This study presents a surrogate model designed to predict the Nusselt number distribution in enclosed impinging jet arrays, where each jet functions independently and where jets can be transformed from inlets to outlets. While computational fluid dynamics (CFD) simulations can predict heat transfer with high fidelity, their cost prohibits real-time application such as model-based temperature control. To address this, we generate a CNN-based surrogate model that predicts the Nusselt distribution in real time. We train it with data from implicit large-eddy simulations (Re < 2000). We train two distinct models, one for a five by one array of jets (83 simulations) and one for a three by three array of jets (100 simulations). We introduce a method to extrapolate predictions to higher Reynolds numbers (Re < 10,000) using a correlation-based scaling. The surrogate models achieve high accuracy, with a normalized mean average error below 2% on validation data for the five by one jets surrogate model and 0.6% for the three by three jets surrogate model. Their predictions are validated experimentally using temperature measurements. This work provides a foundation for model-based control strategies in advanced thermal management applications.

Data-driven emulation of peak storm surge has emerged as a popular strategy for overcoming limitations arising from the computational burden of high-fidelity hydrodynamic numerical models used within coastal risk assessment applications. The surrogate models (also known as metamodels) used for this emulation are developed using suites of synthetic storm simulations, and once calibrated, can replace the original high-fidelity model to establish predictions for new storms. These predictions pertain to the geographic domain, and therefore nodal locations, covered by the original high-fidelity simulation suite. This creates a two-dimensional space for the peak surge predictions, with one (primary) corresponding to the storm features (i.e., the storm parametric description) and the other (secondary) to the spatial domain. Gaussian Process (GP) techniques have emerged as a widely popular surrogate modeling technique for peak surge emulation. In all GP implementations so far, the spatial variability has been incorporated in the analysis through the metamodel output, considering a multi-output GP implementation. This approach fails to explicitly model spatial dependencies for the peak surge. To address this shortcoming, this study examines an alternative implementation that considers spatial and storm feature variability as part of the metamodel input, establishing a surrogate model that simultaneously predicts the peak storm surge (scalar output) across both the spatial domain and the storm features. For computational tractability, a separable covariance function is considered for the GP, establishing separate kernels for the spatial and storm feature spaces. Particularly for the spatial domain, an adaptive covariance tapering formulation, which infuses sparsity in the corresponding covariance matrix, is adopted to support applications with a large number (in the order of thousands) of nodal locations. A simultaneous calibration approach for the hyperparameters of the separate kernels is further proposed to improve emulation accuracy. Comparisons of computational efficiency and accuracy of the alternative GP implementations are established utilizing the Coastal Hazards System–North Atlantic (CHS-NA) database, with those employing the adaptive covariance tapering formulation evaluated under varying sparsity levels. The case study demonstrates that the simultaneous hyperparameter calibration is beneficial for the separable GP’s predictive accuracy, particularly as it relates to the worst-performing nodes in the domain, and that the imposed sparsity level impacts the separable GP’s ability to model non-stationary spatial trends in the domain. • Gaussian-Process based emulation of peak storm surge is examined • Spatial and storm feature variability are considered as part of the metamodel input. • Separable covariance kernel is adopted, establishing separate kernels for the spatial and storm feature spaces • Adaptive sparse covariance tapering is utilized to accommodate large spatial domain applications • The impact of the sparsity level on the established accuracy for the spatial interpolation is examined.

Crystallization is a key separation and purification method governed by thermodynamics, kinetics, and multiphase fluid dynamics. Supersaturation generation, nucleation, crystal growth, aggregation, and breakage determine particle size and shape and are described using population balance models (PBMs). Coupled with computational fluid dynamics (CFD), PBMs enable spatially resolved predictions of particulate behavior in real reactors. This review outlines fundamental crystallization kinetics and PBM formulations, including nucleation and growth expressions, aggregation and breakage kernels, and moment-based solution methods. It also highlights how multiphase CFD represents turbulence, mixing, and phase interactions in stirred tanks, tubular reactors, static mixers, and impinging jets, supporting process design and scale-up. Key computational challenges include stiffness, micro-mixing, reaction–equilibrium networks, and parallelization, along with cost-reduction strategies such as compartment models and CFD-informed reactor representations. Emerging machine-learning tools accelerate parameter estimation and surrogate modeling, with applications from inorganic precipitation to pharmaceutical crystallization.

Co-optimization of area ratio and material fraction in segmented N-type thermoelectric legs using XGBoost–ANN surrogates

Multi-condition optimization of S-Type hydrokinetic turbine driven by surrogate modeling and ocean current LSTM forecasting

Latent attention operator network with augmented representation for complex PDE systems in intricate geometries

Comparison of biaxial mechanical and microstructural properties between human femoral arteries and surrogate models for stent development.

• Indirect cascading Natech events from volcanic ashes are addressed • A model to estimate filtration system clogging from volcanic ash is developed • A matrix to evaluate air intake systems vulnerability to volcanic ash is proposed • Facilitating evaluations for preventive planning and real-time decision-making • A Monte Carlo approach is proposed for the quantitative assessment of risk due to filter clogging Among other consequences of volcanic activity, recent events confirmed that the hazards caused by volcanic ash have a potential impact also at relevant distances from the emission point. The fallout of volcanic ashes may affect several utilities and services at industrial sites, potentially causing Natech events with relevant end-point consequences, e.g., operational failures, business interruption, and environmental contamination. The present study focuses on the vulnerability of process and instrument air intake utility systems to volcanic ash. A detailed model, based on an in-depth characterization of ash properties, is developed to provide accurate time to clogging estimations under varying conditions. A surrogate model is also proposed to enable a real-time assessment using a limited set of input parameters, supporting both preventive planning and real-time decision-making in emergency management. A tailored risk matrix is developed to provide a scenario-specific vulnerability ranking of critical utilities due to volcanic ash accumulation. A novel quantitative approach for the assessment of risk due to filter clogging has also been developed to support the management of the vulnerabilities and critical scenarios identified by the matrix screening. The analysis of test cases confirmed the value of the novel approach in supporting risk management and resilience against volcanic hazards, aimed at the mitigation of operational disruptions and/or more severe process safety accidents.

Highly accurate datasets from numerical or physical experiments are often expensive and time-consuming to acquire, posing a significant challenge for applications that require precise evaluations, potentially across multiple scenarios and in real-time. Even building sufficiently accurate surrogate models can be extremely challenging with limited high-fidelity data. Conversely, less expensive, low-fidelity data can be computed more easily and encompass a broader range of scenarios. By leveraging multi-fidelity information, prediction capabilities of surrogates can be improved. However, in practical situations, data may be different in types, come from sources of different modalities, and not be concurrently available, further complicating the modeling process. To address these challenges, we introduce a progressive multi-fidelity surrogate model. This model can sequentially incorporate diverse data types using tailored encoders. Multi-fidelity regression from the encoded inputs to the target quantities of interest is then performed using neural networks. Input information progressively flows from lower to higher fidelity levels through two sets of connections: concatenations among all the encoded inputs, and additive connections among the final outputs. This dual connection system enables the model to exploit correlations among different datasets while ensuring that each level makes an additive correction to the previous level without altering it. This approach prevents performance degradation as new input data are integrated into the model and automatically adapts predictions based on the available inputs. We demonstrate the effectiveness of the approach on numerical benchmarks and a real-world air pollution case study, showing that it reliably integrates multi-modal data, mitigates low-fidelity imperfections, and provides accurate predictions, while maintaining performance when generalizing across time and parameter variations.

High-performance chiral metasurface sensors optimized by a target-driven active learning framework

Bayesian optimized adaptive control of injection molding machines for plastic recycling material

Interpretable Graph Convolutional Networks for cardiovascular disease risk prediction in patients with Type 2 Diabetes Mellitus.

Active power regulation is essential for wind farm (WF) to perform power reserve and its delivery. Large WFs are required to have such capability as imposed by the grid codes for power system security. The current practice deloads and controls individual wind turbines (WT) without effective and scalable coordination across a WF. In this paper, we propose a fully distributed control framework for the active power regulation of WF integrating a large number of WTs. The proposed framework optimizes the operating points of individual turbines in the torque–speed plane, given the power commands at the WF level. The nonlinear wind aerodynamic equations are approximated by a surrogate model using a data-driven approach, which leads to a mixed-integer reformulation of the optimization problem that is solved in a fully distributed fashion. With the obtained optimal turbine operating points across the WF, individual WTs track the reference trajectory using the existing torque and pitch actuators and guarantee that the requested power is delivered. The case studies comprehensively evaluate the proposed method for delivering the WF power regulation with the desired speed and scalability to fulfill the technical requirements of service. The adaptive reaction against single turbine failures for plug-and-play operation is demonstrated as well. • Distributed framework proposed for wind farm active power regulation. • Neural network surrogate model enables mixed-integer linear programming. • Approach achieves superior scalability over centralized methods. • Plug-and-play capability demonstrated under turbine failure events.

Hybrid experimental-machine learning framework for media optimization enhances antioxidant production in Micrococcus endophyticus SS-1.

The acceleration of physics-based simulations via machine learning has contributed to applications that rely on rapid predictions, such as model predictive control and real-time optimization. Among physics-based simulations, granular flow simulations are of particular interest due to their ubiquity in industrial processes and the high computational cost of traditional models. In this review article, we comment on the various approaches for creating machine learning-based surrogate models from discrete element method (DEM) simulations. This review is specifically focused on methods that accelerate simulations or those that produce the same outputs as high-fidelity models, such as particle positions or void fraction fields. For each method, the underlying concepts behind it were explained and the research articles that used it were discussed. The strengths and weaknesses of each approach in relation to the others were also outlined. After this, knowledge gaps and limitations in the development of surrogate models for granular flow simulations and their practical implementation in industrial applications are presented. Potential solutions to each limitation were suggested based on developments from adjacent fields of study, and these may be taken as directions for future work.

Towards structure-aware surrogate modeling: explicit region interaction improves knee contact stress prediction in young men.

Deep Operator Networks (DeepONets) have emerged as a powerful surrogate modeling framework for learning solution operators in PDE-governed systems. While their use is expanding across engineering disciplines, applications in geotechnical engineering remain limited. This study systematically evaluates several DeepONet architectures for the consolidation problem. We initially consider three architectures: a standard DeepONet with the coefficient of consolidation embedded in the branch net (Models 1 and 2), and a physics-inspired architecture with the coefficient embedded in the trunk net (Model 3). Results show that Model 3 outperforms the standard configurations (Models 1 and 2) but still has limitations when the target solution (excess pore pressures) exhibits significant variation. To overcome this limitation, we propose a Trunknet Fourier feature-enhanced DeepONet (Model 4) that addresses the identified limitations by capturing rapidly varying functions. We further extend Model 4 to 3D scenarios. Although the computational speedup can be modest in the 1D case (1.5-100 × compared with traditional solvers), the speedup becomes more pronounced in 3D, reaching approximately 1000 × . Leveraging this efficiency, we offer a conceptual demonstration of DeepONet’s potential to accelerate uncertainty quantification in a 3D consolidation problem. Overall, the study highlights the potential of DeepONets to enable efficient, generalizable surrogate modeling in geotechnical applications, advancing the integration of scientific machine learning in geotechnics, which is at an early stage.

Differentiable constitutive modelling of granular media: From surrogate material model in explicit FEM-DEM to automatic elastoplastic calibration

Topology and size optimization of trusses using graph neural networks: Towards efficient surrogate modeling

A strategy to construct physics-based local surrogate models for parametric Stokes flows and coupled Stokes-Darcy systems is presented. The methodology relies on the proper generalized decomposition (PGD) method to reduce the dimensionality of the parametric flow fields and on an overlapping domain decomposition (DD) paradigm to reduce the number of globally coupled degrees of freedom in space. The DD-PGD approach provides a non-intrusive framework in which end-users only need access to the matrices arising from the (finite element) discretization of the full-order problems in the subdomains. The traces of the finite element functions used for the discretization within the subdomains are employed to impose arbitrary Dirichlet boundary conditions at the interface, without introducing auxiliary basis functions. The methodology is seamless to the choice of the discretization schemes in space, being compatible with both LBB-compliant finite element pairs and stabilized formulations, and the DD-PGD paradigm is transparent to the employed overlapping DD approach. The local surrogate models are glued together in the online phase by solving a parametric interface system to impose continuity of the subdomain solutions at the interfaces, without introducing Lagrange multipliers to enforce the continuity in the entire overlap and without solving any additional physical problem in the reduced space. Numerical results are presented for parametric single-physics (Stokes-Stokes) and multi-physics (Stokes-Darcy) systems, showcasing the accuracy, robustness, and computational efficiency of DD-PGD, and its capability to outperform DD methods based on high-fidelity finite element solvers in terms of computing times.