model exhibits higher prediction accuracy for both the transition modes than the conventional intermittency factor equation-based transition model. This work demonstrates the effectiveness and promising prospect of machine-learning and data-driven methods in the modeling of complex flow physics.

Study on combined stress failure envelope of CMG based on PSO-BP neural network

The investigation of the failure envelope of control moment gyroscopes (CMGs) under simulated dynamic and thermal vacuum conditions on the ground is crucial for ensuring the reliability of CMGs. In this study, a test rig is employed to simulate the combined stresses in a vacuum environment, including the temperature, CMG gimbal rotating speed, and satellite rotating speed. The objective is to obtain high-fidelity running status data of the CMG. The particle swarm optimization and BP neural network (PSO-BP) model is utilized to learn from these test data and subsequently predict the running status for other stress combinations, ultimately enabling the determination of the failure envelope surface of the CMG. The results demonstrate that this method significantly reduces the cost of testing to detect the CMG failure envelope while achieving high prediction accuracy and adaptability under combined stress situations. By employing the k-fold cross-validation method, the PSO-BP model demonstrates superior generalization performance to the BP model in predicting the running status of CMGs. The optimized hidden layer size and learning rate of the PSO-BP model are also discussed. Furthermore, the method presented in this paper can effectively incorporate the hidden experience data from the test. Finally, the PSO-BP method, in comparison to the BP method, yields a smooth and stable initial predicted value, effectively preventing the prediction result from falling into local optimization.

Opportunities and Challenges of Synthetic Data Generation in Oncology.

Widespread interest in artificial intelligence (AI) in health care has focused mainly on deductive systems that analyze available real-world data to discover patterns not otherwise visible. Generative adversarial network, a new type of inductive AI, has recently evolved to generate high-fidelity virtual synthetic data (SD) trained on relatively limited real-world information. The AI system is fed with a collection of real data, and it learns to generate new augmented data while maintaining the general characteristics of the original data set. The use of SD to enhance clinical research and protect patient privacy has drawn a lot of interest in medicine and in the complex field of oncology. This article summarizes the main characteristics of this innovative technology and critically discusses how it can be used to accelerate data access for secondary purposes, providing an overview of the opportunities and challenges of SD generation for clinical cancer research and health care.

Physics-constrained neural networks with minimax architecture for multiphysics dendritic growth problems in additive manufacturing

Data sparsity is the main barrier to apply deep neural networks to solve complex scientific and engineering problems, where it is expensive to obtain a large amount of high-fidelity training data from experiments or simulations. A new approach of physics-constrained neural networks (PCNNs) was recently developed to alleviate the data sparsity issue by applying the physics-based models as the constraints to guide the training. Constraints are incorporated as different loss terms in the total loss function. However, how to ensure the convergence during the training is still a challenge for PCNNs to predict complex multiphysics phenomena. The causes are the unbalanced gradients from different losses, unstable training resulting from tightly coupled multiphysics, and high dimensionality of the parameter space. In this work, a new sequential training scheme is proposed to accelerate the convergence of our previously developed physics-constrained neural networks with minimax architecture (PCNN-MMs) for solving multiphysics problems. A new training algorithm called the Dual-Dimer with compressive sampling (DD-CS) algorithm is developed to improve the convergence of the training by escaping the local minima. The novel training scheme and algorithm for PCNN-MMs are demonstrated with two examples of dendritic growth in metal additive manufacturing. The comparison shows that the sequential training scheme enables better convergence than the original concurrent training scheme. The DD-CS algorithm has the potential to improve the convergence of the training by escaping the local minimum and converging to a better one after proper parameter tuning.

Recent advances in our understanding of NEC diagnosis, prognosis and surgical approach.

Necrotising enterocolitis (NEC) remains a devasting condition that has seen limited improvement in outcomes in recent years. The incidence of the disease is increasing as more extremely premature infants survive. NEC is responsible for 1 in 10 neonatal deaths and up to 61% of survivors have significant neurodevelopmental delay. The aim of this review is to highlight recent advances in diagnosis, prognosis and surgical approach in this condition. Many recent studies have reported novel methods of diagnosis of NEC with the aim of earlier and more accurate identification. These include imaging and machine learning techniques. Prognostication of NEC is particularly important to allow earlier escalation of therapy. Around 25% of infants with NEC will require surgery and recent data has shown that time from disease onset to surgery is greater in infants whose indication for surgery is failed medical management, rather than pneumoperitoneum. This indication was also associated with worse outcomes compared to pneumoperitoneum. Ongoing research has highlighted several new methods of disease prognostication which includes differentiating surgical from medical NEC. Finally, recent randomised controlled trials in surgical technique are discussed along with the implications of these for practice. Further, high quality research utilising multi-centre collaborations and high fidelity data from electronic patient records is needed to address the issues discussed and ultimately improve outcomes in NEC.

Bayesian Model Calibration Using High-Fidelity Simulations of a Mach 8 Scramjet Isolator and Combustor

ABSTRACT The development of complex physical systems, such as scramjets, requires comprehensive assessments of the relevant performance envelopes. Because high-fidelity simulations and experiments are generally too expensive to perform over a wide range of operating conditions, reduced models can be especially useful for exploring the design space. However, these models often involve empirical parameters which must be calibrated using high-fidelity data sets. The present work develops a calibration procedure for a one-dimensional model of the wall pressure in a scramjet flowpath. Using wall pressure measurements from three-dimensional simulations, the six model parameters are tuned using a Bayesian methodology. The results show that the tuned model is largely able to capture the mean wall pressure in both the isolator and combustor. The calibration procedure is able to converge for both uniform and Gaussian priors, provided that the Gaussian priors are sufficiently wide. Physically-consistent correlations between parameters are obtained, which could inform further reductions in the model’s dimensionality. Application of the model to additional flow conditions shows good agreement to simulations not used in the calibration procedure.

High-fidelity time-series data synthesis based on finite element simulation and data space mapping

Dynamic responses can provide rich information for supporting the entire life cycle of structures, and they can either be measured from actual structures or simulated using the finite element (FE) method. For the FE simulation, insufficient fidelity of simulation data can significantly affect the confidence of analysis results, while FE model updating methods can partially address this problem by reducing the simulation error. However, most FE model updating methods inevitably update the hyperparameters of FE models using sophisticated algorithms with high computational complexities. Thus, one question was raised: whether there is a mapping that can transfer the FE simulation data to the corresponding measurement data directly without performing FE model updating? To achieve this, we proposed a data synthesis method using FE simulation and deep learning space mapping, which can be used to synthesise high-fidelity dynamic responses excited by some unseen load patterns in the measurement. A Dilated Causal Convolutional Neural Network (DCCNN) was designed for realising space mapping. Vibration experiments were conducted on both an I-shaped steel beam and the corresponding FE model to establish datasets and test the proposed method. The quality of the synthetic data was analysed in both the time domain and the frequency domain. The accurate amplitudes, natural frequencies, and mode shapes of the synthetic data successfully demonstrate the effectiveness of the proposed high-fidelity data synthesis method.

SimCAS: an embedding-based method for simulating single-cell chromatin accessibility sequencing data.

Single-cell chromatin accessibility sequencing (scCAS) technology provides an epigenomic perspective to characterize gene regulatory mechanisms at single-cell resolution. With an increasing number of computational methods proposed for analyzing scCAS data, a powerful simulation framework is desirable for evaluation and validation of these methods. However, existing simulators generate synthetic data by sampling reads from real data or mimicking existing cell states, which is inadequate to provide credible ground-truth labels for method evaluation. We present simCAS, an embedding-based simulator, for generating high-fidelity scCAS data from both cell- and peak-wise embeddings. We demonstrate simCAS outperforms existing simulators in resembling real data and show that simCAS can generate cells of different states with user-defined cell populations and differentiation trajectories. Additionally, simCAS can simulate data from different batches and encode user-specified interactions of chromatin regions in the synthetic data, which provides ground-truth labels more than cell states. We systematically demonstrate that simCAS facilitates the benchmarking of four core tasks in downstream analysis: cell clustering, trajectory inference, data integration, and cis-regulatory interaction inference. We anticipate simCAS will be a reliable and flexible simulator for evaluating the ongoing computational methods applied on scCAS data. simCAS is freely available at https://github.com/Chen-Li-17/simCAS.

Artificial neural network based correction for reduced order models in computational fluid mechanics

In this paper we propose a reduced order model (ROM) for incompressible flows based on proper orthogonal decomposition (POD) and having a finite element approximation as full order model (FOM). The main idea is to start from a purely POD-based ROM, projecting the equations onto the ROM space, and then add a nonlinear correction that depends on the ROM unknowns to enhance the final ROM model. This correction is based on the fact that we do have some available high fidelity data, namely, the snapshots. Thus, the correcting term is built as an artificial neural network (ANN) constructed with the snapshots as the training set. This correction is then introduced in the fully discrete ROM system to achieve more accurate solutions. A further feature of our approach is that we construct both the ROM and the FOM using the variational multi-scale (VMS) concept, and this allows us to understand the correction term as an approximation to the scales that are lost when passing from the FOM to the ROM. The resulting corrected ROM has a significant higher accuracy than the original one.

Application of Broadband Receptivity Data to the Amplitude Method for Transition Prediction

Conventional hypersonic transition prediction relies upon the highly empirical method, which is predicated on theoretical approximations of relative disturbance growth. This does not account for the receptivity mechanism and the high degree of variability present in experimental wind-tunnel conditions, which itself leads to high degrees of variability in transition factors. The amplitude method, which better approximates the broadband nature of boundary-layer disturbances, was previously proposed to account for the effects of receptivity. In this study, high-fidelity simulated second-mode receptivity data for two blunt cones at Mach 10 to broadband disturbances are applied to an iterative approximation of the amplitude method tuned for second-mode dominated flows. The studied geometries consist of 9.525- and 5.080-mm-nose-radius 7 deg half-angle straight cones based on experimental cases from Arnold Engineering Development Center’s wind tunnel 9. Amplitude method predictions show improvement over the accuracy of estimates using standard threshold factors. In particular, the less blunt 5.080 mm cone demonstrates the best agreement due to its stronger second-mode response. The results of this work provide a preliminary framework for applying high-fidelity receptivity simulations to the amplitude method for transition prediction in hypersonic flows.

Efficient and highly continuous chromosome-level genome assembly of the first chameleon genome.

Most amniote genomes are diploid, moderate in size (approximately 1-6 Gbp), and contain a large proportion of repetitive sequences. The development of next-generation sequencing technology, especially the emergence of high-fidelity (HiFi) long-read data, has made it feasible to resolve high-quality genome assembly for non-model species efficiently. However, reference genomes for squamate reptiles has lagged behind other amniote lineages. Here we de novo assembled the first genome from the Chameleonidae family, the panther chameleon (Furcifer pardalis). We obtained telomere-to-telomere contigs using only HiFi data, reaching a contig N50 of 158.72 Mbp. The final chromosome-level assembly is 1.61 Gbp in size and 100% of primary contigs were placed to pseudochromosomes using Hi-C interaction data. We also found that sequencing depth > 30 folds can ensure both the integrity and accuracy of the genome, while insufficient depth led to false increase in genome size and proportion of duplicated genes. We provide a high-quality reference genome valuable for evolutionary and ecological studies in chameleons as well as providing comparative genomic resources for squamate reptiles.

A Bayesian analysis for the quantification of strength model uncertainty factor of ship structures in ultimate limit state

The assessment of ship structural reliability involves the quantification of hull girder ultimate bending capacity. Traditionally, Smith’s method is employed for the probabilistic modeling of capacity. The modeling uncertainty associated with Smith’s method is considered by an independent multiplicative random variable Xr. As real-scale data from hull collapses are not available, the quantification of Xr is usually based on the combination of engineering judgment with more objective information, such as non-linear finite element analysis (NLFEA). In this paper, we propose a Bayesian analysis for the determination of Xr. Specifically, we use Bayesian statistical inference to estimate the parameters that characterize the probabilistic model of Xr based on expert judgment and a limited number of high-fidelity NLFEA data. We demonstrate the applicability of the method on container ships, for which two typical load scenarios are considered: (i) pure hogging moment and (ii) combined hogging moment with local bottom loads. For each load scenario, the corresponding Xr is determined. For cases where specific information from a target ship is available, a classical Bayesian updating scheme is proposed to adjust the distribution of Xr based on the individual ship’s structural characteristics. We present a numerical demonstration for the case of the MOL Comfort at the time of the accident. Finally, the impact of the proposed Xr on the failure probability estimate of a 4,400 TEU and a 9,400 TEU container ship is investigated. We show that using the proposed Xr the estimate of the uncertainty in Smith’s model prediction in pure hogging condition is considerably reduced. In addition, we demonstrate that the action of bottom local loads, which considerably decreases the hull girder ultimate strength and the reliability level of container ships in hogging, can be captured by the proposed Xr without any intervention in the conventional Smith method.

Machine learning from RANS and LES to inform coarse grid simulations

Nuclear system thermal hydraulic analysis has historically relied on computationally inexpensive 1D codes. However, such tools are unable to capture multiscale multidimensional effects in large nuclear reactor enclosures. On the other hand, simulations with higher fidelity can be too expensive for such purposes. One of the ways to reduce computational cost is to perform simulations on a coarse grid, which, unfortunately, introduces large discretization errors.In this paper, two high-to-low data-driven approaches are investigated: (1) a coarse grid turbulence model to predict eddy viscosity and (2) correction of errors in coarse grid velocity fields. The approaches aim to reduce grid- and turbulence model-induced errors in coarse grid Reynolds-averaged Navier–Stokes (RANS) simulations. Two sources of high-fidelity data, RANS and large eddy simulations (LES), are explored.To extract the eddy viscosity from the LES data, an inverse optimization problem is solved. However, the LES eddy viscosity is shown to be comparable to the RANS eddy viscosity in terms of error reduction. Therefore, the directly available RANS eddy viscosity was used to develop a coarse grid data-driven turbulence model. Additionally, error correction in velocity is used to reduce the remaining uncertainties and bring the results closer to reality. The performance of the frameworks is demonstrated for a scaled upper plenum of a gas-cooled reactor facility.

Probabilistic neural data fusion for learning from an arbitrary number of multi-fidelity data sets

In many applications in engineering and sciences analysts have simultaneous access to multiple data sources. In such cases, the overall cost of acquiring information can be reduced via data fusion or multi-fidelity (MF) modeling where one leverages inexpensive low-fidelity (LF) sources to reduce the reliance on expensive high-fidelity (HF) data. In this paper, we employ neural networks (NNs) for data fusion in scenarios where data is very scarce and obtained from an arbitrary number of sources with varying levels of fidelity and cost. We introduce a unique NN architecture that converts MF modeling into a nonlinear manifold learning problem. Our NN architecture inversely learns non-trivial (e.g., non-additive and non-hierarchical) biases of the LF sources in an interpretable and visualizable manifold where each data source is encoded via a low-dimensional distribution. This probabilistic manifold quantifies model form uncertainties such that LF sources with small bias are encoded close to the HF source. Additionally, we endow the output of our NN with a parametric distribution not only to quantify aleatoric uncertainties, but also to reformulate the network’s loss function based on strictly proper scoring rules which improve robustness and accuracy on unseen HF data. Through a set of analytic and engineering examples, we demonstrate that our approach provides a high predictive power while quantifying various sources of uncertainty. Our codes and examples can be accessed via GitLab.

Causality analysis of large-scale structures in the flow around a wall-mounted square cylinder

The aim of this work is to analyse the formation mechanisms of large-scale coherent structures in the flow around a wall-mounted square cylinder, due to their impact on pollutant transport within cities. To this end, we assess causal relations between the modes of a reduced-order model obtained by applying proper orthogonal decomposition to high-fidelity simulation data of the flow case under study. The causal relations are identified using conditional transfer entropy, which is an information-theoretical quantity that estimates the amount of information contained in the past of one variable about another. This allows for an understanding of the origins and evolution of different phenomena in the flow, with the aim of identifying the modes responsible for the formation of the main vortical structures. Our approach unveils that vortex-breaker modes are the most causal modes, in particular, over higher-order modes, and no significant causal relationships were found for vortex-generator modes. We validate this technique by determining the causal relations present in the nine-equation model of near-wall turbulence developed by Moehliset al.(New J. Phys., vol. 6, 2004, p. 56), which are in good agreement with literature results for turbulent channel flows.

Multi-fidelity machine learning based uncertainty quantification of progressive damage in composite laminates through optimal data fusion

Recently machine learning (ML) based approaches have gained significant attention in dealing with computationally intensive analyses such as uncertainty quantification of composite laminates. However, high-fidelity ML model construction is computationally demanding for such high-dimensional problems due to the required large amount of high-fidelity training data. We propose to address this issue effectively through multi-fidelity ML based surrogates which can use a training dataset consisting of optimally distributed high- and low-fidelity simulations. For forming multi-fidelity surrogates of progressive damage in composite laminates, we combine low-fidelity finite element analysis data obtained using Matzenmiller damage model with Hasin failure criteria and high-fidelity finite element analysis data obtained using three-dimensional continuum damage mechanics based model with P Linde’s failure criteria. It is shown that there is a significant computational advantage to using the multi-fidelity surrogate approach as compared to conventional single-fidelity surrogates. Such computational advantage through optimal data fusion without compromising accuracy becomes crucial for the subsequent data-driven uncertainty quantification and sensitivity analysis of composites involving thousands of realizations. Ply orientations come out to be the most sensitive parameters to matrix damage, fibre damage and reaction force in composite laminates. The degree of uncertainty in the output quantities depend on the input-level stochastic variations. For example, a combined stochastic variation of ±10% in material properties and ±10° in ply orientations lead to 1.85%, 16.98% and 11.24% coefficient of variation in the matrix damage, fibre damage and reaction force respectively. In general, the numerical results obtained based on the efficient data-driven approach strongly suggest that source-uncertainty of composites significantly influences the progressive damage evolution and global mechanical behaviour, leading to the realization of the importance of adopting an inclusive analysis framework considering such inevitable random variabilities.

Multi-fidelity prediction of fluid flow based on transfer learning using Fourier neural operator

Data-driven prediction of laminar flow and turbulent flow in marine and aerospace engineering has received extensive research and demonstrated its potential in real-time prediction recently. However, usually large amounts of high-fidelity data are required to describe and accurately predict the complex physical information, while reality, only limited high-fidelity data are available due to the high experimental/computational cost. Therefore, this work proposes a novel multi-fidelity learning method based on the Fourier neural operator by jointing abundant low-fidelity data and limited high-fidelity data under transfer learning paradigm. First, as a resolution-invariant operator, the Fourier neural operator is first and gainfully applied to integrate multi-fidelity data directly, which can utilize the limited high-fidelity data and abundant low-fidelity data simultaneously. Then, the transfer learning framework is developed for the current task by extracting the rich low-fidelity data knowledge to assist high-fidelity modeling training, to further improve data-driven prediction accuracy. Finally, three engineering application problems are chosen to validate the accuracy of the proposed multi-fidelity model. The results demonstrate that our proposed method has high effectiveness when compared with other high-fidelity models and has the high modeling accuracy of 99% for all the selected physical field problems. Additionally, the low-fidelity model without transfer learning has the modeling accuracy of 86%. Significantly, the proposed multi-fidelity learning method has the potential of a simple structure with high precision for fluid flow problems, which can provide a reference for the construction of the subsequent model.