Search NASA⌕ Search

SEARCH · Search NASA

Results for “Deep generative model”

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 811 records · Page 45

Search for vector-like leptons coupling to first- and second-generation Standard Model leptons in pp collisions at s = 13 TeV with the ATLAS detector

A search for pair production of vector-like leptons coupling to first- and second-generation Standard Model leptons is presented. The search is based on a dataset of proton-proton collisions at s$$ \sqrt{s} $$ = 13 TeV recorded with the ATLAS detector during Run 2 of the Large Hadron Collider, corresponding to an integrated luminosity of 140 fb−1. Events are categorised depending on the flavour and multiplicity of leptons (electrons or muons), as well as on the scores of a deep neural network targeting particular signal topologies according to the decay modes of the vector-like leptons. In each of the signal regions, the scalar sum of the transverse momentum of the leptons and the missing transverse momentum is analysed. The main background processes are estimated using dedicated control regions in a simultaneous fit with the signal regions to data. No significant excess above the Standard Model background expectation is observed and limits are set at 95% confidence level on the production cross-sections of vector-like electrons and muons as a function of the vector-like lepton mass, separately for SU(2) doublet and singlet scenarios. The resulting mass lower limits are 1220 GeV (1270 GeV) and 320 GeV (400 GeV) for vector-like electrons (muons) in the doublet and singlet scenarios, respectively.

Aad, G↗

Classification of machine learning frameworks for data-driven thermal fluid models

Thermal fluid processes are inherently multi-physics and multi-scale, involving mass-momentum-energy transport phenomena at multiple scales. Thermal fluid simulation (TFS) is based on solving conservative equations, for which – except for “first-principles” direct numerical simulation – closure relations (CRs) are required to provide microscopic interactions or so-called sub-grid-scale physics. In practice, TFS is realized through reduced-order modeling, and its CRs as low-fidelity models can be informed by observations and data from relevant and adequately evaluated experiments and high-fidelity simulations. This paper is focused on data-driven TFS models, specifically on their development using machine learning (ML). Five ML frameworks are introduced including physics-separated ML (PSML or Type I ML), physics-evaluated ML (PEML or Type II ML), physics-integrated ML (PIML or Type III ML), physics-recovered (PRML or Type IV ML), and physics-discovered ML (PDML or Type V ML). The frameworks vary in their performance for different applications depending on the level of knowledge of governing physics, source, type, amount and quality of available data for training. Notably, outlined for the first time in this paper, Type III models present stringent requirements on modeling, substantial computing resources for training, and high potential in extracting value from “big data” in thermal fluid research. Here the current paper demonstrates and investigates ML frameworks in three examples. First, we utilize the heat diffusion equation with a nonlinear conductivity model formulated by convolutional neural networks (CNNs) and feedforward neural networks (FNNs) to illustrate the applications of Type I, Type II, Type III, and Type V ML. The results indicate a preference for Type II ML under deficient data support. Type III ML can effectively utilize field data, potentially generating more robust predictions than Type I and Type II ML. CNN-based closures exhibit more predictability than FNN-based closures, but CNN-based closures require more training data to obtain accurate predictions. Second, we illustrate how to employ Type I ML and Type II ML frameworks for data-driven turbulence modeling using reference works. Third, we demonstrate Type I ML by building a deep FNN-based slip closure for two-phase flow modeling. The results show that deep FNN-based closures exhibit a bounded error in the prediction domain.

42 ENGINEERING↗

Local atomic structure and lattice defect analysis in heavily Co-doped ZnS thin films using X-ray absorption fine structure spectroscopy

Iin this work, X-ray absorption fine structure (XAFS) spectroscopy was used to address both the local atomic structure and the local electronic structure around Zn, Co, and S in Co-doped ZnS thin films. X-ray absorption near-edge spectroscopy (XANES) revealed changes in the strong pre-edge feature of the S K-edge spectra that depend on the Co concentration and are related to atomic bound-state transitions through hybridization between $3d$ transition metal and $sp$ host semiconductor electronic states. These changes reveal intrinsic effects, such as S vacancies and dopant effects. From the Co K-edge XANES spectra, substitution of Co 2+ into ZnS is observed. The local atomic structure around the dopant, obtained by XAFS spectroscopy, indicates a reduction of the Co–S bond length. Optical analysis shows deep absorption centers or intermediate states in the bandgap, ascribed to $d$-$d$ transitions in the Co atoms that modify the electronic band structure and cause a decrease in the bandgap as a function of Co concentration. This work provides direct evidence of dopant effects in heavily Co-doped ZnS, including bond distances and Debye-Waller factors, providing key elements to generate a realistic atomic structural model, which is crucial to understand the observed electronic properties of this material.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Simulated Feasibility of 3-D Lightning Mapping From Space

In addition to the awe it inspires, lightning can illuminate the microphysical processes hidden away within deep convection. The current generation of space-based lightning mapping uses mostly 2-D optical imaging to connect overall flash characteristics to their parent storm dynamics, but are missing a dimension’s worth of information. With lightning now classified as an essential climate variable, future spaceborne mappers will need improved capabilities to take advantage of the 3-D structure of lightning flashes to support meteorological and climate modeling. We report here on a study of the feasibility of high-resolution 3-D lightning mapping using a radio frequency (RF)-based network of satellites from low-Earth orbit (LEO). Lightning sources are simulated using existing lightning mapping array (LMA) tools, modified for orbital detection, and spatially reconstructed using a Levenberg–Marquardt geolocation algorithm to assess sources of uncertainty in these solutions. We analyze the benefits and limitations of this approach compared to existing orbital and ground-based methods. Results of this study show that lightning can be mapped in 3-D with a vertical location accuracy better than 2 km using as few as five satellites in LEO capable of measuring the time-of-arrival of impulsive RF signals in the very high-frequency (VHF) band. The consequence of this study is that high-resolution, spaceborne 3-D mapping of lightning is achievable across most of the globe, having crucial implications for our understanding of not only lightning, but also severe weather development, climate science, and more.

47 OTHER INSTRUMENTATION↗

Ultra-deep mutant spectrum profiling: improving sequencing accuracy using overlapping read pairs

Backgound: High throughput sequencing is beginning to make a transformative impact in the area of viral evolution. Deep sequencing has the potential to reveal the mutant spectrum within a viral sample at high resolution, thus enabling the close examination of viral mutational dynamics both within- and between-hosts. The challenge however, is to accurately model the errors in the sequencing data and differentiate real viral mutations, particularly those that exist at low frequencies, from sequencing errors. Results: We demonstrate that overlapping read pairs (ORP) -- generated by combining short fragment sequencing libraries and longer sequencing reads -- significantly reduce sequencing error rates and improve rare variant detection accuracy. Using this sequencing protocol and an error model optimized for variant detection, we are able to capture a large number of genetic mutations present within a viral population at ultra-low frequency levels (<0.05%). Conclusions: Our rare variant detection strategies have important implications beyond viral evolution and can be applied to any basic and clinical research area that requires the identification of rare mutations.

59 BASIC BIOLOGICAL SCIENCES↗

Evaluating E3SM Global Storm‐Resolving Model Simulations of Deep Convection: Insights From DP‐SCREAM During TRACER

Global Storm-Resolving Models (GSRMs) are becoming increasingly vital for advancing climate modeling and improving the prediction of extreme weather events. Houston, a coastal region frequently affected by deep convective storms, offers an ideal setting to evaluate the ability of GSRMs to simulate deep convection. This study assesses the performance of the Doubly Periodic Simple Cloud-Resolving E3SM (Energy Exascale Earth System Model) Atmosphere Model (DP-SCREAM) using observations from the TRacking Aerosol Convection interactions ExpeRiment (TRACER) campaign. DP-SCREAM effectively reproduces the diurnal cycles of clouds and precipitation, demonstrating much greater skill than the E3SM single column model. The DP-SCREAM is demonstrated to be applicable to coastal regions, partially due to the forcing data sets already capturing the influence of breezes. DP-SCREAM also replicates biases persistent in the global version of SCREAM: the underrepresentation of boundary layer shallow clouds, a lack of mid-level congestus clouds, and the popcorn convection, characterized by small and disorganized convective cells generating the strongest precipitation. To investigate these issues, two sensitivity experiments were conducted: increasing the mixing length and scaling up the buoyancy flux within the Simplified Higher Order Closure scheme. Increasing the mixing length improved mid-level congestus representation and reduced unrealistic early morning fog occurrence. Enhancing buoyancy flux only marginally improved the bias of underproduced big convective cells. In conclusion, an additional resolution sensitivity test at 0.5 km grid spacing demonstrated that a refined horizontal resolution alone is insufficient to resolve these biases.

54 ENVIRONMENTAL SCIENCES↗

Moment method and continued fraction expansion in Floquet operator Krylov space

Recursion methods such as Krylov techniques map complex dynamics to an effective noninteracting problem in one dimension. For example, the operator Krylov space for Floquet dynamics can be mapped to the dynamics of an edge operator of the one-dimensional Floquet inhomogeneous transverse field Ising model (ITFIM), where the latter, after a Jordan-Wigner transformation, is a Floquet model of noninteracting Majorana fermions and the couplings correspond to Krylov angles. We present an application of this showing that a moment method exists where given an autocorrelation function, one can construct the corresponding Krylov angles and from that the corresponding Floquet ITFIM. Consequently, when no solutions for the Krylov angles are obtained, it indicates that the autocorrelation is not generated by unitary dynamics. We highlight this by studying certain special cases: stable m-period dynamics derived using the method of continued fractions, exponentially decaying, and power-law decaying stroboscopic dynamics. Remarkably, our examples of stable m-period dynamics correspond to m-period edge modes for the Floquet ITFIM where, deep in the chain, the couplings correspond to a critical phase. Furthermore our results pave the way to engineer Floquet systems with desired properties of edge modes and also provide examples of persistent edge modes in gapless Floquet systems.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

RX-ADS: Interpretable Anomaly Detection Using Adversarial ML for Electric Vehicle CAN Data

Recent year has brought considerable advancements in Electric Vehicles (EVs) and associated infrastructures/ communications. Intrusion Detection Systems (IDS) are widely deployed for anomaly detection in such critical infrastructures. This paper presents an Interpretable Anomaly Detection System (RX-ADS) for intrusion detection in CAN protocol communication in EVs. Contributions include: 1) window based feature extraction method; 2) deep Autoencoder based anomaly detection method; and 3) adversarial machine learning based explanation generation methodology. The presented approach was tested on two benchmark CAN datasets: OTIDS and Car Hacking. The anomaly detection performance of RX-ADS was compared against the state-of-the-art approaches on these datasets: HIDS and GIDS. The RX-ADS approach presented performance comparable to the HIDS approach (OTIDS dataset) and has outperformed HIDS and GIDS approaches (Car Hacking dataset). Further, the proposed approach was able to generate explanations for detected abnormal behaviors arising from various intrusions. Furthermore, these explanations were later validated by information used by domain experts to detect anomalies. Other advantages of RX-ADS include: 1) the method can be trained on unlabeled data; 2) explanations help experts in understanding anomalies and root course analysis, and also help with AI model debugging and diagnostics, ultimately improving user trust in AI systems.

42 ENGINEERING↗

Computationally efficient models for aqueous organic redox flow batteries

The rising usage of intermittent energy has garnered the need for large scale energy storage systems. Redox flow batteries (RFB) based energy storage system shows promising potential. Numerical simulations and machine learning approaches have been widely used to study RFB performance. The development of autonomous material discovery framework and digital twin of energy storage system usually needs to query cell performance through fast response models. In this study, two computationally efficient models are introduced: a physics-based analytical flow battery model (EZBattery), and a machine learning operator model (Deep Operator Network, denoted by DeepONet). Both models can provide cell performance near instantly, and prediction accuracy was systematically examined on an application of evaluating the performances of a 780 cm 2 aqueous organic redox flow battery (AORFB), using potential anolyte candidates in dihydroxyphenazine (DHP)-based family of organic materials. A validated computationally expansive 3-dimensional multi-physics finite element model by COMSOL was used as the ground truth and provided the training data set for the DeepONet. 1280 samples were generated with 10 properties to mimic the different possible anolyte candidates, and the cell performances were evaluated under 10 different combined operating conditions. The accuracy comparisons for the two computationally efficient models show that both models can provide comparable accuracy in predicting cell charging/discharging voltage curves. DeepONet can provide slightly higher overall accuracy than EZBattery with faster calculation speed, but highly relies on the training dataset. EZBattery does not need a training dataset and can provide interpretable physics-based explanations of the results, while being more flexible to adjust to adapt any different cell designs, flow battery architectures, and electrolyte materials.

Analytical model↗

Linear and Nonlinear Solvers for Simulating Multiphase Flow within Large-Scale Engineered Subsurface Systems

Simulation of multiphase flow in the subsurface is well-known to be computationally challenging. While there have been many studies that have explored approaches to overcoming these challenges, they often utilize relatively simple case studies. In this paper, we focus on the unique numerical challenges posed by modeling large-scale engineered subsurface systems, characterized by discrete features embedded in a heterogeneous natural subsurface setting. The man-made features such as shafts, tunnels, and barriers often cause multiple challenges in modeling the domain for multiphase porous media flow. This flow scenario can have a wide range of applications such as nuclear waste repositories, enhanced recovery of a petroleum reservoir, geothermal engineering, and carbon sequestration. An example of these severe numerical challenges is the case of performance assessment (PA) for Waste Isolation Pilot Plant (WIPP), the only operating deep geological repository in the US, which simulates extreme material properties of bedded salt rock formation and extreme contrast due to open excavation next to the formation. The models have extremes not only of permeability and porosity but also of the constitutive models needed for multiphase flow; additionally, they have process models like salt creep closure reducing porosity over time, fracturing in clay and anhydrite interbeds of the bedded salt, gas generation from the waste materials, and unintentional human borehole intrusions in some scenarios. Numerical simulations require the solution of coupled systems of nonlinear PDEs; in our work, we use the open-source simulator PFLOTRAN which is based on Finite Volume discretization. The solution of the nonlinear equations requires use of the Newton-Raphson iteration at each time step, which entails the solution of the linearized Jacobian system at each iteration. The effects of all the processes (i.e., large number of unknowns, highly nonlinear constitutive relations, large contrasts in material properties in short distances) lead to an ill-conditioned Jacobian matrix that severely challenges traditional linear solver, i.e., stabilized biconjugate gradient with block Jacobi incomplete LU preconditioner (BCGS-ILU) leading to non-convergence for traditional Newton-Raphson nonlinear solver causing unacceptably long computation time for each model. This paper presents linear solvers such as constrained pressure residual (CPR) two-stage preconditioner with alternate-block-factorization (ABF) and quasi- implicit pressure and explicit saturation (QIMPES) decouplers and flexible generalized residual solver (FGMRES). The new general-purpose nonlinear solver, Newton trust-region dogleg Cauchy (NTRDC), is also introduced to resolve extreme nonlinearities in the models. We demonstrate the effectiveness of each method relative to the default BCGS-Newton solver. The two best cases had nearly 50 times speed-up and achieved completion of a simulation in 14 hours that never completed due to non-convergence with the default solver. We also investigate the strong scalability of each method and discuss some of the deficiencies found for Block Jacobi preconditioner using parallel domain decomposition, and node packing effects of modern processor architecture.

Preconditioner, Nonlinear, Porous media, Multiphas↗

Bond-centric modular design of protein assemblies

Directional interactions that generate regular coordination geometries are a powerful means of guiding molecular and colloidal self-assembly, but implementing such high-level interactions with proteins remains challenging due to their complex shapes and intricate interface properties. Here we describe a modular approach to protein nanomaterial design inspired by the rich chemical diversity that can be generated from the small number of atomic valencies. We design protein building blocks using deep learning-based generative tools, incorporating regular coordination geometries and tailorable bonding interactions that enable the assembly of diverse closed and open architectures guided by simple geometric principles. Experimental characterization confirms the successful formation of more than 20 multicomponent polyhedral protein cages, two-dimensional arrays and three-dimensional protein lattices, with a high (10%–50%) success rate and electron microscopy data closely matching the corresponding design models. Due to modularity, individual building blocks can assemble with different partners to generate distinct regular assemblies, resulting in an economy of parts and enabling the construction of reconfigurable networks for designer nanomaterials.

Biomaterials – proteins↗

Dual-Channel Densenet for Hyperspectral Image Classification

Deep neural networks provide deep extracted features for image classification. As a high dimension data, hyperspectral image (HSI) feature extraction is unlike an RGB image whose feature representation could not be simply generated in the spatial domain. To take full advantage of HSI, a dualchannel convolutional neural network (CNN) is applied, 1D convolution for the spectral domain and 2D convolution for spatial domain. For pixel-wise classification of HSI, in our network model, one-dimensional customized DenseNet is for extracting the hierarchical spectral features and another customized DenseNet is applied to extract the hierarchical spatial-related feature. Furthermore, we experimentally tuned the several widen factors and dense-net growth rates to evaluate the impact of hyper-parameter. Furthermore, to compare our proposed method with HSI classification methods, we test other three DNNs based method in two real-world HSI dataset. The result demonstrated our approach outperformed the state-of-art method

97 MATHEMATICS AND COMPUTING↗

Diverse Super-Resolution (diversity_SR) [SWR-21-60]

Deep learning tools for enhancing the spatial resolution of wind data. The software is developed in Python using the TensorFlow deep learning package. Models for diversity super-resolution is provided. Included in the package are pretrained models with example code/data to perform the super-resolution as well as tools of training models for different enhancement- or data-types. The super-resolution is an inherently ill-conditioned problem, with multiple high-resolution fields plausibly mapping to the same coarse field. Considitional GANs provide a framework for generating a distribution of high-resolution realizations from a given low-resolution input. Stochastic estimation is used to inform the network of the expected degree and location of sub-grid diversity. The package includes a pretrained network to generate distributions of 10x-enhanced fields of wind data.

Glaws, Andrew↗

Deep Learning Based Superconducting Radio-Frequency Cavity Fault Classification at Jefferson Laboratory

This work investigates the efficacy of deep learning (DL) for classifying C100 superconducting radio-frequency (SRF) cavity faults in the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab. CEBAF is a large, high-power continuous wave recirculating linac that utilizes 418 SRF cavities to accelerate electrons up to 12 GeV. Recent upgrades to CEBAF include installation of 11 new cryomodules (88 cavities) equipped with a low-level RF system that records RF time-series data from each cavity at the onset of an RF failure. Typically, subject matter experts (SME) analyze this data to determine the fault type and identify the cavity of origin. This information is subsequently utilized to identify failure trends and to implement corrective measures on the offending cavity. Manual inspection of large-scale, time-series data, generated by frequent system failures is tedious and time consuming, and thereby motivates the use of machine learning (ML) to automate the task. This study extends work on a previously developed system based on traditional ML methods (Tennant and Carpenter and Powers and Shabalina Solopova and Vidyaratne and Iftekharuddin, Phys. Rev. Accel. Beams, 2020, 23, 114601), and investigates the effectiveness of deep learning approaches. The transition to a DL model is driven by the goal of developing a system with sufficiently fast inference that it could be used to predict a fault event and take actionable information before the onset (on the order of a few hundred milliseconds). Because features are learned, rather than explicitly computed, DL offers a potential advantage over traditional ML. Specifically, two seminal DL architecture types are explored: deep recurrent neural networks (RNN) and deep convolutional neural networks (CNN). We provide a detailed analysis on the performance of individual models using an RF waveform dataset built from past operational runs of CEBAF. In particular, the performance of RNN models incorporating long short-term memory (LSTM) are analyzed along with the CNN performance. Furthermore, comparing these DL models with a state-of-the-art fault ML model shows that DL architectures obtain similar performance for cavity identification, do not perform quite as well for fault classification, but provide an advantage in inference speed.

97 MATHEMATICS AND COMPUTING↗

Super-Resolution for Renewable Energy Resource Data with Wind from Reanalysis Data and Application to Ukraine

With a potentially increasing share of the electricity grid relying on wind to provide generating capacity and energy, there is an expanding global need for historically accurate, spatiotemporally continuous, high-resolution wind data. Conventional downscaling methods for generating these data based on numerical weather prediction have a high computational burden and require extensive tuning for historical accuracy. In this work, we present a novel deep learning-based spatiotemporal downscaling method using generative adversarial networks (GANs) for generating historically accurate high-resolution wind resource data from the European Centre for Medium-Range Weather Forecasting Reanalysis version 5 data (ERA5). In contrast to previous approaches, which used coarsened high-resolution data as low-resolution training data, we use true low-resolution simulation outputs. We show that by training a GAN model with ERA5 as the low-resolution input and Wind Integration National Dataset Toolkit (WTK) data as the high-resolution target, we achieved results comparable in historical accuracy and spatiotemporal variability to conventional dynamical downscaling. This GAN-based downscaling method additionally reduces computational costs over dynamical downscaling by two orders of magnitude. We applied this approach to downscale 30 km, hourly ERA5 data to 2 km, 5 min wind data for January 2000 through December 2023 at multiple hub heights over Ukraine, Moldova, and part of Romania. With WTK coverage limited to North America from 2007–2013, this is a significant spatiotemporal generalization. The geographic extent centered on Ukraine was motivated by stakeholders and energy-planning needs to rebuild the Ukrainian power grid in a decentralized manner. This 24-year data record is the first member of the super-resolution for renewable energy resource data with wind from the reanalysis data dataset (Sup3rWind).

17 WIND ENERGY↗

Techno-economic implications and cost of forecasting errors in solar PV power production using optimized deep learning models

Accurate solar Photovoltaic (PV) power forecasting is important for enhancing both the performance and economic feasibility of PV systems. This study evaluates several deep learning models, including Dense Neural Networks (DNN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and a hybrid LSTMCNN model, for predicting PV power production one day in advance. Prior to optimization, the models exhibited relatively high errors, with the best model (DNN) achieving a Root Mean Square Error (RMSE) of 31.13 kW and a coefficient of determination (R 2 ) of 62.15 %. After employing Bayesian optimization, the LSTM-CNN model demonstrated the best performance, with the RMSE reduced to 9.79 kW and R 2 improved to 97.62 %, showcasing significant enhancement in predictive accuracy. Here, the economic evaluation considered three cases: rewards for underestimation (0.08 USD/kWh), no rewards, and penalties for both over-and underestimation (120 % of the utility tariff). In the rewards scenario, the LSTM-CNN model reduced the Levelized Cost of Electricity (LCOE) by 4 %, while in the penalty scenario, a backup diesel generator would have increased the LCOE by 49 %. Additionally, the LSTM-CNN model minimized financial losses, achieving the lowest penalties and maximizing net cash flow compared to other models, demonstrating its overall technical and economic superiority.

Deep learning↗

Formation and annihilation of E4 centers in ZnO: Influence of hydrogen

Hydrothermally grown n-type ZnO bulk samples have been implanted with protons and deuterium ions to fluences in the range of 8 × 10{sup 10} to 8 × 10{sup 11 }cm{sup −2}. The implantations were performed at the temperature of 285 K, and the samples were then analyzed in-situ by deep level transient spectroscopy (DLTS) using a setup connected to the implanter beam line. The concentration of the so-called E4 center, having an apparent energy level at ∼0.57 eV below the conduction edge, is found to increase linearly with the ion fluence and the generation rate is proportional to the elastic energy deposition, as expected for a primary defect. Isothermal annealing of the E4 center at temperatures between 290 and 315 K reveals first-order kinetics with the activation energy of ∼0.6 eV. The annealing rate is strongly enhanced with increasing hydrogen fluence, and a model invoking migration of interstitial hydrogen and subsequent reaction with E4 exhibits close agreement with the experimental data. The rate of electron capture by E4 during the DLTS filling pulse depends on temperature, and it displays a thermal barrier of ∼0.15 eV. Most of these experimental results for E4 conform to the theoretically predicted properties of the oxygen vacancy (V{sub O}) and a tentative assignment of E4 to V{sub O} is made, corroborating previous suggestions in the literature. In particular, the 0.57 eV level is ascribed to the double donor state of V{sub O}.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Deep reinforcement learning for dynamic control of fuel injection timing in multi-pulse compression ignition engines

Conventional compression-ignition (CI) engines have long offered high thermal efficiencies and torque across a wide range of loads, but often require extensive exhaust gas treatment that decreases efficiency to meet ever-increasing emissions regulations. One strategy to decrease emissions is to split the fuel injection into a series of smaller injections. In this paper, we explore a new way of discovering optimal control strategies for the next generation of CI engines using deep reinforcement learning (DRL). We outline a DRL procedure to maximize the weighted reward of engine work while minimizing end-of-cycle NO x emissions. Through the procedure outlined in this paper, we show that the DRL agent is able to reduce NO x emissions threefold while only decreasing network by 2%. We demonstrate the use of transfer learning (TL) across hierarchies of physical models to accelerate the learning process, making this approach feasible for a range of control problems within this space. This paper presents a framework and demonstration for using DRL to design control systems in technology areas such as multi-pulse engine control where a hierarchy of models combined with multi-objective rewards are used for optimal operation.

33 ADVANCED PROPULSION SYSTEMS↗