Physics-informed neural network simulation of thermal cavity flow

3D concrete printing (3DCP) is an innovative construction method that extrudes cementitious materials layer-by-layer to fabricate building components based on a digital model. 3DCP has gained increasing adoption globally for projects like buildings, bridges, retaining walls, and stormwater management systems. However, 3DCP places stringent demands on the rheological properties of the printable cementitious materials. Rheology is critical for the flow and buildability of fresh concrete during extrusion. The mixture must have a low yield stress for pumping, moderate viscosity to hold its shape after deposition, minimal bleeding and segregation of aggregates, as well as responsive rheology that can transition from fluid to solid state as layers are printed sequentially. In this dissertation, we aim to solve the challenges mentioned above including: (1) accurate and efficient quantification of rheological and thixotropic properties of cementitious materials, (2) evaluation of flow behavior of fresh concrete during extrusion. Using traditional experimental methods to evaluate rheological properties of cement paste is labor-intensive and time-consuming. Numerical simulations like Finite Element Method (FEM) or Computational Fluid Dynamics (CFD) can help increase the speed of finding the most suitable material for 3DCP. However, these simulation methods are costly in computing power, making them less desirable for simulating the dynamic and complex process of 3DCP. Moreover, mesh-based approaches are not suitable for complex geometries as the mesh generation would be challenging with curved boundaries, small features, holes or thin regions. Physics-Informed Neural Network (PINN) is introduced in the work as an innovative solution to simulate and model the whole process of 3DCP. It can leverage the flexibility and computational efficiency of data-based neural networks, which can accelerate the learning and convergence speed. At the same time, Partial Differential Equations (PDEs) would be embedded in the network structure to increase the accuracy of its prediction. The integrated governing physical laws can help the neural network understanding the underlying physics of the 3DCP process, potentially increasing the accuracy and reliability of simulations. A significant part of the research is dedicated to understanding the rheological and thixotropic behavior of cementitious materials, which is critical for the material design in 3DCP. We developed a Rheology-informed Neural Network, RheologyNet, where we embedded the rheological constitutive laws into the loss function. These physical laws can help the network focus on simulating the rheological properties of cementitious materials. Moreover, a thixotropic evaluation system has been established based on the proposed model to evaluate the thixotropic property of cement paste. This study is presented in Chapter 2. Then, we proposed a Navier-Stokes Informed Neural Network (NSINN) to study the flow behavior of cement paste in the 3DCP barrel and nozzle since it involves complicated Multiscale or Multiphysics behaviors, such as anisotropy and non-uniform shear rate distribution. The Navier Stokes Equation and the rheological constitutive equations are coupled and embedded into the NSINN architecture to simulate the flow behavior in the barrel during 3D-printing extrusion process. This approach demonstrates the potential of PINNs to provide high-accuracy simulations of velocity and pressure fields in a computationally efficient manner compared to traditional mesh-based models. Also, we used the NSINN to learn the relationship between the nozzle size and the printing quality. The details of the study are presented in Chapter 3. At last, a multi-subnetwork PINN architecture (MultiSubPINN) is proposed to address the limitations of traditional PINNs. This is because traditional PINNs used Fully-connected Neural Network (FNN) as main bone. It has problems like different outputs must share parameters within the network structure. However, this parameter sharing nature might lead to potential errors and decrease the model performance. The main architecture of this network is consisted of multiple separated sub-networks and the loss function of each sub-network is calculated separately to update the parameters in each sub-network. By doing this, we can ensure that the optimization for one output doesn't negatively impact others, therefore providing a more flexible, scalable, and potentially more accurate alternative for the evaluation. Besides, using randomly distributed training points in the computational domain for PINNs is a common way. However, many physical problems exhibit non-uniform behavior, with certain regions of the domain having more complex dynamics or sharper gradients than others. Randomly distributed training points might not adequately capture these regions, leading to a model that misrepresents the underlying physics in critical areas. We proposed a Physical-based sampling strategy to optimize the training of the PINNs. This study is presented in Chapter 4. In summary, this dissertation has established a new platform that is enabled by interpretable NN to accurately quantify cement paste’s rheological properties and simulate the flow behavior of 3D concrete printing process. The findings suggest that PINNs can not only improve the efficiency and accuracy of simulations but also provide a more profound understanding of the material behaviors and process dynamics critical to 3DCP. The novel PINN structure developed from this dissertation is a powerful general platform that could be applicable to a variety of material domains for complex rheological behavior predictions.

In this paper, we propose an extrinsic approach based on physics-informed neural networks (PINNs) for solving the partial differential equations (PDEs) on surfaces embedded in high dimensional space. PINNs are one of the deep learning-based techniques. Based on the training data and physical models, PINNs introduce the standard feedforward neural networks (NNs) to approximate the solutions to the PDE systems. Using automatic differentiation, the PDEs information could be encoded into NNs and a loss function. To deal with the surface differential operators in the loss function, we combine the extrinsic approach with PINNs and then express that loss function in extrinsic form. Subsequently, the loss function could be minimized extrinsically with respect to the NN parameters. Numerical results demonstrate that the extrinsic approach based on PINNs for surface problems has good accuracy and higher efficiency compared with the embedding approach based on PINNs. In addition, the strong nonlinear mapping ability of NNs makes this approach robust in solving time-dependent nonlinear problems on more complex surfaces.

Influence of two- and three-dimensional simulations on bubble behavior in gas–solid fluidized beds with and without immersed horizontal tubes

We used ultrasonic shear waves for nondestructive defect detection in a billet using transmitted waves. We utilized the deviation the time-of-flight (TOF) obtained by cross-correlation of transmitted waves of a defect-free reference plane and that of a measurement plane containing a defect. We compared the performance of longitudinal waves and shear waves at different wavelengths in detecting the diameter of a circular defect in two-dimensional (2D) simulation and the TOF for a cylindrical defect while changing the vibration direction of shear waves in three-dimensional (3D) simulation. Shear waves detected defects better than longitudinal waves in the 2D simulation, especially at wavelengths of 1.4–2.4 mm. In the 3D simulation, the maximum TOF was larger when the vibration direction was perpendicular to the defect’s major axis than when it was parallel in the measurement using shear waves. This suggests a defect’s shape can be estimated by measurement using shear waves.

We present two methods to estimate bottom topography in a shallow water flow using only surface deformation measurements. One is based on Physics‐Informed Neural Networks (PINNs) and the other on the Adjoint State Method. We test both methods using synthetic data in 1D and 2D cases. Both are able to successfully reconstruct not only the bottom topography but also the surface velocity. Both also show robustness against noise and data sparsity up to reasonable levels.

In spite of the importance of a detailed description of the filamentary current constriction of the IGBT during the turn-off operation that could lead to the device’s failure, there are to date no quantitative 3D simulation results of the filament growth and dynamic that can be compared with experimental results. In this paper we present 3D numerical simulations on the failure mode in the Unclamped Inductance Switching (UIS) test operation, extended to the full device area, which will be usefully compared with detailed experimental results on a large number of trench IGBT test samples. For the first time extended 3D dynamic electrothermal simulations of the whole die are made, to take into account both the electric and thermal effects of the filamentary conduction in avalanche mode. The onset of a filament growth condition for a current level just above the turnover voltage evaluated by the 3D simulations, and the area of the filament, obtained for the first time, are well in agreement with the quantitative data extracted by the experimental evaluations. Moreover, the thermal heating due to the filament formation is found to be quite independent from the current level, because it depends on the current density in the filament, rather than on the injected current. The delay time between the filament formation and the final failure time seen in the experimental results is verified to be due to the movement of the filament all around the chip surface in search of a cooler spot. The movement of the filament along the whole die area is verified for the first time by full area 3D electrothermal dynamic simulations, with times in agreement with the experimental delay between filament formation and final failure seen in all the failure reports.

There have been remarkable advances in semiconductor industry but the miniaturization of magnetic devices is progressed very slowly. So many studies are carried out overcoming the gap. To develop small size devices, proper design and property simulation of the devices are requested to spare the efforts before fabrication processes. However, there is a few of simulation studies have been reported. In this work, we fabricated thin film toroidal inductors using permalloy as a core material and measured the properties to estimate 2D and 3D simulation in the magnetic devices. The 3D simulation shows more agreeable results than the 2D simulation. On the other hand, when we changed core materials, the inductor using Fe-Hf-C-N nanocrystalline alloys as core material shows the better properties than using permalloy and Fe-Zr-C-N alloys.

Two- and three-dimensional Direct Numerical Simulation of particle-laden gravity currents

In this study, E-Scaffolding Enhance Learning (ESEL) using Web-based simulation has been designed and implemented. The scaffolding approach in this research is a conceptual scaffolding of the moment of inertia. Web-based simulation consists of 2D and 3D simulation. The 3D simulation was developed using unity software and 2D simulation was developed using scratch and articulate storyline software. All simulation was published as an Html file, and finally inserted into the website. The validation of the product was carried out by experts such as a physics lecturer and a physics teacher. The implementation phase was carried out through collaborative distance learning between one Senior High School in Bandung and one school in West Nusa Tenggara. The validation results show that ESEL using Web-based simulation on moment of inertia is deserved with an average score of 92% in the very good category. Nevertheless, there are several improvements still need to be made in terms of the appearance of the Web-based 3D simulation especially when it is accessed by phone cell. The appearance of the Web-based 3D simulation is still not responsive for smartphones. But based on the questionnaire result of students, ESEL using Web-based simulation on moment of inertia has a value 71% which means that ESEL has a good category of validation and is interesting for the students. ESEL using Web-based simulation also improves students’ activities in the distance learning of physics classroom.

Highlights Bin filler design and bin dimensions determine apple distribution Performance of new apple distribution systems was estimated with 3D simulations Simulated lemniscate scattering path demonstrated improved apple distribution in rectangular bins 3D simulation provides an efficient, informed design decision tool before going into production Abstract. Mechanical harvest platform systems equipped with conveyors were meant to replace the traditional apple-picking ladder and bucket; however, their feasibility has not yet been exhausted. The conveyor belts transport the apples to the bin filler, whose role is to distribute the fruit into the bins. Commercial mechanical harvest platforms are designed to perform well with square bins; however, Israeli apple industry bins are rectangular. This results in a non-uniform apple distribution, which requires human intervention and contributes to apple bruising and economic loss. In the present study, a 3D simulation was used to quantitatively evaluate the performance of three novel bin filler designs (back-and-forth, double disk, and lemniscate) and compare them to commercial bin fillers. The bin filler models were tested for the uniformity of the apples distributed into square and rectangular bins. The angle of repose and the standard deviation of the height of the top layer of apples were calculated. The commercial bin filler model performed better when filling square bins (angle = 12°, SD height = 31 mm) than rectangular bins (angle = 18°, SD height = 71 mm). The lemniscate model (angle = 10°, SD height = 41 mm) outperformed the two others as well as the commercial bin filler model (collapsed: angle = 18°, height = 67 mm). A machine design for the lemniscate model is proposed. The quantitative methods described in this study point to the potential benefit of 3D simulation as an inexpensive means to explore the feasibility of new harvest devices.

Large progress has been realized in modeling conduction heat transfer problems over the past decade by the introduction of high performance compact thermal models (CTMs) mainly developed for thermal design of complicated electronic systems. The objective of this paper is to generalize these advances to convective heat transfer. A new convective CTM is proposed, which offers many advantages over the traditional approach using the heat transfer coefficient (HTC). The latter is simply a zeroth order CTM. The HTC is quite handy and simple, but with unpredictable errors. It can be suitable for hand calculations of simple systems giving rather crude estimates. For a higher precision, users have no other option than time consuming 3D simulations. For large systems, in terms of number of components, 3D simulations can be rapidly impractical. The CTM bridges the gap between both approaches going gradually from “HTC” levels (low precision and calculations time) at the zeroth order, to 3D simulation precision and computing time levels at large orders. Fortunately, like for conduction, a CTM of order of few tens quickly approaches 3D simulation precision levels, while keeping computation time significantly lower than 3D simulation. Moreover, the CTM approach solves conjugate heat transfer problems in a quite elegant way. A “black box” model, developed for fluid domain alone, can be easily combined with classical CTM conduction models to generate good precision predictions for any combination of fluid/solid domains.

A parametric, three-dimensional, computational study examining steady state performance of a planar plug micronozzle is conducted. Nozzle depth and throat Reynolds number are varied for a single plug contour designed using an approximation for the Method of Characteristics. The Reynolds number is varied from 80 − 820, for nozzle depths of 90μm-360μm. Thrust and specific impulse are used as metrics to assess nozzle performance. Results from 3D simulation are compared to 2D simulation for the same planar geometry and 3D results for a 30◦ linear-walled nozzle of the same throat size. For all Reynolds numbers examined, 3D nozzles were found to under-perform the 2D case. The margin between the 3D and 2D cases decreases with increasing nozzle depth. The 3D plug nozzle outperforms the 3D 30◦ linear-walled nozzle in both 2D and 3D simulation for the range of Reynolds numbers examined.

A comparative study between two- and three-dimensional (2D and 3D) modeling is carried out on arc discharge phenomena inside a thermal plasma torch with hollow electrodes, in order to evaluate the effects of arc root configuration characterized by either 2D annular or 3D highly localized attachment on the electrode surface. For this purpose, a more precise 3D transient model has been developed by taking account of 3D arc current distribution and arc root rotation. The 3D simulation results apparently reveal that the 3D arc root attachment brings about the inherent 3D and turbulence nature of plasma fields inside the torch. It is also found that the constricted arc column near the vortex chamber plays an important role in heating and acceleration of injected arc gases by concentrating arc currents on the axis of the hollow electrodes. The inherent 3D nature of arc discharge is well preserved inside the cathode region, while these 3D features slowly diminish behind the vortex chamber where the turbulent flow begins to be developed in the anode region. Based on the present simulation results, it is noted that the mixing effects of the strong turbulent flow on the heat and mass transfer are mainly responsible for the gradual relaxation of the 3D structures of plasma fields into the 2D axisymmetric ones that eventually appear in the anode region near the torch exit. From a detailed comparison of the 3D results with the 2D ones, the arc root configuration seems to have a significant effect on the heat transfer to the electrode surfaces interacting with the turbulent plasma flow. That is, in the 2D simulation based on an axisymmetric stationary model, the turbulence phenomena are fairly underestimated and the amount of heat transferred to the cold anode wall is calculated to be smaller than that obtained in the 3D simulation. For the validation of the numerical simulations, calculated plasma temperatures and axial velocities are compared with experimentally measured ones, and the 3D simulation turns out to be more accurate than the 2D simulation as a result of a relatively precise description of the turbulent phenomena inside the torch using a more realistic model of arc root attachment. Finally, it is suggested that the 3D transient formulation is indeed required for describing the real arc discharge phenomena inside the torch, while the 2D stationary approach is sometimes useful for getting practical information about the time-averaged plasma characteristics outside the torch because of its simplicity and rapidness in computation.

The purpose of this paper is to shed light on how and when 3D simulations are preferable over 2D simulations in order to predict the replicated height of nanocavities via plastic injection moulding. Simulations are necessary since not all the cavity is copied depending on geometrical and injection conditions. A 3D simulation is preferable over a 2D simulation in two cases: (i) when the heat transfer through the walls parallel to the simulated domain is not negligible and (ii) when the polymer final shape is relevant. This work explains when 2D simulations like the more commonly found in the literature are not enough to obtain reliable results and how to obtain such results by means of a 3D simulation.

This paper introduces a novel approach to solve inverse problems by leveraging deep learning techniques. The objective is to infer unknown parameters that govern a physical system based on observed data. We focus on scenarios where the underlying forward model demonstrates pronounced nonlinear behaviour, and where the dimensionality of the unknown parameter space is substantially smaller than that of the observations. Our proposed method builds upon physics-informed neural networks (PINNs) trained with a hybrid loss function that combines observed data with simulated data generated by a known (approximate) physical model. Experimental results on an orbit restitution problem demonstrate that our approach surpasses the performance of standard PINNs, providing improved accuracy and robustness.