arXiv subjects

Jasmin Jelovica

Publications and source records attributed to Jasmin Jelovica.

15 recordsLinked to original sources

arXiv2026

Neural topology optimization of ship structures under propulsion machinery vibrations

Ship structural vibrations contribute to noise, fatigue, and equipment damage, while dynamic-compliance topology optimization can produce pathological designs near resonance. This study extends neural-reparameterized topology optimization using a convolutional Kolmogorov-Arnold network (KATO) to forced-vibration design with active input power (AIP) as the objective. Applications include a 100 Hz engine-supporting deck panel and an 18 Hz thruster foundation frame. Helmholtz PDE filtering and Heaviside projection control feature sizes and manufacturing tolerance. Across both deck families, all eight optimized layouts reduce AIP relative to size-optimized references and, after finite-depth extrusion, also achieve lower static compliance. For unrestricted, manufacturing-aware, and stress-aware frame variants, KATO matches GCMMA in AIP within 0.5 dB while yielding 22-36x lower static compliance after matched-volume binary re-analysis. In a near-resonant 300 Hz case, both methods reduce initial AIP by more than 32 dB; KATO maintains a connected design, achieves 59x lower binary static compliance, and reduces maximum AIP over 1-500 Hz by 2.7 dB. KATO runs 6.4-10.4x faster than GCMMA for the implemented stress-aware formulations. The results demonstrate neural AIP-driven topology optimization as an efficient approach for designing connected, feature-size-controlled ship structures with improved forced-vibration performance.

cs.CE↗

arXiv2026

KATOsuper: Surrogate-accelerated neural topology optimization with sensitivity-consistent Fourier neural operators

Topology optimization (TO) remains computationally intensive due to repeated finite element analysis (FEA) evaluations required at each iteration. While neural network-based surrogates offer potential acceleration, existing approaches often suffer from gradient inconsistency between predicted objectives and sensitivities, leading to optimization instability. This work presents KATOsuper, an objective-agnostic framework that couples neural-reparameterized topology optimization with a Sensitivity-Consistent Fourier Neural Operator (SC-FNO). The framework employs the forward_split architecture, which derives deployed sensitivities via automatic differentiation through the predicted objective field and thereby preserves consistency between the predicted objective and the gradient used for optimization. The case studies include three 2D benchmark problems and three 3D structures considering compliance or stress minimization. A physics-informed multi-channel input encoding with Fourier position embedding enables resolution-invariant learning, supporting zero-shot extrapolation beyond the training resolution, with useful performance at moderate scaling factors and topology-preserving exploration at up to 64x without retraining. The framework extends to 3D through KATO3D, featuring novel KANConv3D blocks with learnable B-spline activations. KATOsuper demonstrates 15--110x deployment-time speedup over MATLAB baselines while maintaining competitive optimality, with the clearest gains observed in complex 3D and stress-optimization cases. The insight that sensitivity direction matters more than magnitude enables robust optimization even with approximate physics evaluation, extensible to other differentiable physics-driven design objectives.

cs.CE↗

arXiv2026

A new dual-population constrained multi-objective evolutionary optimization algorithm with repair constraint handling for structural optimization

Structural optimization problems often involve a large number of decision variables and highly non-convex feasible regions, making convergence to the true Pareto front extremely challenging. Even when convergence is achievable, it typically requires thousands of function evaluations, resulting in significant computational cost. This highlights the need for efficient and robust optimization algorithms for real-world engineering applications. In this study, we introduce a novel constrained multi-objective evolutionary algorithm, termed DPCME. The algorithm employs two interacting populations that exchange information, enabling effective global exploration and reducing the risk of convergence to local optima. To further enhance performance, a recent repair-based constraint-handling technique is incorporated, and alternative repair approaches are proposed and systematically evaluated. The proposed algorithm is tested on three engineering problems: the 72-bar truss, the 120-bar truss, and a chemical tanker structure, each involving hundreds of nonlinear failure constraints. Its performance is evaluated against state-of-the-art constrained multi-objective optimization algorithms from the latest PlatEMO package. A total of 43 algorithms are initially tested, from which the 12 best-performing methods are selected for detailed comparison. The results demonstrate that DPCME achieves superior or competitive convergence and diversity across all test cases, and that the inclusion of repair-based constraint handling further improves its performance.

cs.NE↗

arXiv2026

One-Step Flow Matching for Generative Modeling of Path-Dependent Physical Fields

Physical simulations for intricate geometries with path-dependent constitutive models face difficulties due to the enormous computational cost they require. Recently, the emergence of generative AI models, which succeed in image and video synthesis tasks, has provided a promise to further improve simulations. Although U-Net-based denoising diffusion probabilistic models (DDPMs) have been adopted for elastic stress field generation, they typically require hundreds of sampling steps, and applications of generative models to path-dependent, e.g. plastic, stress fields remain very limited. In this work, we propose a novel flow matching (FM) model based on a transformer backbone for high-resolution path-dependent stress field generation with stochastic loading-unloading paths and geometry. The proposed model operates within the latent space of a variational autoencoder (VAE) and formulates the simulation of plastic fields as a video synthesis task, directly generating the stress fields across all time steps. Meanwhile, we design a non-Gaussian source distribution for flow matching, such that crossings among conditional transport paths are reduced during training. This enables our model to generate satisfactory samples in one step without relying on distillation. In addition, we introduce token-level loading embeddings and two auxiliary networks to further enhance the model performance in path-dependent simulation. The results demonstrate that, even with a limited training dataset, our model can accurately generate high-resolution path-dependent fields. It is much more computationally efficient than finite element analysis, providing a speedup of 6 to 7 times over FEM on CPUs and approximately two orders of magnitude speedup on consumer-grade GPUs.

cs.LG↗

arXiv2026

Physics-Guided Dual-Stream Heterogeneous Graph Neural Network for Predicting Full-Field Structural Response of Stiffened Panels

Iterative design and optimization of large, complex structures require fast and accurate prediction of stress, displacement, and other fields. Finite element analysis (FEA) is computationally expensive for this task. Existing neural network surrogates often struggle with varying topologies and complex boundary conditions. This study proposes the novel Dual-Stream Heterogeneous Graph Neural Network (DS-HGNN) for full-field stress and displacement prediction in thin-walled structures, demonstrated on box beams made of stiffened panels. DS-HGNN operates on panel-level heterogeneous graph representations and introduces physics-guided edge states initialized from edge types, spatial information, and boundary kinematics. These states are updated through dual-stream message passing that separates longitudinal and transverse structural information while allowing cross-stream exchange. Geometry and loading effects are incorporated through Feature-wise Linear Modulation (FiLM)-conditioned 1-D spectral convolutions, and physical fields are reconstructed using a spectral-bypass low-rank readout. The model is evaluated on stiffened panel datasets with different geometries, boundary kinematics, loading conditions, and material nonlinear responses. DS-HGNN achieves the lowest stress and displacement RMSE compared with six benchmark heterogeneous graph neural network models. It also reaches comparable accuracy to the strongest benchmark models using 19%-38% fewer training samples. A targeted evaluation further shows that DS-HGNN captures yield and post-yield stress features.

cs.LG↗

arXiv2025

A hybrid global local computational framework for ship hull structural analysis using homogenized model and graph neural network

This study presents a computational framework for global local structural analysis of ship hull girders that integrates an equivalent single layer (ESL) model with a graph neural network (GNN). A coarse mesh homogenized ESL model efficiently predicts the global displacement field, from which degrees of freedom (DOFs) along stiffened panel boundaries are extracted. A global to local DOF mapping and reconstruction procedure is developed to recover detailed boundary kinematics for local analysis. The reconstructed DOFs, together with panel geometry and loading, serve as inputs to a heterogeneous graph transformer (HGT), a subtype of GNN, which rapidly and accurately predicts the detailed stress and displacement fields for any panel within the hull girder. The HGT is trained using high fidelity 3D panel finite element model with reconstructed boundary conditions, enabling it to generalize across varying panel geometries, loadings, and boundary behaviors. Once trained, the framework requires only the global ESL solution in order to generate detailed local responses, making it highly suitable for optimization. Validation on three box beam case studies demonstrates that the global prediction error is governed by the coarse mesh ESL solution, while the HGT maintains high local accuracy and clearly outperforms conventional ESL based stress estimation method.

cs.CE↗

arXiv2025

Heterogeneous Graph Representation of Stiffened Panels with Non-Uniform Boundary Conditions and Loads

Surrogate models are essential in structural analysis and optimization. We propose a heterogeneous graph representation of stiffened panels that accounts for geometrical variability, non-uniform boundary conditions, and diverse loading scenarios, using heterogeneous graph neural networks (HGNNs). The structure is partitioned into multiple structural units, such as stiffeners and the plates between them, with each unit represented by three distinct node types: geometry, boundary, and loading nodes. Edge heterogeneity is introduced by incorporating local orientations and spatial relationships of the connecting nodes. Several heterogeneous graph representations, each with varying degrees of heterogeneity, are proposed and analyzed. These representations are implemented into a heterogeneous graph transformer (HGT) to predict von Mises stress and displacement fields across stiffened panels, based on loading and degrees of freedom at their boundaries. To assess the efficacy of our approach, we conducted numerical tests on panels subjected to patch loads and box beams composed of stiffened panels under various loading conditions. The heterogeneous graph representation was compared with a homogeneous counterpart, demonstrating superior performance. Additionally, an ablation analysis was performed to evaluate the impact of graph heterogeneity on HGT performance. The results show strong predictive accuracy for both displacement and von Mises stress, effectively capturing structural behavior patterns and maximum values.

cs.CE↗

arXiv2024

A multi-objective optimization framework for reducing the impact of ship noise on marine mammals

The underwater radiated noise (URN) emanating from ships presents a significant threat to marine mammals, given their heavy reliance on hearing for essential life activities. The intensity of URN from ships is directly correlated to the speed, making speed reduction a crucial operational mitigation strategy. This paper presents a new multi-objective optimization framework to optimize the ship speed for effective URN mitigation without compromising fuel consumption. The proposed framework addresses a fixed-path voyage scheduling problem, incorporating two objective functions namely (i) noise intensity levels and (ii) fuel consumption. The optimization is performed using the state-of-the-art non-dominated sorting genetic algorithm under voyage constraints. A 2D ocean acoustic environment, comprising randomly scattered marine mammals of diverse audiogram groups and realistic conditions, including sound speed profiles and bathymetry, is simulated. To estimate the objective functions, we consider empirical relations for fuel consumption and near-field noise modeling together with a ray-tracing approach for far-field noise propagation. The optimization problem is solved using the genetic algorithm to determine the Pareto solutions and subsequently the trade-off solution. The effectiveness of the optimization framework is demonstrated via both simplified tests and practical case studies involving a large container ship. A comparative analysis illustrates the adaptability of the optimization framework across different oceanic environments, affirming its potential as a robust tool for reducing the URN from shipping.

math.OC↗

arXiv2024

Optimizing sandwich panels with graded tubular cell core for enhanced sound transmission loss

In this research a theoretical approach is proposed to evaluate sound transmission loss (STL) of a sandwich panel with tubular cells core (TCC) in the in-plane orientation. Afterwards, the core geometric parameters are optimized by implementing the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to maximize the STL through the sandwich panel while minimizing the total mass and height of the panel. The sound transmission response of the sandwich panel is parameterized in terms of four independent geometric parameters of the core, which are: number of tubular cells in the length and height directions of the core, thickness of the tubular cell wall and radii of the tubular cells. It is assumed that every tubular cell has a similar thickness while their radius in each row differs. Using the Abaqus scripting interface, a structural acoustic finite element model of the sandwich panel with tubular cell core is generated based on the specified parametric parameters. The study designs are obtained within a frequency range of 50Hz to 350Hz, which includes the first few resonance regions of the sandwich panel. In order to decrease the computational cost, a machine learning technique based on gradient boosting decision tress (GBDT) is employed to predict the STL of the sandwich panel. Findings reveal that the GBDT model is successfully trained. Data comparison shows that the discrepancies between the GBRT model and the FE model are confined to a margin of -3.09% to 2.7% in 95% of instances. The training is comprised of 500 examples, and an extra 150 samples are employed for testing to guarantee the reliability of the outcomes. The optimized panel using tubular cells demonstrates 52% and 23% reduction in the mass and height of the panel, respectively, when compared to the reference panel with the same area under the STL vs. frequency curve.

physics.app-ph↗

arXiv2023

Stress and frequency optimization of prismatic sandwich beams with joints: Performance improvements through topology optimization

Prismatic sandwich panels fabricated from metals offer a compelling alternative to more traditional panels across diverse industries, primarily due to their superior strength-to-weight ratio. Although several core types were proposed in the past, further improvements in performance could be achieved by devising the topology of the core through a topology optimization framework, which is explored in this article for the first time. Another novelty is the inclusion of joints between the sandwich beams and its surroundings in the analysis and optimization. Stress is minimized under uniform pressure loading on the beams and natural frequency maximized using the Method of Moving Asymptotes. The results are compared with X-core, Y-core, corrugated-core, and web-core sandwich beams, a few conventional prismatic sandwich types, which are optimized using a prominent global evolutionary algorithm. Manufacturing requirements are considered through practical limitations on the design variables. It is shown that structures produced by topology optimization outperform the conventional sandwich beams by up to 44% at intermediate to high mass levels, where volume fraction is between 0.2 and 0.4, but often through increased topological complexity. The new core topologies bear a certain resemblance with the conventional core types, underscoring engineering ingenuity that went into their development over the years. The topology of the optimized joints differs from the conventional joint. The results also show some limitations of the topology optimization framework, for example that it does not offer better-performing beams for volume fractions below 0.2.

math.OC↗

arXiv2023

Graph Neural Network for Stress Predictions in Stiffened Panels Under Uniform Loading

Machine learning (ML) and deep learning (DL) techniques have gained significant attention as reduced order models (ROMs) to computationally expensive structural analysis methods, such as finite element analysis (FEA). Graph neural network (GNN) is a particular type of neural network which processes data that can be represented as graphs. This allows for efficient representation of complex geometries that can change during conceptual design of a structure or a product. In this study, we propose a novel graph embedding technique for efficient representation of 3D stiffened panels by considering separate plate domains as vertices. This approach is considered using Graph Sampling and Aggregation (GraphSAGE) to predict stress distributions in stiffened panels with varying geometries. A comparison between a finite-element-vertex graph representation is conducted to demonstrate the effectiveness of the proposed approach. A comprehensive parametric study is performed to examine the effect of structural geometry on the prediction performance. Our results demonstrate the immense potential of graph neural networks with the proposed graph embedding method as robust reduced-order models for 3D structures.

cs.LG↗

arXiv2023

A parametric level set method with convolutional encoder-decoder network for shape optimization with fluid flow

In this article, we present a new data-driven shape optimization approach for implicit hydrofoil morphing via a polynomial perturbation of parametric level set representation. Without introducing any change in topology, the hydrofoil morphing is achieved by six shape design variables associated with the amplitude and shape of the perturbed displacements. The proposed approach has three to four times lower design variables than shape optimization via free-form deformation techniques and almost two orders lower design variables compared to topology optimization via traditional parametric level sets. Using the fixed Cartesian level set mesh, we also integrate deep convolutional encoder-decoder networks as a surrogate of high-fidelity Reynolds-averaged Navier-Stokes (RANS) simulations for learning the flow field around hydrofoil shapes. We show that an efficient shape representation via parametric level sets can enable online convolutional encoder-decoder application for the shape optimization of hydrofoils. The generalized flow field prediction of the convolutional encoder-decoder is demonstrated by a mean and minimum structural similarity index measure of 0.985 and 0.95, respectively, for predicted solutions compared to RANS predictions for out-of-training shapes. The convolutional encoder-decoder predictions are performed nearly five orders of magnitude faster compared to RANS. This enables a computationally tractable surrogate-based drag minimization of fifty different hydrofoils for two different design lift coefficients. Furthermore, the best local minimum obtained via the surrogate-based optimization lie in the neighbourhood of the RANS-predicted counterparts for both the design lift cases. The present findings show promise for the future shape optimization via parametric level sets with convolutional encoder-decoder over a broader spectrum of flow conditions and shapes.

physics.flu-dyn↗

arXiv2022

Assessment of convolutional recurrent autoencoder network for learning wave propagation

It is challenging to construct generalized physical models of wave propagation in nature owing to their complex physics as well as widely varying environmental parameters and dynamical scales. In this article, we present the convolutional autoencoder recurrent network (CRAN) as a data-driven model for learning wave propagation phenomena. The CRAN consists of a convolutional autoencoder for learning low-dimensional system representation and a long short-term memory recurrent neural network for the system evolution in low dimension. We show that the convolutional autoencoder significantly outperforms the dimension-reduction of complex wave propagation phenomena via projection-based methods as it can directly learn subspaces resembling wave characteristics. On the other hand, the projection-based modes are restricted to the Fourier subspace. Geometric priors of the convolutional autoencoder enabling selective scale separation of complex wave dynamics further enhance its dimension-reduction capability. We also demonstrate that geometric priors such as translation equivariance and translational invariance of the convolutional autoencoder enable generalized learning of low-dimensional maps. Thus, the composite CRAN model connecting the convolutional autoencoder with a long short-term memory network specially designed for autoregressive modeling can perform generalized wave propagation prediction over the desired time horizon. Numerical experiments display 90% mean structural similarity index measure of CRAN predictions compared to true solutions for out-of-training cases, and less than 10% pointwise $L_1$ error for most cases, verifying such generalization claims. Finally, the CRAN predictions offer similar wave characteristic patterns to the target solutions indicating not only their generalization but also their kinematical consistency.

physics.flu-dyn↗

arXiv2022

Deep convolutional neural network for shape optimization using level-set approach

This article presents a reduced-order modeling methodology via deep convolutional neural networks (CNNs) for shape optimization applications. The CNN provides a nonlinear mapping between the shapes and their associated attributes while conserving the equivariance of these attributes to the shape translations. To implicitly represent complex shapes via a CNN-applicable Cartesian structured grid, a level-set method is employed. The CNN-based reduced-order model (ROM) is constructed in a completely data-driven manner thus well suited for non-intrusive applications. We demonstrate our ROM-based shape optimization framework on a gradient-based three-dimensional shape optimization problem to minimize the induced drag of a wing in low-fidelity potential flow. We show a good agreement between ROM-based optimal aerodynamic coefficients and their counterparts obtained via a potential flow solver. The predicted behavior of the optimized shape is consistent with theoretical predictions. We also present the learning mechanism of the deep CNN model in a physically interpretable manner. The CNN-ROM-based shape optimization algorithm exhibits significant computational efficiency compared to the full-order model-based online optimization applications. The proposed algorithm promises to develop a tractable DL-ROM-driven framework for shape optimization and adaptive morphing structures.

math.OC↗

arXiv2021

Kinematically consistent recurrent neural networks for learning inverse problems in wave propagation

Although machine learning (ML) is increasingly employed recently for mechanistic problems, the black-box nature of conventional ML architectures lacks the physical knowledge to infer unforeseen input conditions. This implies both severe overfitting during a dearth of training data and inadequate physical interpretability, which motivates us to propose a new kinematically consistent, physics-based ML model. In particular, we attempt to perform physically interpretable learning of inverse problems in wave propagation without suffering overfitting restrictions. Towards this goal, we employ long short-term memory (LSTM) networks endowed with a physical, hyperparameter-driven regularizer, performing penalty-based enforcement of the characteristic geometries. Since these characteristics are the kinematical invariances of wave propagation phenomena, maintaining their structure provides kinematical consistency to the network. Even with modest training data, the kinematically consistent network can reduce the $L_1$ and $L_\infty$ error norms of the plain LSTM predictions by about 45% and 55%, respectively. It can also increase the horizon of the plain LSTM's forecasting by almost two times. To achieve this, an optimal range of the physical hyperparameter, analogous to an artificial bulk modulus, has been established through numerical experiments. The efficacy of the proposed method in alleviating overfitting, and the physical interpretability of the learning mechanism, are also discussed. Such an application of kinematically consistent LSTM networks for wave propagation learning is presented here for the first time.

cs.LG↗