Search NASA⌕ Search

SEARCH · Search NASA

Results for “trigger algorithms”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 469 records · Page 26

The artificial retina processor for track reconstruction at the LHC crossing rate

We present results of an R&D study for a specialized processor capable of precisely reconstructing, in pixel detectors, hundreds of charged-particle tracks from high-energy collisions at 40 MHz rate. We apply a highly parallel pattern-recognition algorithm, inspired by studies of the processing of visual images by the brain as it happens in nature, and describe in detail an efficient hardware implementation in high-speed, high-bandwidth FPGA devices. This is the first detailed demonstration of reconstruction of offline-quality tracks at 40 MHz and makes the device suitable for processing Large Hadron Collider events at the full crossing frequency.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Swift Burst Alert Telescope (BAT) Instrument Response

The Burst Alert Telescope (BAT), a large coded aperture instrument with a wide field-of-view (FOV), provides the gamma-ray burst triggers and locations for the Swift Gamma-Ray Burst Explorer. In addition to providing this imaging information, BAT will perform a 15 keV - 150 keV all-sky hard x-ray survey based on the serendipitous pointings resulting from the study of gamma-ray bursts and will also monitor the sky for transient hard x-ray sources. For BAT to provide spectral and photometric information for the gamma-ray bursts, the transient sources and the all-sky survey, the BAT instrument response must be determined to an increasingly greater accuracy. In this talk, we describe the BAT instrument response as determined to an accuracy suitable for gamma-ray burst studies. We will also discuss the public data analysis tools developed to calculate the BAT response to sources at different energies and locations in the FOV. The level of accuracy required for the BAT instrument response used for the hard x-ray survey is significantly higher because this response must be used in the iterative clean algorithm for finding fainter sources. Because the bright sources add a lot of coding noise to the BAT sky image, fainter sources can be seen only after the counts due to the bright sources are removed. The better we know the BAT response, the lower the noise in the cleaned spectrum and thus the more sensitive the survey. Since the BAT detector plane consists of 32768 individual, 4 mm square CZT gamma-ray detectors, the most accurate BAT response would include 32768 individual detector response functions to separate mask modulation effects from differences in detector efficiencies! We describe OUT continuing work to improve the accuracy of the BAT instrument response and will present the current results of Monte Carlo simulations as well as BAT ground calibration data.

Parsons, A.↗

In situ monitoring of the stopped muon flux at Mu2e

The specific object of Mu2e search is the neutrino-less coherent $\mu^-\rightarrow e^-$ conversion in the field of an aluminum nucleus and the signal is a monoenergetic electron of energy $ \approx 104.97$ MeV. This process is forbidden in the Standard Model but allowed in many of its extensions: with minimal changes to include neutrino masses and oscillation, the Branching Ratio of this process, or the similarly interesting $\mu\rightarrow e \gamma$ decay, is expected to be of the order of $\mathcal{O}(10^{-54})$. Values like these are below any currently achievable experimental sensitivity, and the observation of a signal would be unambiguous evidence of New Physics. The upper limit on the muon conversion was set by SINDRUM II at $7\times10^{-13}$ (90\% C.L.) and the goal of the Mu2e collaboration is an improvement of 4 order of magnitudes.\\ The Mu2e experiment can be conceptually divided into three stages: interaction of the primary proton pulse with the tungsten target a nd production of $\pi$ and $\mu$; collection and transport of the produced particles down to the aluminum stopping target; interaction with the stopping target and measurement of the output particles. The measurement of the stopped muon flux is of cardinal importance for the normalization of the Mu2e results. The baseline design includes two detectors developed to measure this flux, counting $\gamma$ emitted by the muonic atoms: HPGe and LaBr$_3$(Ce). The studies performed by the Mu2e Collaboration show that these two systems can be reliably used to determine the overall normalization.\\ The number of stopped muons is proportional to the number of protons on target, which itself depends on the extraction system for the proton beam. The Mu2e beam delivery system will use a resonant extraction to create the proton pulses and this method is characterized by intensity fluctuations on the time scale of milliseconds. These fluctuations, aside from the proportional fluctuation of t he number of muons, have a non-trivial impact on the overall! performance of the apparatus. For example, higher intensity proton bunches translate into higher veto rates by the Cosmic Rays Veto system, which is triggered by secondary particles generated by the proton interaction. Another example is the effect of the instantaneous luminosity on the reconstruction efficiency: higher intensity translates into higher detector occupancy, which in turn reduces the reconstruction efficiency.\\ The two cited detectors encounter limitations when trying to monitor the flux at the millisecond timescales: HPGe is by construction a slow detector while the rate of emission of the $\gamma (1809\textrm{ keV})$ the LaBr$_3$(Ce) will measure is too low. As of today, no system allows to monitor these fluctuations and the goal of this Thesis is to try filling this gap. The method we developed to monitor the fluctuations of the stopped muon flux, and by extension, the fluctuation in the proton pulses intensity is based on counting the number of muons capt ured in the stopping target by counting the number of protons produced in the process of the nuclear muon capture. This is one of the possible processes a stopped muon can undergo and it can lead to the ejection of charged particles.\\ The reconstruction algorithms were developed by the Mu2e collaboration with the explicit (and sometimes implicit) aim to reconstruct tracks of electrons in an energy range of a conversion electron. The protons we are interested in behave quite differently from these electrons and the cardinal task of this study has been to tailor the reconstruction routines to these particles. This has been done by analyzing simulated events consisting of a single proton to optimize the procedure and understand the features of these tracks. A major difference with respect to the electrons is the non-relativistic nature of $100\div300$ MeV$/c$ protons: accounting for the low velocity is both a challenge and a unique signature of the tracks.\\ The use of a pulse d beam and the need to allow produced $\pi$ to decay and $\m! u$ to int! eract in the stopping target forces the definition of a Mu2e \textit{event}: an event is comprised by everything which happens in a $1.7\ \mu$s window between the incidence two consecutive proton pulses. This means that the reconstruction will be performed in an environment crowded by many different particles. The next step of this study has been to estimate the performance of the method in a fully-fledged Mu2e simulated event and to study what kind of selections would improve it.\\ Our results are satisfactory since the number of reconstructed protons is significant (few per event), and show that a monitor on the timescale of milliseconds is indeed possible.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Reactive burn models and ignition & growth concept

Plastic-bonded explosives are heterogeneous materials. Experimentally, shock initiation is sensitive to small amounts of porosity, due to the formation of hot spots (small localized regions of high temperature). This leads to the Ignition and Growth concept, introduced by Lee and Tarver in 1980, as the basis for reactive burn models. A homogeneized burn rate needs to account for three mesoscale physical effects (i) the density of burnt hot spots, which depends on the lead shock strength; (ii) the growth of the burn fronts triggered by hot spots, which depends on the local deflagration speed; (iii) a geometric factor that accounts for the overlap of deflagration wavelets from adjacent hot spots. These effects can be combined and the burn model defined by specifying the reaction progress variable {lambda}(t) as a function of a dimensionless reaction length {tau}{sub hs}(t)/{ell}{sub hs}, rather than by xpecifying an explicit burn rate. The length scale {ell}{sub hs} is the average distance between hot spots, which is proportional to [N{sub hs}(P{sub s})]{sup -1/3}, where N{sub hs} is the number density of hot spots activated by the lead shock. The reaction length {tau}{sub hs}(t) = {line_integral}{sub 0}{sup t} D(P(t'))dt' is the distance the burn front propagates from a single hot spot, where D is the deflagration speed and t is the time since the shock arrival. A key implementation issue is how to determine the lead shock strength in conjunction with a shock capturing scheme. They have developed a robust algorithm for this purpose based on the Hugoniot jump condition for the energy. The algorithm utilizes the time dependence of density, pressure and energy within each cell. The method is independent of the numerical dissipation used for shock capturing. It is local and can be used in one or more space dimensions. The burn model has a small number of parameters which can be calibrated to fit velocity gauge data from shock initiation experiments.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A novel Lagrangian approach for the stable numerical simulation of fault and fracture mechanics

The simulation of the mechanics of geological faults and fractures is of paramount importance in several applications, such as ensuring the safety of the underground storage of wastes and hydrocarbons or predicting the possible seismicity triggered by the production and injection of subsurface fluids. However, the stable numerical modeling of ground ruptures is still an open issue. The present work introduces a novel formulation based on the use of the Lagrange multipliers to prescribe the constraints on the contact surfaces. The variational formulation is modified in order to take into account the frictional work along the activated fault portion according to the principle of maximum plastic dissipation. The numerical model, developed in the framework of the Finite Element method, provides stable solutions with a fast convergence of the non-linear problem. The stabilizing properties of the proposed model are emphasized with the aid of a realistic numerical example dealing with the generation of ground fractures due to groundwater withdrawal in arid regions. - Highlights: • A numerical model is developed for the simulation of fault and fracture mechanics. • The model is implemented in the framework of the Finite Element method and with the aid of Lagrange multipliers. • The proposed formulation introduces a new contribution due to the frictional work on the portion of activated fault. • The resulting algorithm is highly non-linear as the portion of activated fault is itself unknown. • The numerical solution is validated against analytical results and proves to be stable also in realistic applications.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Electron Cyclotron power management for control of Neoclassical Tearing Modes in the ITER baseline scenario

Time-dependent simulations are used to evolve plasma discharges in combination with a Modified Rutherford equation (MRE) for calculation of Neoclassical Tearing Mode (NTM) stability in response to Electron Cyclotron (EC) feedback control in ITER. The main application of this integrated approach is to support the development of control algorithms by analyzing the plasma response with physics-based models and to assess how uncertainties in the detection of the magnetic island and in the EC alignment affect the ability of the ITER EC system to fulfill its purpose. These simulations indicate that it is critical to detect the island as soon as possible, before its size exceeds the EC deposition width, and that maintaining alignment with the rational surface within half of the EC deposition width is needed for stabilization and suppression of the modes, especially in the case of modes with helicity (2,1). A broadening of the deposition profile, for example due to wave scattering by turbulence fluctuations or not well aligned beams, could even be favorable in the case of the (2,1)-NTM, by relaxing an over-focussing of the EC beam and improving the stabilization at the mode onset. Pre-emptive control reduces the power needed for suppression and stabilization in the ITER baseline discharge to a maximum of 5 MW, which should be reserved and available to the Upper Launcher during the entire flattop phase. By assuming continuous triggering of NTMs, with pre-emptive control ITER would be still able to demonstrate a fusion gain of Q=10.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Brain-Computer Interfaces for 1-D and 2-D Cursor Control: Designs Using Volitional Control of the EEG Spectrum or Steady-State Visual Evoked Potentials

We have developed and tested two EEG-based brain-computer interfaces (BCI) for users to control a cursor on a computer display. Our system uses an adaptive algorithm, based on kernel partial least squares classification (KPLS), to associate patterns in multichannel EEG frequency spectra with cursor controls. Our first BCI, Target Practice, is a system for one-dimensional device control, in which participants use biofeedback to learn voluntary control of their EEG spectra. Target Practice uses a KF LS classifier to map power spectra of 30-electrode EEG signals to rightward or leftward position of a moving cursor on a computer display. Three subjects learned to control motion of a cursor on a video display in multiple blocks of 60 trials over periods of up to six weeks. The best subject s average skill in correct selection of the cursor direction grew from 58% to 88% after 13 training sessions. Target Practice also implements online control of two artifact sources: a) removal of ocular artifact by linear subtraction of wavelet-smoothed vertical and horizontal EOG signals, b) control of muscle artifact by inhibition of BCI training during periods of relatively high power in the 40-64 Hz band. The second BCI, Think Pointer, is a system for two-dimensional cursor control. Steady-state visual evoked potentials (SSVEP) are triggered by four flickering checkerboard stimuli located in narrow strips at each edge of the display. The user attends to one of the four beacons to initiate motion in the desired direction. The SSVEP signals are recorded from eight electrodes located over the occipital region. A KPLS classifier is individually calibrated to map multichannel frequency bands of the SSVEP signals to right-left or up-down motion of a cursor on a computer display. The display stops moving when the user attends to a central fixation point. As for Target Practice, Think Pointer also implements wavelet-based online removal of ocular artifact; however, in Think Pointer muscle artifact is controlled via adaptive normalization of the SSVEP. Training of the classifier requires about three minutes. We have tested our system in real-time operation in three human subjects. Across subjects and sessions, control accuracy ranged from 80% to 100% correct with lags of 1-5 seconds for movement initiation and turning.

Trejo, Leonard J.↗

Composited structure of non-precipitating shallow cumulus clouds

The normalized distributions of thermodynamic and dynamical variables both within and outside shallow clouds are investigated through a composite algorithm using large-eddy simulations of oceanic and continental cases. The normalized magnitude is maximum near the cloud centre and decreases outwards. While relative humidity (RH) and cloud liquid water ( q l ) decrease smoothly to match the environment, the vertical velocity, virtual potential temperature (θ v ), and potential temperature (θ) perturbations have more complicated behaviour towards the cloud boundary. Below the inversion layer, θ' v becomes negative before the vertical velocity has turned from an updraft to a subsiding shell outside the cloud, indicating the presence of a transition zone where the updraft is negatively buoyant. Due to the downdraft outside the cloud and enhanced horizontal turbulent mixing across the edge, the normalized turbulent kinetic energy (TKE) and horizontal turbulent kinetic energy (HTKE) decrease more slowly from the cloud centre outwards than the thermodynamic variables. The distributions all present asymmetric structures in response to the vertical wind shear, with more negatively buoyant air, stronger downdrafts, and larger TKE on the downshear side. We discuss several implications of the distributions for theoretical models and parameterizations. Positive buoyancy near the cloud base is mostly due to the virtual effect of water vapour, emphasizing the role of moisture in triggering. The mean vertical velocity is found to be approximately half the maximum vertical velocity within each cloud, providing a constraint to achieve possible power-law distributions for some models. Finally, the normalized distributions for different variables are used to estimate the vertical heat and moisture fluxes within clouds. The results suggest that distributions near the cloud edge and variability of maximum perturbations need careful treatment. The fluxes are underestimated in the inversion layer because cloud-top downdrafts cannot be captured well.

54 ENVIRONMENTAL SCIENCES↗

Deep Learning Estimation of Daily Ground–Level NO 2 Concentrations from Remote Sensing Data

The limited number of nitrogen dioxide (NO 2 ) surface measurements calls for the development of highly accurate approaches to estimating surface NO 2 concentrations. In this study, we leverage a new satellite instrument, the TROPOspheric Monitoring Instrument (TROPOMI), along with other predictor variables, to estimate daily surface NO 2 concentrations over Texas in 2019. We use the deep convolutional neural network (Deep-CNN), an advanced deep learning algorithm, to obtain estimates and achieve a correlation coefficient (R) of 0.91, an index of agreement (IOA) of 0.95, and a mean absolute bias (MAB) of 1.75 ppb in surface NO 2 estimation. Additionally, we leverage a novel approach, SHapley Additive exPlanations (SHAP), to describe how Deep-CNN understands each predictor variable. The SHAP results show that the Deep-CNN model has an advanced understanding of the dataset, revealing that TROPOMI closely captures levels of NO 2 . In addition, we show the superiority of our Deep-CNN model at estimating surface NO 2 over other well-known machine learning and regression models in the field, including the support vector machines (SVM), random forest (RF), and multiple linear regression (MLR). Although SVM and RF show strong capabilities at estimating surface NO 2 concentrations, their accuracy is inferior to that of the Deep-CNN model, ranking second and third in model accuracy in this study. The MLR, however, shows a poor ability at NO 2 estimation and ranks last among all models. Furthermore, testing the impact of sample size on model performance, we also show that, compared to other models, Deep-CNN needs more samples to trigger its strength at surface NO 2 estimation.

54 ENVIRONMENTAL SCIENCES↗

The Center for Momentum Transport and Flow Organization in Plasmas - Final Scientific Report

Overview of University of Colorado Efforts: The University of Colorado group has focused on two primary fronts during the grant period: development of a variety of multi-point diagnostic and/or imaging analysis techniques, and momentum-transport related experiments on a variety of devices (NSTX at PPPL, CSDX at UCSD, LAPD at UCLA, DIII-D at GA). Experimental work has taken advantage of several diagnostic instruments, including fast-framing cameras for imaging of electron density fluctuations (either directly or using injected gas puffs), ECEI for imaging of electron temperature fluctuations, and multi-tipped Langmuir and magnetic probes for corroborating measurements of Reynolds and Maxwell stresses. Mode Characterization in CSDX: We have performed a series of experiments at the CSDX linear device at UCSD, in collaboration with Center PI G. Tynan's group. The experiments included a detailed study of velocity estimation techniques, including direct comparisons between Langmuir probes and image-based velocimetry from fast-framing camera data. We used the camera data in a second set of studies to identify the spatial and spectral structure of coherent modes, which illuminates wave behavior to a level of detail previously unavailable, and enables direct comparison of dispersion curves to theoretical estimates. In another CSDX study, similar techniques were used to demonstrate a controlled transition from nonlinearly coupled discrete eigenmodes to fully developed broadband turbulence. The axial magnetic field was varied from 40-240 mT, which drove the transition. At low magnetic fields, the plasma is dominated by drift waves. As the magnetic field is increased, a strong potential gradient at the edge introduces an ExB shear-driven instability. At the transition, another mode with signatures of a rotation-induced Rayleigh–Taylor instability appears at the central plasma region. Concurrently, large axial velocities were found in the plasma core. For larger magnetic fields, all of the instabilities co-exist, leading to rich plasma dynamics and fully developed broadband turbulence. Edge-Turbulence and Flow Experiments in NSTX: A series of Gas Puff Imaging (GPI) observations on NSTX revealed a quasi-periodic oscillation in the plasma edge preceding the L-H transition in a limited set of neutral beam heated plasmas. These ~3 kHz flow oscillations exhibit both long wavelength and long correlation lengths, suggesting they are zonal-flow-like. The flow oscillations are strongly correlated with modulations of the level of edge turbulence, thus the system appears to undergo a predator--prey-type limit-cycle preceding the L-H transition. However, a clear trigger for the L-H transition was not observed. Reynolds stress profiles were obtained directly from image velocimetry for L-mode periods ELM-Precursor Studies in NSTX: A separate study based on NSTX-GPI data captured the two-dimensional evolution of edge-localized mode (ELM) precursors. Precursor events were observed preceding ELMs and ELM-induced H–L back-transitions in radio-frequency heated H-mode plasmas, and the growth of the precursor mode through the ELM filamentation was imaged in the plane perpendicular to the local B-field. Strong edge intensity modulations appeared to propagate in the electron diamagnetic direction while steadily drifting radially outwards. Intensity fluctuations were observed at frequencies around 20 kHz and wavenumbers of 0.05-0.2 cm -1 . Upon growing to a trigger point, precursor fluctuations were seen to form filamentary structures and move into the scrape-off layer (SOL) explosively with radial velocities peaking at 8 km/s. Once in the SOL, filaments reverse their propagation direction and travel in the ion diamagnetic direction. Edge intensity fluctuations were strongly correlated with magnetic signals from Mirnov coils, and toroidally distributed coils estimated toroidal mode numbers of n=5-10. Quantitatively similar precursors have been observed in ohmic H-mode plasmas as well, though significantly fewer events are seen in the ohmic cases and none were observed in the near-threshold NBI H-modes studied. Development of Velocimetry and Image Analysis Techniques: Along with the experiments listed above, the Colorado group has continued to explore various velocimetry techniques and their range of validity. We have developed a “linear optical flow” code, which calculates smooth velocity maps while accurately assessing local regions of high curl. This is critical for separating spatial scales of velocity behavior, and thus transport. This code has recently come on-line, and we are currently using it to revisit a number of older datasets. Additionally, we have worked on developing pattern-recognition techniques for imaging diagnostics, based on established digital image compression algorithms. This has the potential to open the analysis of turbulent plasma behavior beyond the well-trodden Fourier and wavelet approaches. Finally, we have extended several of these image-analysis routines to multiple other diagnostic sets, including GPI datasets in NSTX and ECEI imaging on DIII-D. In one study, correlation maps were used to measure the full 2-D mode structure of drift-wave level structures using ECEI for the first time. This enabled direct comparison to gyrokinetic simulations using the GEM code, from which is was determined that TEM modes were being measured in DIII-D.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A Heterogeneous System for Eagle Detection, Deterrent, and Wildlife Collision Detection for Wind Turbines (Final Technical Report)

This report summarizes the design, implementation, and test of an integrated system for automated detection and deterrence of eagles, with included wind turbine blade strike detection and imaging functionality. A machine learning approach was used in conjunction with a 360° camera system for automated detection and classification of golden eagles. This was developed using footage obtained from trained golden eagles and other raptors, in collaboration with wildlife biologists and professional bird handlers. Oregon State University developed a visual deterrent system, which uses inflatable anthropomorphic sculptures with random, kinetic motion to deter eagles, and conducted limited field testing on live eagles; the deterrent can be triggered by the visual detection of eagles using the vision system. Finally, a multi-sensor module was developed that is mounted at the turbine blade root. This module measures vibration and other motions to detect blade strikes, and an integrated on-blade camera captures an image of any impacting objects. Long-term, this blade strike detection system is intended to support an automatic monitoring and certification system for the eagle detection and deterent system. Independent field testing of each system component is described. Testing of the integrated system on an operational wind turbine was conducted across three separate field tests. This includes multi-day fields tests on a General Electric 1.5MW wind turbine at the National Renewable Energy Laboratory (NREL) National Wind Technology Center (NWTC) in Boulder, CO in October 2018 and July 2019; installation procedures, test procedures, and a summary of collected data are presented. A third multi-day on-turbine field test is also presented, which was performed using a General Electric 1.5MW wind turbine at the North American Wind Research and Training Center (NAWRTC) at Mesalands Community College, Tucumcari, NM in April 2019. Across these field tests, the vision system was demonstrated using unmanned aerial vehicles (UAV), and the eagle classification algorithm was not tested; the visual deterrent system was demonstrated, including automatic, remote deployment following surrogate visual detections; and, multi-sensor on-blade data was recorded across multiple wind turbine operational conditions and through more than 100 surrogate blade strikes using soft projectiles, including the successful demonstration of automatic image capture of striking objects. This data set was also used for offline development and validation of enhanced collision detection algorithms. As summarized in this report, the development and field validation of an integrated detection, deterrent, and blade collision detection system represents a critical proof of concept for future technology development of related detection and deterrent technologies, where both deterrent as well as collision detection recording devices are needed for future siting, monitoring, and operation of wind turbine installations, both onshore and offshore.

17 WIND ENERGY↗

Even Higher-Level Synthesis: An Exploration of AI Hardware Accelerators using HLS4ML

With the rise of artificial intelligence, the popularization of deep learning, and a constantly evolving industry, the demand for flexible and efficient tools has never been greater. As algorithms grow more complex, their runtime and energy consumption increase exponentially. Customized hardware accelerators, long used for specific mathematical operations, remain essential for managing modern applications' computational and power demands. Hardware accelerators can speed up complex computations by orders of magnitude, but their manual design and verification processes are often challenging and time-consuming. High-Level Synthesis (HLS) provides a solution by transforming high-level algorithm descriptions, typically written in C++ or SystemC, into synthesizable RTL suitable for hardware implementation. This approach reduces development time for RTL engineers while offering flexibility beyond what traditional handwritten RTL can provide. We extended this capability to the machine-learning domain with the open-source framework hls4ml, which allows neural networks trained in Python frameworks like Tensorflow or PyTorch to be synthesized into efficient hardware representations for the traditional FPGA and ASIC flows. This breakthrough addresses the growing need for reduced design turnaround and easy verification of ML hardware accelerators with low latency and power efficiency constraints. During this tutorial, we will demonstrate how Python complements HLS by simplifying the ML design process, bridging the gap between software and hardware development. Attendees will explore how we translate neural networks modeled in Python into fixed-point C++ models suitable for HLS workflows. We will dive into strategies like Value-Range Analysis and Quantization-Aware Training, which optimize these designs for deployment and evaluate their accuracy, power consumption, and energy efficiency. To exemplify these concepts, experts from Fermilab will share their experiences applying this technology to high-energy physics experiments, where real-time, low-latency processing is critical. Over the years, Fermilab engineers have demonstrated how deep neural networks, optimized for hardware using hls4ml, can meet the stringent requirements of trigger systems at the CERN Large Hadron Collider. These systems rely on rapid decision-making to process immense data volumes while retaining only the most relevant events for further analysis. The application of hls4ml has also been extended to innovative technologies like smart pixel arrays. These smart pixels integrate ML inference capabilities directly into sensor devices, enabling localized data processing at the pixel level. This approach drastically reduces the need to transmit raw data to external processing units, significantly decreasing power consumption and latency. By embedding neural networks within the pixel architecture, the smart pixels can identify and prioritize relevant data in real time, providing a highly efficient solution for edge computing in scenarios such as particle detectors and imaging systems. Fermilab's work highlights the potential of hardware-accelerated ML in scenarios where both speed and power efficiency are mission-critical. Through this tutorial, attendees will gain valuable insights into the challenges and solutions of deploying ML in hardware. Understanding how HLS and hls4ml streamline the development of neural network-based hardware accelerators is fundamental for the industry's future. Participants will learn how these technologies are shaping the future of AI and scientific computing.

Di Guglielmo, Giuseppe [Fermilab]↗

Spatiotemporal Patterns of Precipitation-Modulated Landslide Deformation From Independent Component Analysis of InSAR Time Series

We present that long-term landslide deformation is disruptive and costly in urbanized environments. We rely on TerraSAR-X satellite images (2009–2014) and an improved data processing algorithm (SqueeSAR™) to produce an exceptionally dense Interferometric Synthetic Aperture Radar ground deformation time series for the San Francisco East Bay Hills. Independent and principal component analyses of the time series reveal four distinct spatial and temporal surface deformation patterns in the area around Blakemont landslide, which we relate to different geomechanical processes. Two components of time-dependent landslide deformation isolate continuous motion and motion driven by precipitation-modulated pore pressure changes controlled by annual seasonal cycles and multiyear drought conditions. Two components capturing more widespread seasonal deformation separate precipitation-modulated soil swelling from annual cycles that may be related to groundwater level changes and thermal expansion of buildings. Lastly, high-resolution characterization of landslide response to precipitation is a first step toward improved hazard forecasting.

54 ENVIRONMENTAL SCIENCES↗

Hyper-Velocity Nanoparticle Plasma Jet as Fast Probe for Runaway Electrons in Tokamak Disruptions

The large energy focalized deposition of the runaway electron (RE) beam on the first wall can potentially yield to destructive effects on the reactor vessel during the explosive instability of a major disruption in a large tokamak (e.g. RE beam estimation is multi-MA of ~10 MeV energy in ITER.) Impurity injection is seen as an effective way to suppress REs after thermal quench (TQ). Injected material in solid state is advantageous, but it must be in small size grain form, with large surface-to-mass ratio to facilitate its rapid ablation within plasma. However, the results using shattered pellet injection (SPI), a technique aiming to satisfy these condition, indicated that the relatively unexplored RE evolution phases and RE beam-plasma interaction are far from being properly diagnosed and fully understood. We proposed the idea to probe REs with a pulsed power hyper-velocity large-mass nanoparticle plasma jet (NPPJ). NPPJ provides a high specific surface area and swift injection to core, hence fast increase of three basic parameters: 1) the electron density ne (bound and free through ionization, if the cold post-TQ plasma has a still high enough T e ), 2) the effective charge (Z eff ) which rises both the dynamical friction force and plasma resistivity, and, finally, 3) the value of the (Dreicer and critical) threshold E-field above which electrons are accelerated and become REs. But the fast transient phenomena of RE ”seed” creation, RE ”avalanche” multiplication, and RE beam formation and interaction with the background plasma all take place dominantly in the core plasma, where the induction E-field has its maximum. Thus, the RE probe success requires a rapid (1 - 2 ms) injection of sufficient mass able to penetrate through the strong toroidal magnetic field (2 - 5 T) over a long distance (~1 - 2 m) with a large assimilation fraction in the core plasma. As a RE probe, NPPJ is a unique combination of fast response with trigger-to-delivery time of few milliseconds, large mass-velocity (>75 mg at several km/s) and deep injection into post-TQ central plasma. The Objectives of this Category #1 project (concept development) were as follows: 1: We investigated the essential basic physics aspects of creating a plasma with ~10 15 to 10 16 cm -3 boron nitride (BN) NP density suited for a pulsed power plasma gun acceleration to several km/s. We concluded that, for RE NPPJ probe, BN (diameter Ø10 to 100 nm NPs) are in all respects usable. 2: We advanced solutions to get BN NPPJ by adapting FAR-TECH’s technology for C 60 NPPJ. 3: We developed self-consistent models for RE current density j RE (R; t) evolution in the TQ aftermath, with fast and simultaneous increase of n e (R; t) and Z eff (R; t) > 1 upon the specific injection of C 60 (i.e. C ions) and BN NP (i.e. B and N ions), respectively. We showed that prompt post-TQ injection and high velocity are of essence in RE probe diagnostics and suppression. 4: We used the in-house HEM-2D code to simulate the C$_{60}^{q+}$ NPPJ transverse penetration through tokamak-level B-field (BT ~2 to 5 T) and distance (~1 to 2m), with fragmentation of C$_{60}^{q+}$ and developed the algorithm for BN Q+ NPPJ ablative sublimation of BN Q+ NP by the collisions with post-TQ plasma electrons. HEM-2D results indicated that fragmenting C$_{60}^{1+}$ NPPJ penetrates up to ~60 cm in 100 us when injecting at 1.6 ms after TQ and gradually released C 2 fragments and C ions.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Tracking Object Existence From an Autonomous Patrol Vehicle

An autonomous vehicle patrols a large region, during which an algorithm receives measurements of detected potential objects within its sensor range. The goal of the algorithm is to track all objects in the region over time. This problem differs from traditional multi-target tracking scenarios because the region of interest is much larger than the sensor range and relies on the movement of the sensor through this region for coverage. The goal is to know whether anything has changed between visits to the same location. In particular, two kinds of alert conditions must be detected: (1) a previously detected object has disappeared and (2) a new object has appeared in a location already checked. For the time an object is within sensor range, the object can be assumed to remain stationary, changing position only between visits. The problem is difficult because the upstream object detection processing is likely to make many errors, resulting in heavy clutter (false positives) and missed detections (false negatives), and because only noisy, bearings-only measurements are available. This work has three main goals: (1) Associate incoming measurements with known objects or mark them as new objects or false positives, as appropriate. For this, a multiple hypothesis tracker was adapted to this scenario. (2) Localize the objects using multiple bearings-only measurements to provide estimates of global position (e.g., latitude and longitude). A nonlinear Kalman filter extension provides these 2D position estimates using the 1D measurements. (3) Calculate the probability that a suspected object truly exists (in the estimated position), and determine whether alert conditions have been triggered (for new objects or disappeared objects). The concept of a probability of existence was created, and a new Bayesian method for updating this probability at each time step was developed. A probabilistic multiple hypothesis approach is chosen because of its superiority in handling the uncertainty arising from errors in sensors and upstream processes. However, traditional target tracking methods typically assume a stationary detection volume of interest, whereas in this case, one must make adjustments for being able to see only a small portion of the region of interest and understand when an alert situation has occurred. To track object existence inside and outside the vehicle's sensor range, a probability of existence was defined for each hypothesized object, and this value was updated at every time step in a Bayesian manner based on expected characteristics of the sensor and object and whether that object has been detected in the most recent time step. Then, this value feeds into a sequential probability ratio test (SPRT) to determine the status of the object (suspected, confirmed, or deleted). Alerts are sent upon selected status transitions. Additionally, in order to track objects that move in and out of sensor range and update the probability of existence appropriately a variable probability detection has been defined and the hypothesis probability equations have been re-derived to accommodate this change. Unsupervised object tracking is a pervasive issue in automated perception systems. This work could apply to any mobile platform (ground vehicle, sea vessel, air vehicle, or orbiter) that intermittently revisits regions of interest and needs to determine whether anything interesting has changed.

Wolf, Michael↗

Smart Pixels: In-pixel AI for on-sensor data filtering

We present a smart pixel prototype readout integrated circuit (ROIC) designed in CMOS 28 nm bulk process, with in-pixel implementation of an artificial intelligence (AI) / machine learning (ML) based data filtering algorithm designed as proof-of-principle for a Phase III upgrade at the Large Hadron Collider (LHC) pixel detector. The first version of the ROIC consists of two matrices of 256 smart pixels, each 25$\times$25 $\mu$m$^2$ in size. Each pixel consists of a charge-sensitive preamplifier with leakage current compensation and three auto-zero comparators for a 2-bit flash-type ADC. The frontend is capable of synchronously digitizing the sensor charge within 25 ns. Measurement results show an equivalent noise charge (ENC) of $\sim$30e$^-$ and a total dispersion of $\sim$100e$^-$ The second version of the ROIC uses a fully connected two-layer neural network (NN) to process information from a cluster of 256 pixels to determine if the pattern corresponds to highly desirable high-momentum particle tracks for selection and readout. The digital NN is embedded in-between analog signal processing regions of the 256 pixels without increasing the pixel size and is implemented as fully combinatorial digital logic to minimize power consumption and eliminate clock distribution, and is active only in the presence of an input signal. The total power consumption of the neural network is $\sim$ 300 $\mu$W. The NN performs momentum classification based on the generated cluster patterns and even with a modest momentum threshold, it is capable of 54.4% - 75.4% total data rejection, opening the possibility of using the pixel information at 40MHz for the trigger. The total power consumption of analog and digital functions per pixel is $\sim$ 6 $\mu$W per pixel, which corresponds to $\sim$ 1 W/cm$^2$ staying within the experimental constraints.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Smart Pixels: In-pixel AI for on-sensor data filtering

We present a smart pixel prototype readout integrated circuit (ROIC) designed in CMOS 28 nm bulk process, with in-pixel implementation of an artificial intelligence (AI) / machine learning (ML) based data filtering algorithm designed as proof-of-principle for a Phase III upgrade at the Large Hadron Collider (LHC) pixel detector. The first version of the ROIC consists of two matrices of 256 smart pixels, each 25$\times$25 µm\textsuperscript{2} in size. Each pixel consists of a charge-sensitive preamplifier with leakage current compensation and three auto-zero comparators for a 2-bit flash-type ADC. The frontend is capable of synchronously digitizing the sensor charge within 25 ns. Measurement results show an equivalent noise charge (ENC) of $\sim$30e\textsuperscript{-} and a total dispersion of $\sim$100e\textsuperscript{-} The second version of the ROIC uses a fully connected two-layer neural network (NN) to process information from a cluster of 256 pixels to determine if the pattern corresponds to highly desirable high-momentum particle tracks for selection and readout. The digital NN is embedded in-between analog signal processing regions of the 256 pixels without increasing the pixel size and is implemented as fully combinatorial digital logic to minimize power consumption and eliminate clock distribution, and is active only in the presence of an input signal. The total power consumption of the neural network is $\sim$ 300 $\mu$W. The NN performs momentum classification based on the generated cluster patterns and even with a modest momentum threshold, it is capable of 54.4\% – 75.4\% total data rejection, opening the possibility of using the pixel information at 40MHz for the trigger. The total power consumption of analog and digital functions per pixel is $\sim$ 6 $\mu$W per pixel, which corresponds to $\sim$ 1 W/cm\textsuperscript{2} staying within the experimental constraints.

Parpillon, Benjamin↗

A real-time machine learning-based disruption predictor in DIII-D

A disruption prediction algorithm, named DPRF (Disruption Prediction using Random Forests), has run in real-time in the DIII-D plasma control system (PCS) over more than 900 discharges. DPRF naturally provides a probability mapping associated to its predictions, i.e. the disruptivity signal, now incorporated in the DIII-D PCS. Here, we report on disruption prediction accomplishments in terms of shot-by-shot performances, by simulating alarms on each discharge as in the PCS framework. Due to the optimised performance metric chosen to evaluate DPRF, we find that almost all disruptive discharges are detected in average with few hundred of milliseconds of warning time, but this comes at a high cost of false alarms produced. Performances are not satisfying ITER requirements, where the success rate has to be higher than 95%, but this is not completely unexpected. DPRF is trained on many years of major disruptions occurring during the flattop phase of the plasma current on DIII-D, but without any differentiation by cause. Furthermore, we find that DPRF is affected by a relatively high fraction of false alarms occurring during the first 500 milliseconds from the flattop onset. This subtle effect, more evident on the discharges where DPRF ran in real-time, can be marginalized by taking specific precautions on the validity range of the predictions, and performances do improve. Even if presently burdened by some limitations, DPRF provides an incredible and novel advantage. Thanks to the feature contribution analysis (e.g., the identification of which signals contributed to triggering an alarm), it is possible to interpret and explain DPRF predictions. It is the first time that such interpretability features is exploited by a disruption predictor: by uncovering the disruption events' causes, a better understanding of disruption dynamics is achieved, and a clear path toward the design of disruption avoidance strategies can be provided.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗