Engineering with Computers最新文献

This paper introduces a methodology designed to augment the inverse design optimization process in scenarios constrained by limited compute, through the strategic synergy of multi-fidelity evaluations, machine learning models, and optimization algorithms. The proposed methodology is analyzed on two distinct engineering inverse design problems: airfoil inverse design and the scalar field reconstruction problem. It leverages a machine learning model trained with low-fidelity simulation data, in each optimization cycle, thereby proficiently predicting a target variable and discerning whether a high-fidelity simulation is necessitated, which notably conserves computational resources. Additionally, the machine learning model is strategically deployed prior to optimization to compress the design space boundaries, thereby further accelerating convergence toward the optimal solution. The methodology has been employed to enhance two optimization algorithms, namely Differential Evolution and Particle Swarm Optimization. Comparative analyses illustrate performance improvements across both algorithms. Notably, this method is adaptable across any inverse design application, facilitating a synergy between a representative low-fidelity ML model, and high-fidelity simulation, and can be seamlessly applied across any variety of population-based optimization algorithms.

We propose a method for reducing the spatial discretization error of coarse computational fluid dynamics (CFD) problems by enhancing the quality of low-resolution simulations using deep learning. We feed the model with fine-grid data after projecting it to the coarse-grid discretization. We substitute the default differencing scheme for the convection term by a feed-forward neural network that interpolates velocities from cell centers to face values to produce velocities that approximate the down-sampled fine-grid data well. The deep learning framework incorporates the open-source CFD code OpenFOAM, resulting in an end-to-end differentiable model. We automatically differentiate the CFD physics using a discrete adjoint code version. We present a fast communication method between TensorFlow (Python) and OpenFOAM (c++) that accelerates the training process. We applied the model to the flow past a square cylinder problem, reducing the error from 120% to 25% in the velocity for simulations inside the training distribution compared to the traditional solver using an x8 coarser mesh. For simulations outside the training distribution, the error reduction in the velocities was about 50%. The training is affordable in terms of time and data samples since the architecture exploits the local features of the physics.

Finite element (FE) analysis is one of the most accurate methods for predicting electromagnetic field scatter; however, it presents a significant computational overhead. In this study, we propose a data-driven parametric model-order reduction (pMOR) framework to predict the scattered electromagnetic field of FE analysis. The surface impedance of a coated component is selected as parameter of analysis. A physics-aware (PA) neural network incorporated within a least-squares hierarchical-variational autoencoder (LSH-VAE) is selected for the data-driven pMOR method. The proposed PA-LSH-VAE framework directly accesses the scattered electromagnetic field represented by a large number of degrees of freedom (DOFs). Furthermore, it captures the behavior along with the variation of the complex-valued multi-parameters. A parallel computing approach is used to generate the training data efficiently. The PA-LSH-VAE framework is designed to handle over 2 million DOFs, providing satisfactory accuracy and exhibiting a second-order speed-up factor.

The Volume-of-Fluid (VOF) method is commonly used for numerical simulations of phase change phenomena, such as nucleate boiling or droplet evaporation. A key issue with the standard VOF method is the averaging of the liquid and vapor properties in interface cells, which causes non-physical conjugate heat transfer with a solid wall. Therefore, we aim at a physical model for conjugate heat transfer between a solid and a multiphase fluid. The first measure for higher quality simulations is the splitting of the single temperature field in the fluid region into separate liquid and vapor temperature fields. The second measure is the development of a new, more physical temperature boundary condition for conjugate heat transfer between a solid region and a multiphase fluid, based on experimental results, theoretical models and theoretical considerations. In interface cells, the vapor phase is excluded from the conjugate heat transfer because only heat transfer to the liquid phase occurs resp. dominates. Additionally, the conjugate heat transfer between solid and liquid in the interface cells is performed with virtual subcells, depending on the respective volume fraction of the liquid phase. This new approach (we name it distinctive approach) is successfully validated for energy conservation, and stability issues are discussed for the first time. Significant differences to simulations with averaged properties are observed due to the (now) physically correct modeling of conjugate heat transfer. In our boiling cases, the more accurate numerical simulations lead to considerably larger bubble growth rates. Higher quality simulations are also expected for nearly all applications, where there is a three-phase contact line, be it vapor bubbles in nucleate boiling or droplets impacting on a heated surface.

The purpose of this paper is to point out that it is possible to evaluate the approximation solution of elliptic Partial differential equations (PDEs) on regular and irregular domains where no boundary conditions are defined on some part of the boundary domain. In the presence of trigonometric basis functions (TBFs), the backward substitution method (BSM) coupled with the radial basis functions neural network (RBFNN) is implemented very easily and works well. As a result, the approximation of the boundary conditions and the approximation of the PDE inside the solution domain is separated. The particular solution with an ungiven part of the inhomogeneous boundary condition is completely analyzed by the RBFNN method, and the efficiency and accuracy of the developed algorithms are discussed.

Many thermomechanical processes, such as rolling, turning, grinding, welding or additive manufacturing, involve either a material flowing through a fixed load system or a heat source moving with respect to the material. In many situations, these processes involve a constant speed translational, rotational or helical movement of the loading with respect to the material so that a (quasi-) steady thermo-mechanical state is achieved quickly. Classical Lagrangian steady state finite element simulation of these processes in the material’s frame is a heavy task requiring large meshes refined all along the load path. This article presents a nodal-integration-based finite element method for solving transient and steady-state elastoplastic problems associated with these processes. The simulation is carried out step by step in a frame linked to the loading. As the nodes of the mesh do not represent material points, the computation procedure requires determining the position at the previous time step of the material point associated with each node (anterior point) in order to perform the time-integration of the constitutive equations. The anterior points are located anywhere in the mesh and therefore interpolation techniques are required to get the previous mechanical state there. As all the mechanical variables are calculated at nodes with the method proposed, this approach makes the interpolation more straightforward. Applications to 3D forming and welding are presented to illustrate the efficiency of the proposed method. The results of finite element simulations in the frame tied to the loading are compared to those of Lagrangian calculations simulating the load motion in the material’s frame. The applications demonstrate that the proposed model can significantly accelerate simulations, achieving a maximum acceleration of around 40 in 3D forming and about 4 in welding. These results highlight the substantial efficiency improvements enabled by the proposed method.

Assessing the structural integrity of ageing structures that are affected by climate-induced stressors, challenges traditional engineering methods. The reason is that structural degradation often initiates and advances without any notable warning until visible severe damage or catastrophic failures occur. An example of this, is the conventional inspection methods for prestressed concrete bridges which fail to interpret large permanent deflections because the causes—typically tendon loss—are barely visible or measurable. In many occasions, traditional inspections fail to discern these latent defects and damage, leading to the need for expensive continuous structural health monitoring towards informed assessments to enable appropriate structural interventions. This is a capability gap that has led to fatalities and extensive losses because the operators have very little time to react. This study addresses this gap by proposing a novel machine learning approach to inform a rapid non-destructive assessment of bridge damage states based on measurable structural deflections. First, a comprehensive training dataset is assembled by simulating various plausible bridge damage scenarios associated with different degrees and patterns of tendon losses, the integrity of which is vital for the health of bridge decks. Second, a novel General Regression Neural Network (GRNN)-based cascade ensemble model, tailored for predicting three interdependent output attributes using limited datasets, is developed. The proposed cascade model is optimised by utilising the differential evolution method. Modelling and validation were conducted for a real long-span bridge. The results confirm the efficacy of the proposed model in accurately identifying bridge damage states when compared to existing methods. The model developed demonstrates exceptional prediction accuracy and reliability, underscoring its practical value in non-destructive bridge damage assessment, which can facilitate effective restoration planning.

Laser-directed energy deposition (L-DED) enables the creation of near-net-shape parts with location-specific materials, repair of machine components, and addition of features to existing parts. However, gathering sufficient experimental L-DED data to establish process maps is challenging especially when expensive materials are being investigated. Despite the interest in data-driven modeling for developing such maps, few studies have considered reusing knowledge across multiple materials including uncertainty quantification (UQ). To address this, knowledge transfer methods based on Gaussian process (GP) are proposed. Melt pool data for SS316L and IN718 are used to emulate data-rich and data-scarce conditions, respectively. Three avenues are explored: (i) mixing the data of both materials to train a single GP regression model (the mixed-input model), (ii) relation-based transfer learning (RB-TL) model, and (iii) multi-fidelity GP-based transfer learning (MFGP-TL) model. Results show that the mixed-input model outperforms the baseline or no-transfer model under data-deficient conditions. Compared to the baseline model, the RB-TL model exhibits a general improvement in accuracy and confidence while consuming the least computation time among all proposed models. The MFGP-TL model achieves the best performance, which is only half the error and standard deviation observed for the RB-TL model, albeit resulting in longer computation times. Finally, the proposed transfer learning models, when used on experimental data obtained from the literature, show 22–31% and 24–40% improvement over the baseline model for IN718 and IN625, respectively. This work, therefore, facilitates data- and cost-effective UQ-based knowledge transfer in reconstructing process maps in L-DED.

The stabilized conforming nodal integration (SCNI) is currently widely employed in Galerkin meshfree formulation. A key ingredient of SCNI is the strain or gradient smoothing defined within a set of conforming nodal representative domains, which usually are formed by the auxiliary points in addition to the meshfree nodes. Nonetheless, these auxiliary points may significantly increase the storage requirement and computational cost of SCNI, in comparison with the direct nodal integration. In order to address this issue, a purely node-based consistent non-conforming gradient smoothing (CNGS) scheme is proposed herein to accelerate the Galerkin meshfree computation. In the proposed method, only the meshfree nodes are employed to construct overlapping and non-conforming nodal representative domains, which are then adopted for the nodal gradient smoothing operation. However, unlike the existing non-conforming gradient smoothing algorithms that commonly violate the integration consistency, the proposed method maintains the desirable integration consistency through a proportional separation between the nodal gradient smoothing domains and the nodal integration domains, which essentially ensures the meshfree solution accuracy. Meanwhile, due to the absence of auxiliary points in the gradient smoothing evaluation, the computational efficiency is substantially improved by the proposed method of CNGS compared with SCNI. The effectiveness of the proposed methodology is well demonstrated by numerical results.