Hongjun Ran

https://lnkd.in/gBwt_e2r Physicists have been hoping for this moment for a long time: For many years, scientists all around the world have been searching for a very specific state of thorium atomic nuclei that promises revolutionary technological applications. It could be used, for example, to build a nuclear clock that could measure time more precisely than the best atomic clocks available today. It could also be used to answer completely new fundamental questions in physics—for example, the question of whether the constants of nature are actually constant or whether they change in space and time.

47

Starting the 4th of July weekend with a bit of a celebration: new paper on ML-accelerated mod/sim out!💥 Materials simulations based on direct numerical solvers is the primary way we conduct computational materials research today. These solvers are accurate but computationally expensive. In this paper, we are rethinking materials simulation by blending direct numerical solvers with neural operators. This approach is based on a U-Net neural network conditioned on time. The idea is to learn the multiple dynamics occurring during the simulation via this machine-learned approach to speed up the simulation time and feed the predictions back to the direct numerical solvers to keep the error bonded and course-correct the physics from time to time. These are some of the first steps towards the future of mod/sim. Great collaboration with George Karniadakis group, notably Vivek Oommen leading the charge. More details in our open access paper here:

101

I'm thrilled to announce the release of FAST: Future Aircraft Sizing Tool, the first open-source software out of my group, IDEAS Lab at the University of Michigan! So, what sets FAST apart from other tools? Many years ago, I created one of the first conventional/electrified aircraft sizing tools as part of my PhD thesis, E-PASS, which has been used extensively in industry and government projects. While many academic tools followed, most remain black boxes. FAST is open-source, democratizing access to knowledge. Unlike some of the existing tools that require extensive design knowledge, FAST leverages a historical database and predictive modeling, making aircraft design accessible with just four inputs. And if you're an expert user, FAST allows you to customize your design with more control and granularity suitable for a conceptual design tool. FAST excels at early-stage design space exploration, rapidly evaluating countless input combinations. When developing novel concepts, it's crucial to explore all possibilities and understand the consequences at the system level. The early design stage often lacks detailed information, making big decisions premature. FAST mitigates this by enabling rapid analysis and exploration, empowering users to navigate uncertainty effectively. We hope that FAST will be a valuable resource for educators and researchers alike. While we’re continuing development on new capabilities, you’re welcome to develop your own and contribute to the open-source movement and join us in advancing sustainable aviation!

156

5 Comments

Here’s some more cool info on physics-informed neural networks, or, 2020’s version of reduced order modeling. Most applications are in the CFD space, but now more research is looking at hypersonics, middle defense, and PNT in space systems. I’m stoked to see how far this type of work goes - I’ll always be a physicist at heart. https://lnkd.in/gEsKRaU4

1

Can iterative methods for linear systems be efficient for gate-based quantum computers? The no-cloning theorem has often been pointed out as a barrier, since copies of iterates are required for conjugate gradient or Krylov subspace methods. But what if the iterative scheme is linear? Our new paper tackling this issue, qRLS, has been published today. We show that if the iterative scheme is linear, a quantum state encoding the iterate can indeed be prepared with exponentially fewer resources compared to classical computing. This implies that relaxation methods like Jacobi, Gauss-Seidel, etc. can be exponentially efficient on quantum computers. https://rdcu.be/dFGDq This paper aims to provide a foundation to build more advanced and refined iterative methods for quantum computers and provides explicit circuit descriptions for example problems. The question remains to decrease the number of iterations. For this we have proposed a quantum multigrid approach. This idea is built on the same footing, that all operations in the multigrid algorithm are linear. Stay tuned for more developments! https://lnkd.in/d22pVA4E #quantum #quantumcomputing #engineering #finiteelement #differentialequations #pdes

40

Electron Microscopy Data: The study produced 1.4 petabytes of electron microscopy data from a cubic millimetre of the human temporal cortex. Neural Reconstruction: Thousands of neurons, over a hundred million synapses, and other brain matter elements like glial cells and blood vessels were reconstructed. New Discoveries: A previously unrecognized class of neurons and rare, powerful multisynaptic connections between neurons were identified. Open Resource: All data and analysis tools have been made freely available online for further research into human brain function.

3

1 Comment

You're invited to an insightful seminar in Industrial & Systems Engineering featuring Dr. Jia (Peter) Liu from Auburn University. Dr. Jia (Peter) Liu will speak about "Understand fatigue performance of metal additively manufactured parts with data-driven modeling". Dr. Jia Liu is an Assistant Professor in the Department of Industrial and Systems Engineering at Auburn University. He graduated from Virginia Tech with a Ph.D. in Industrial and Systems Engineering and an M.S. in Statistics in 2017. He also received a B.S. and an M.S. in Electrical Engineering from Zhejiang University, China, in 2005 and 2007, respectively. He focuses on interpretable data-driven modeling for complex manufacturing processes with heterogeneous sensor data to achieve online process monitoring, product quality prediction, and control. He has published over 40 journal and conference papers. His research contribution has been featured in IISE magazine. He has been honored with several awards, including the 2023 National Science Foundation (NSF) Career Award, the 2023 Ginn Achievement Faculty Fellow at Auburn University, and the 2020 Early Career Development Grant at Auburn University. Abstract: Additive manufacturing (AM) offers unprecedented flexibility in freeform fabrication using a layer-wise strategy for product innovation, mass customization and sustainable production. With a laser scanning metal powder layer by layer, laser beam powder bed fusion (L-PBF) provides high geometry accuracy and strength for different applications. Though L-PBF fabricated parts have comparable static mechanical properties with their counterparts from wrought materials, their dynamic properties under cyclic loading are inferior due to their nonhomogeneous microstructure from inconstant process characteristics. Their fatigue performance has more uncertainty and larger variability than their wrought counterparts do. Since fatigue failure has historically caused the majority of engineering failure incidents, the uncertainty in the fatigue performance of L-PBF parts thwarts the adoption of such innovative technology in fatigue-critical applications. In this talk, I will introduce our work in data-driven modeling to advance the understanding of microstructural traits contributing to fatigue fracture of L-PBF parts and entail non-destructive fatigue life prediction. 🗓️ Date: Tuesday, April 23, 2024 🕥 Time: 12:10PM-1:30PM 📍 Location: CoRE-101 #Rutgers #RutgersSoE #IndustrialandSystemsEngineering #Seminar #AdditiveManufacturing

5

The relation between dislocation (line defects in materials) glide and shear stress (driving force) is known as the mobility relation. In this new study we investigate the motion of sonic, transonic, and supersonic twinning dislocations with a view to improving understanding of twinning during high strain rate deformation. The twinning-dislocation mobility relation is composed of two branches separated by a band of forbidden velocities. Our simulation results also show that twinning-dislocation motion is kink-mediated. Great collaboration lead by Michael Demkowicz and his student Ta Duong at TAMU. Link to those findings can be found here:

40

📢 📢 📢 📢 New Publication Alert 📢 📢 📢 📢 Our, then PhD student, now officially a colleague, Lequn Chen, did an exquisite job in capturing the current state of affairs in all-things-process-monitoring related to Laser-based Additive Manufacturing. Free access to the publication in the comments below (valid for 50 days). Click to have a look on how Sensors and Machine Learning can be deployed to enhance quality in 3D printing. #AdditiveManufacturing #MachineLearning #ProcessMonitoring #QualityAssurance #AI

28

1 Comment

Enhancing Anomaly Detection with Adaptive Noise: A Pseudo Anomaly Approach Researchers proposed a method that involves creating pseudo anomalies by incorporating adaptive noise into normal data. This is achieved by training another network to generate noise based on normal input, which is then added to the input to create pseudo anomalies. By training an AE to poorly reconstruct these pseudo anomalies, the reconstruction boundary evolves, improving anomaly detection. Unlike methods with strong inductive bias, this approach is generic and applicable across diverse domains, as demonstrated across various datasets. Their method aims to enhance anomaly detection by training an AE to reconstruct normal input well while poorly reconstructing anomalies, achieved by incorporating learned adaptive noise into the normal data. This noise, generated by another autoencoder, creates pseudo anomalies, which are reconstructed poorly by the AE. The training process ensures that the AE evolves its reconstruction boundary towards improved anomaly detection. Unlike methods with strong inductive bias, this approach is generic and applicable across various domains. During testing, anomaly scores are computed based on reconstruction error, offering efficient inference without additional computational cost.

4

Magnetic fields can create mass! Mathematically, the permeability of free space times the square of fundamental electric charge of electron all divided by the classic Compton radius of the electron times 4 pi will yield the mass of the electron. The permeability of free space is henry per meter so there is the magnetic connection. Note that charge squared is also involved. See electrogravity.com

13

6 Comments

THERMODYNAMICS AS A SCIENCE OF SYMMETRY “The most immediately evident conserved coordinate is, of course, the energy (time-translation symmetry). Its relevance as a thermodynamic coordinate underlies the "first law" of thermodynamics. Time-translation, spatial translation, and spatial rotation symmetries are interrelated in a single class of continuous space-time symmetries. The symmetry interpretation of thermodynamic immediately suggests, then, that energy, linear momentum, and angular momentum should play fully analogous roles in thermodynamics. The equivalence of these roles is rarely evident in conventional treatments, which appear to grant the energy a misleadingly unique status. The momentum and the angular momentum are generally suppressed by restricting the theory to systems at rest, constrained by external "clamps." Nevertheless, it is evident that in principle the linear momentum does appear in the formalism in a form fully equivalent to the energy, for relativistic considerations imply that the energy in one frame appears partially as linear momentum in another frame. Similarly, the angular momentum is only occasionally introduced explicitly into thermodynamic formalisms (as in astrophysical applications to rotating galaxies); it appears, for instance, in the "Boltzmann factor", additively and symmetrically with the energy. To stress these facts we might well amend the first law to read that "the extended first law of thermodynamics is the symmetry of the laws of physics under space and time translations and under spatial rotation.".” - Herbert B. Callen in Thermodynamics as a Science of Symmetry, Foundations of Physics, Vol. 4, No. 4, 1974

14

1 Comment

AI Starts to Sift Through String Theory’s Near-Endless Possibilities Using machine learning, string theorists are finally showing how microscopic configurations of extra dimensions translate into sets of elementary particles — though not yet those of our universe.

2

My research paper, titled "Confinement Phenomena in Topological Stars," is now available on the arXiv. The study explores the implications of Quantum Chromodynamics (QCD) in extreme density regimes. Under certain conditions, such as those present in black holes, quarks may undergo a transition to a deconfined quark-gluon plasma phase. Moreover, there exist regions around black holes characterized by pronounced gluon-gluon and quark-gluon interactions. In environments with high QCD density, the energy-momentum tensor might exhibit anomalous transport phenomena, such as the chiral magnetic effect, which could be reflected in the unique electromagnetic signatures of topological stars. The investigation of QCD contributions to the energy-momentum tensor in extreme density environments provides profound insights into the formation, behavior, and characteristics of topological stars. It introduces paradigm-shifting concepts that redefine current understanding and open new avenues for exploration, thereby altering the trajectory of astrophysical research and enhancing our comprehension of the universe's deepest mysteries. This research was conducted during my undergraduate sophomore year in 2022-2023. Note:- This is just a preprint, I have submitted the original document to a journal. It is under a rigorous review process. Link: https://lnkd.in/gQu6mQZa

154

8 Comments

[50 days' free access!] [A paper with Yuan L. and Arjit Seth] Expanding the aerodynamic shape optimization (ASO) formulation to a mission-based one has enabled us to investigate the tradeoff between fuel consumption and trailing-edge noise (TEN), while taking into account the comprehensive flight operational spectrum of the aircraft. TEN---despite not being the most important noise source for aircraft---is a direct consequence of airfoil shapes, which should be considered in ASO. Our work demonstrates the interplay between aerodynamics, aeroacoustics, and flight operational variations, which was made possible by the mission-based platform. The results provide a strong basis and motivation to further enriching the platform with higher-fidelity models and more noise components. https://lnkd.in/ghiGv8Fk

52

1 Comment

In my "free time" I attempted to create a video that explains how active learning in Gaussian Process Regression (GPR) works. The aim of active learning is clear: You want to minimize the number of samples while obtaining an accurate model or the optimal point. Imagine that one evaluation takes hours (e.g., from expensive computational fluid dynamics simulations)- surely, reducing the number of samples will be super beneficial. In this video, you can see that new samples are added at certain locations and the quality of the GPR prediction improves every time we do so. How do we add new samples? In the context of obtaining an accurate model, you will add new samples at locations where the uncertainty is at its highest. Such information can be obtained from GPR's uncertainty structure, which is naturally provided by GPR thanks to its probabilistic interpretation. In fact, we routinely deploy the active learning strategy in our research- a lot! One challenge about the application of machine learning in engineering is that we often face situations where we are constrained by a limited computational budget. In other words, it is not always about big data: We want to be as smart as possible with limited resources that we have. If we can achieve our objective with a small budget, then why not?

103

7 Comments

X-RAY SOURCE DESIGN This tutorial illustrates techniques to characterize a high-power, rotating-target X-ray sources. The figure shows a specific examples. Critical information was required in two areas: * The power flux on a beryllium window from backscattered electrons. * The X-ray flux and spectrum exiting the window. I used a code suite that handled electric field calculations, self-consisent electron beam dynamics and Monte Carlo simulations of electron/photon/positron interactions in matter to perform several calculations: * Two-dimensional backscatter calculation * Electric field calculations * Three-dimensional electron backscatter distribution * Backscattered electron flux on the X-ray window * X-rays transmitted through the beryllium window Link to the tutorial: https://lnkd.in/gNjXJyMn

26

Engineering energy levels showing thrust and power, at each energy level are added. Also, quoting the paper, "95% of all the energy in the universe is due to stochastic fluctuations of spacetime, caused by "a fundamental breakdown in predictability." Gravity is classical general relativity, dark matter effects are entropic, equivalent to QI [20], and emergent classical gravity is asymptotically free because stochastic white noise vanishes at short distances [21]." And references [20] and [21] are added. Every laboratory needs a standard thrust capacitor. 😎 https://lnkd.in/gWunjcJH

3