Cellular automata (CA) are powerful discrete models for complex spatiotemporal systems, yet estimating hidden parameters governing their dynamics remains challenging. We propose an attention-enhanced convolutional neural network (Attention-CNN) as a surrogate optimizer for identifying the jump parameter ( σ ) in a two-phase CA model. The jump parameter regulates neighborhood radius and cell mobility, shaping emergent spatial structures. The architecture combines convolutional layers for local feature extraction with a spatial-attention module that captures global dependencies, enabling adaptive focus on informative regions of the domain. Training data were generated across varying domain sizes ( 25× 25 – 150× 150 ) and CA iterations (0, 5, 25, and 50), yielding 80,000 labeled samples. The model achieved 90.53 × faster inference than AlexNet, demonstrating robustness and efficiency across resolutions. Attention heatmaps further reveal interpretable focus patterns aligned with CA dynamics. This framework provides a fast, data-driven surrogate for inverse parameter estimation in CA-based models.
This paper introduces a Gaussian Process method that predicts the outcome of a set of chemical reactions. These reactions are traditionally described in terms of Ordinary Differential Equations and modelled using numerical methods, which can be computationally expensive and time-consuming. Since the time-scales over which reactions occur can vary drastically within a single process, the resulting numerical systems are regularly very stiff, compounding the computation expense requirement. For many applications, especially those involving a large number of predictions, speeding up the calculations would be desirable. In this work, a Sparse Variational Gaussian Process has been trained on a relatively small amount of high-dimensional data to provide both predictions and the associated uncertainty. Results show that using this Gaussian Process on a commodity personal computer (PC) can provide estimates of species concentrations and thermodynamic properties significantly faster than a traditional solver, while remaining highly accurate and providing a statistically consistent quantification of that accuracy.
This position paper explores healthcare operations and logistics from a multicriteria decision-making (MCDM) perspective, emphasizing the integration of operations research (OR) models with healthcare information technology (HIT). It reviews key OR problem classes, including timetabling, facility location, resource allocation, inventory management, supply chains, and routing, and synthesizes the performance criteria and trade-offs among efficiency, cost, patient outcomes, equity, resilience, and sustainability. We argue that classical and modern MCDM techniques are shown to reveal and balance these trade-offs, guiding model formulation and governance in complex healthcare systems. Beyond methodological aspects, we call for two complementary integrations. The first is horizontal integration: interoperability-by-design across peer healthcare IT and logistics platforms (e.g., EHRs, laboratory and imaging systems, asset and inventory management, and supplier interfaces) so that OR models can reliably read and write executable data and actions. The second is vertical integration: alignment of decisions, uncertainty treatments, and information hand-offs across planning horizons, operationalized through a four-level hierarchy of real-time (H1), operational (H2), tactical (H3), and strategic (H4). This hierarchy aligns problems, decision methods, and information hand-offs across horizons, linking rapid response and scheduling at lower levels with long-term capacity planning and governance at higher levels. The paper concludes with a research agenda focused on (i) multiobjective modeling that elevates equity and sustainability as primary design goals, (ii) interoperability-by-design frameworks connecting OR tools with healthcare IT infrastructures, and (iii) time-scale-aware optimization combining robust, stochastic, and learning-augmented methods to achieve adaptive, reliable, and data-driven healthcare logistics.
Laminar Separation Bubbles (LSBs) often occur on airfoils at low-Reynolds numbers ( 10^4-10^5 ), involving laminar separation, transition, and turbulent reattachment that challenge traditional turbulence models. This work introduces a data-driven framework combining Field Inversion and Machine Learning (FIML) to enhance the SST κ -ω -γ model in LSB regimes. Field Inversion first infers a spatial corrective field β (x) that adjusts turbulent kinetic energy production to match time-averaged LES data, with a NURBS-based regularization ensuring smoothness and dimensionality reduction. Then, a DeepONet neural operator learns the functional dependence of β on local RANS variables and fluxes, enabling adaptive model correction within the solver. Applications to the SD7003 and Eppler 387 airfoils show accurate predictions of separation, reattachment, skin friction, and pressure recovery, comparable to those obtained by LES. The proposed approach advances physics-informed, data-driven turbulence closures for regimes where high-fidelity data are limited.
pythonOCC is a library that provides Python bindings for the Open CASCADE Technology (OCCT) C++ geometric modelling kernel. To integrate it into a gradient-based shape optimization, one requires to compute the so-called geometric sensitivities, e.g., derivatives of surface nodes with respect to the design parameters. To obtain this information, pythonOCC and OCCT were algorithmically differentiated. Here, they are modularly integrated in the form of a CAD plugin into a framework for multidisciplinary design analysis and optimization (MDAO) based on the DLR’s FlowSimulator HPC ecosystem. The CAD plugin allows a robust and metadata-enabled mesh-to-CAD association between MPI domain-decomposed mesh objects and the underlying CAD patches, as well as the computation of geometric sensitivities. The framework integration of pythonOCC is demonstrated in a context of gradient-based, aerodynamic shape optimization for a RAE2822 configuration in fully turbulent, transonic flow.
An adjoint-based shape optimization method is developed for multiphase flows with phase change due to boiling; this can have numerous applications, such as two-phase immersion cooling of batteries, power electronics and similar applications. Simulation of the two-phase flow in immersion cooling is performed using a mixture model. Both the liquid and the vapor phases are assumed to be incompressible, whereas the mixture behaves as a compressible fluid; Laminar flow is assumed based on the Reynolds number present. The phase-change is modeled with the Rohsenow correlation, since nucleate boiling is expected as dominating boiling regime. For the optimization problem, a new formulation that considers the two-phase flow terms in the adjoint terms has been derived and implemented within the framework of the commercial CFD software AVL Fire. As objective function, the overall pressure drop through a flow channel where boiling occurs, is considered. For testing of the developed adjoint method, a case study of a 2D S-shaped duct is considered. The adjoint-based method for multiphase flows with phase change due to boiling is non-existent in the literature and is a novel contribution from the present work.
Fuel cell electric vehicles (FCEVs) are a promising solution for reducing global emissions. As the number of FCEVs grows, operations at hydrogen refueling stations (HRSs) must be optimized to support the hydrogen mobility ecosystem. This is challenging because station operators face multiple sources of uncertainty, such as fluctuating electricity prices and variable hydrogen demand, which are further exacerbated when renewable energy sources are integrated. The nonlinear behavior of hydrogen production technologies also makes it difficult to formulate tractable decision models. A variety of optimization approaches for HRSs operation have been proposed in the literature. These studies differ in how they formulate the problem, in the objectives and constraints they consider, and in the solution methods they employ. For example, heuristics and metaheuristics are often used for offline optimization problems, while learning-based optimization methods are increasingly applied to online problems, especially when the problem size is large. In this paper, we review component-level modeling techniques and optimization-based operational strategies proposed in the literature, and summarize how HRSs operation can be modeled and solved as an optimization problem that accounts for these uncertainties and nonlinearities. Moreover, based on existing research, we identify research gaps in this domain and outline directions for future work.
The polytopic approach offers a powerful framework for addressing the modeling, stability and control design of systems with parameter dependent dynamics, especially when these parameters vary within a known range. Applying this to the Saint-Venant equations, which are nonlinear partial differential equations (PDEs) modeling shallow water flows (such as in rivers, channels, or urban drainage), is a way to handle uncertainties and nonlinearities. Indeed, the Saint-Venant equations describe the conservation of mass and momentum in open-channel flow. They are typically nonlinear and can be quite challenging for stability and control design due to their inherent nonlinearities and the presence of varying physical parameters. These parameters are often subject to uncertainties and spatial-temporal variations, making classical linear control methods inadequate or overly conservative. The polytopic approach helps by approximating these nonlinear PDEs with a set of linear parameter-varying (LPV) models, whose parameters stay within a convex polytope.In this first contribution, we focus on the modeling of Saint-Venant equations using the polytopic framework, with a comparison study between the nonlinear, linearized and polytopic models. Theoretical arguments are provided to justify a methodology for the derivation of the polytopic model, particularly its activation functions.
Multi-objective optimization has a wide range of applications across engineering and design. In this work, we address non-orthogonal Building Spatial Design (BSD) optimization using the evolutionary algorithm NSGA-II, which has demonstrated strong performance in many optimization tasks. For representing building layouts, we employ the Prism-net representation, which is suited to capturing both thermal and structural characteristics in BSD. We initially adopted a previously proposed mutation operator that follows established guidelines for representation and operator design in metric-based evolutionary algorithms. However, we identified limitations for this operator that hindered effective exploration of the design space. Therefore, we propose an alternative delete-and-split mutation operator specifically tailored to BSD. Comparative experiments demonstrate that our mutation operator discovers better distributed and more diverse Pareto-optimal solutions. Because high-fidelity simulations in real-world building design are computationally expensive, we further introduce a low-fidelity surrogate model based on rule-of-the-thumb engineering calculations. Within a multi-model optimization framework, this low-fidelity model accelerates the early search process, enabling effective exploration of the design space and the generation of promising near–Pareto optimal solutions. These solutions can subsequently be refined using high-fidelity simulations during the detailed engineering stages.
In this work, a two-step structural optimization approach of a particle collider has been proposed. The two-step phases the optimization problem at macro and micro levels, ensuring a better overall result. The first approach proves the advantages of subdividing a structure into different parts and optimizing each part separately, combining all the optimized parts to create the final structure. The optimization flexibility is increased: it is possible to subdivide the structure according to design necessities and use different optimization methods for each part. For all the analyzed scenarios, the subdivide optimization achieves better results than the global optimization, and the best result obtained is a mass reduction 38 % higher than the global structure optimization, respecting stress and displacement constraints. The second approach involves the building of minimal surface-based lattice structures, using the hypercube as the initial domain. An iterative particle-spring dynamic model was used to generate the lattice, not the resolution of the Lagrangian equation, considering symmetry requirements related to boundary conditions. The ten obtained lattice structures have been evaluated using homogenization techniques to characterize their mechanical properties and compared in terms of material budget and stress intensification among themselves and with the Gyroid and Schwartz cells, achieving a stiffness-surface ratio 40 % higher. The two chosen lattices, 5-1 and 5-2, have been used to create a custom infill between faces, inside a volume, and on a pre-optimized support structure as a potential second optimization step.
