SEARCH · Engineering Papers

Results for “Batch processing”

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.

SourceResearch connectionFrom yearThrough year

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

At least 955 records · Page 53

All sources NASA NTRS DOE OSTI

DOE OSTI2016

Power Plant Model Validation Tool

The PPMV is used to validate generator model using disturbance recordings. The PPMV tool contains a collection of power plant models and model validation studies, as well as disturbance recordings from a number of historic grid events. The user can import data from a new disturbance into the database, which converts PMU and SCADA data into GE PSLF format, and then run the tool to validate (or invalidate) the model for a specific power plant against its actual performance. The PNNL PPMV tool enables the automation of the process of power plant model validation using disturbance recordings. The tool uses PMU and SCADA measurements as input information. The tool automatically adjusts all required EPCL scripts and interacts with GE PSLF in the batch mode. The main tool features includes: The tool interacts with GE PSLF; The tool uses GE PSLF Play-In Function for generator model validation; Database of projects (model validation studies); Database of the historic events; Database of the power plant; The tool has advanced visualization capabilities; and the tool automatically generates reports.

Source record↗

DOE OSTI2026

Component-Level Inverse Design of Transmon Qubits Using Neural Networks

Designing a superconducting qubit to realize specific Hamiltonian parameters typically requires iterating through a time and compute-intensive forward loop in which the designer chooses a layout geometry, simulates it, extracts circuit parameters such as capacitances, and refines the geometry. We study the inverse version of this task using a neural-network workflow that maps target Hamiltonian parameters directly to component-level layout parameters, which we subsequently demonstrate on a planar transmon layout. During training, we pair the inverse model with a frozen forward surrogate model and evaluate the loss in Hamiltonian space rather than in layout-parameter space. In validation against a conventional EM solver, 97% of generated designs produce usable geometries, and the inverse-plus-surrogate pipeline reaches mean percent errors of 0.73% for qubit frequency and 1.58% for anharmonicity, comparable to or below the fabrication and simulation-to-measurement uncertainty expected for academic-process transmon devices of this type. A single pipeline query takes ~60 ms on CPU, versus ~2 min for a conventional EM capacitance extraction on the same hardware, a speedup of approximately 2,000x. Batching minimizes the AI model inference overhead, reducing the runtime to 3.1 microseconds per sample on CPU and 2.6 microseconds per sample on GPU at a batch size of 2048, resulting in speedups of 3.9 x 10^7 and 4.6 x 10^7, respectively, relative to a single conventional CPU EM extraction. Our results indicate that component-level inverse design usefully extends and complements conventional EM simulation, including for small datasets on the order of 1,000 samples.

Seidel, Olivia [Fermilab; Texas U., Arlington]↗

DOE OSTI2019

Improvement of Sugarcane Stillage (Vinasse) Anaerobic Digestion with Cheese Whey as its Co-substrate: Achieving High Methane Productivity and Yield

This study investigated methane production in an anaerobic sequencing batch biofilm reactor (AnSBBR) by co-digesting sugarcane vinasse and cheese whey. The assessment was based on the influence of feed strategy, interaction between cycle time and influent concentration, applied volumetric organic load (OLR{sub A}), and temperature over system stability and performance. The system showed flexibility with regard to the feed strategy, but the reduction of cycle time and influent concentration, at the same OLR{sub A}, resulted in lower methane productivity. Increasing organic load, up to the value of 15.27 gCOD L{sup −1} day{sup −1}, favored the process, increasing methane yield and productivity. Temperature reduction from 30 to 25 °C resulted in worse performance, although increasing it to 35 °C provided similar results to 30 °C. The best results were achieved at an OLR{sub A} of 15.27 gCOD L{sup −1} day{sup −1}, cycle time of 8 h, fed-batch operation, and temperature of 30 °C. The system achieved soluble COD removal efficiency of 89%, methane productivity of 208.5 molCH{sub 4} m{sup −3} day{sup −1} and yield of 15.76 mmolCH{sub 4} gCOD{sup −1}. The kinetic model fit indicated methanogenesis preference for the hydrogenotrophic route. At the industrial scale estimative, considering a scenario with a sugarcane ethanol plant with ethanol production of 150,896 m{sup 3} year{sup −1}, it was estimated energy production of 25,544 MWh month{sup −1}.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2025

Harvesting Energy from Wastewater by Converting Sewage

This project aims were to develop and demonstrate a scalable, integrated process to convert sewage sludge into renewable natural gas (RNG), enabling wastewater treatment plants (WWTPs) to become net energy producers. The system proposal integrates autothermal hydrothermal liquefaction (AT-HTL), supercritical salt precipitation (SCSP), and hydrothermal gasification (HTG), collectively forming the Supercritical Sludge-to-Gas (SC-S2G) platform. Initially, batch hydrothermal liquefaction reactions were used to screen sewage sludge using AT-HTL (later termed RI-HTL) conversion to biocrude, aqueous and char phases compared to hydrothermal liquefaction (HTL). Significant improvement in biocrude yield using peroxide addition at O:C ratio of 0.05 and under conditions of 300°C for 10 minutes gave 57% biocrude yield and 85% fluid carbon yield (biocrude plus aqueous), while minimizing the loss of carbon to char solids (~7%). Hence, RI-HTL was shown to be effective for conversion of real sewage sludge. The corrosion of the alloy reactor tubes or vessels is an important factor when developing a process that includes an oxidant and a chemically complex feed like sewage sludge. We investigated the corrosion rates on metal alloys at 350°C for 240 hours. Corrosion rates of 0.21 and 0.26 mpy for 304L and 316L stainless steel were measured respectively. The corrosion information obtained in this investigation was utilized by PNNL for design, materials sourcing and construction of the pilot scale continuous flow system.

09 BIOMASS FUELS↗

DOE OSTI2021

Simulated hydrological dynamics and coupled iron redox cycling impact methane production in an Arctic soil: Modeling Archive

This Modeling Archive is in support of an NGEE Arctic publication "Simulated hydrological dynamics and coupled iron redox cycling impact methane production in an Arctic soil" in the Journal of Geophysical Research-Biogeosciences. We simulated biogeochemical cycling in arctic soils using the PFLOTRAN geochemical model combined with measurements from previous NGEE Arctic incubations of polygonal permafrost soils in northern Alaska (Zheng et al., 2018). Simulated iron cycling, carbon dioxide production, and methane production were compared with incubation measurements and the parameterized model was then used to simulate coupled iron and carbon cycling over repeated oxic-anoxic cycles at different levels of carbon substrate availability and pH. The most recent data version (2.0) in the archive incorporates changes to the model and simulations as suggested by reviewers during the manuscript review process. These changes include an updated parameterization of the model; a new set of simulations omitting the iron cycle for direct evaluation of how iron cycle processes affect modeled outcomes; and a set of simulations testing different scenarios of carbon substrate availability in addition to scenarios of initial soil pH. This archive contains simulation code, model output, and analysis code for PFLOTRAN simulations. All scripts are python except the batch script for submitting multiprocessor jobs. Note that the model also requires compiled versions of the Alquimia interface and the NGEE Arctic fork of the PFLOTRAN geochemical simulator (see the README_INSTALL document for basic instructions). The Output directory contains eight data files in netCDF format generated by the model. The Next-Generation Ecosystem Experiments: Arctic (NGEE Arctic), was a 10-year research effort (2012-2022) to reduce uncertainty in Earth System Models by developing a predictive understanding of carbon-rich Arctic ecosystems and feedbacks to climate. NGEE Arctic was supported by the Department of Energy's Office of Biological and Environmental Research. The NGEE Arctic project had two field research sites: 1) located within the Arctic polygonal tundra coastal region on the Barrow Environmental Observatory (BEO) and the North Slope near Utqiagvik (Barrow), Alaska and 2) multiple areas on the discontinuous permafrost region of the Seward Peninsula north of Nome, Alaska. Through observations, experiments, and synthesis with existing datasets, NGEE Arctic provided an enhanced knowledge base for multi-scale modeling and contributed to improved process representation at global pan-Arctic scales within the Department of Energy's Earth system Model (the Energy Exascale Earth System Model, or E3SM), and specifically within the E3SM Land Model component (ELM).

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2012

DWPF STARTUP FRIT VISCOSITY MEASUREMENT ROUND ROBIN RESULTS

A viscosity standard is needed to replace the National Institute of Standards and Technology (NIST) glasses currently being used to calibrate viscosity measurement equipment. The current NIST glasses are either unavailable or less than ideal for calibrating equipment to measure the viscosity of high-level waste glasses. This report documents the results of a viscosity round robin study conducted on the Defense Waste Processing Facility (DWPF) startup frit. DWPF startup frit was selected because its viscosity-temperature relationship is similar to most DWPF and Hanford high-level waste glass compositions. The glass underwent grinding and blending to homogenize the large (100 lb) batch. Portions of the batch were supplied to the laboratories (named A through H) for viscosity measurements following a specified temperature schedule with a temperature range of 1150 C to 950 C and with an option to measure viscosity at lower temperatures if their equipment was capable of measuring at the higher viscosities. Results were used to fit the Vogel-Tamman-Fulcher and Arrhenius equations to viscosity as a function of temperature for the entire temperature range of 460 C through 1250 C as well as the limited temperature interval of approximately 950 C through 1250 C. The standard errors for confidence and prediction were determined for the fitted models.

36 MATERIALS SCIENCE↗

DOE OSTI2019

Retained Hydrogen Literature Study for DWPF

During Defense Waste Processing Facility (DWPF) operation in the Slurry Receipt and Adjustment Tank (SRAT) and the Slurry Mix Evaporator (SME), hydrogen is primarily produced by radiolysis and catalytic generation. The hydrogen is released from the sludge by agitation and removed from the vapor space by purging with air. Upon loss of agitation, the hydrogen begins accumulating in the sludge, and releases rapidly upon the resumption of agitation. If enough hydrogen accumulated in the sludge, and released rapidly upon the start of agitation, the hydrogen concentration in the vapor space could exceed the lower flammability limit (4 vol %). Savannah River Remediation (SRR) currently manages the risk for retained hydrogen in the DWPF vessels by implementing a Retained Hydrogen Program. The program primarily relies on the timely operation of vessel agitators to ensure that the retained hydrogen eruptions that could lead to flammable vapor spaces do not occur. The current program relies on conservative assumptions concerning gas release and retention (e.g., instantaneous release, conservative hydrogen bubble fraction, no release during quiescent periods) to develop allowable vessel quiescent times (Q-times). SRR would like to relax the conservative assumptions to the Retained Hydrogen Program and allow the Retained Hydrogen Program to extend past Sludge Batch 9. Relaxing the assumptions could potentially allow longer Q-times during agitator stoppages. SRR requested Savannah River National Laboratory (SRNL) personnel to perform a literature study to extract industry data pertaining to retained hydrogen and apply it to the DWPF process. The authors conducted a review of the technical literature, including work from the Savannah River Site (SRS), SRNL, the Hanford site, Pacific Northwest National Laboratory (PNNL), and the Sellafield site to identify literature describing the retention and release of gases with non-Newtonian slurries. The study was undertaken with the knowledge that SRS wastes are unique which makes using non-SRS data challenging. The literature search focused on the following topics: Mixing or blend time in Newtonian and non-Newtonian fluids, Quiescent release of gas from an unagitated non-Newtonian fluid, Release of gas from an agitated non-Newtonian fluid, and Changes in fluid rheology as a function of settling time.

08 HYDROGEN↗

DOE OSTI2021

CsPbI 3 Nanocrystals Go with the Flow: From Formation Mechanism to Continuous Nanomanufacturing

Despite the groundbreaking advancements in the synthesis of inorganic lead halide perovskite (LHP) nanocrystals (NCs), stimulated from their intriguing size-, composition-, and morphology-dependent optical and optoelectronic properties, their formation mechanism through the hot-injection (HI) synthetic route is not well-understood. Here in this work, for the first time, in-flow HI synthesis of cesium lead iodide (CsPbI 3 ) NCs is introduced and a comprehensive understanding of the interdependent competing reaction parameters controlling the NC morphology (nanocube vs nanoplatelet) and properties is provided. Utilizing the developed flow synthesis strategy, a change in the CsPbI 3 NC formation mechanism at temperatures higher than 150 °C, resulting in different CsPbI 3 morphologies is revealed. Through comparison of the flow- versus flask-based synthesis, deficiencies of batch reactors in reproducible and scalable synthesis of CsPbI 3 NCs with fast formation kinetics are demonstrated. The developed modular flow chemistry route provides a new frontier for high-temperature studies of solution-processed LHP NCs and enables their consistent and reliable continuous nanomanufacturing for next-generation energy technologies.

36 MATERIALS SCIENCE↗

DOE OSTI2024

Thermophilic biocatalysts for one-step conversion of citrus waste into lactic acid

Agri-food residues offer significant potential as a raw material for the production of L-lactic acid through microbial fermentation. Weizmannia coagulans, previously known as Bacillus coagulans, is a spore-forming, lactic acid-producing, gram-positive, with known probiotic and prebiotic properties. This study aimed to evaluate the feasibility of utilizing untreated citrus waste as a sustainable feedstock for the production of L-lactic acid in a one-step process, by using the strain W. coagulans MA-13. By employing a thermophilic enzymatic cocktail (Cellic CTec2) in conjunction with the hydrolytic capabilities of MA-13, biomass degradation was enhanced by up to 62%. Moreover, batch and fed-batch fermentation experiments demonstrated the complete fermentation of glucose into L-lactic acid, achieving a concentration of up to 44.8 g/L. These results point to MA-13 as a microbial cell factory for one-step production of L-lactic acid, by combining cost-effective saccharification with MA-13 fermentative performance, on agri-food wastes. Moreover, the potential of this approach for sustainable valorization of agricultural waste streams is successfully proven.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2023

Stochastic projective splitting

Here, we present a new, stochastic variant of the projective splitting (PS) family of algorithms for inclusion problems involving the sum of any finite number of maximal monotone operators. This new variant uses a stochastic oracle to evaluate one of the operators, which is assumed to be Lipschitz continuous, and (deterministic) resolvents to process the remaining operators. Our proposal is the first version of PS with such stochastic capabilities. We envision the primary application being machine learning (ML) problems, with the method’s stochastic features facilitating “mini-batch” sampling of datasets. Since it uses a monotone operator formulation, the method can handle not only Lipschitz-smooth loss minimization, but also min–max and noncooperative game formulations, with better convergence properties than the gradient descent-ascent methods commonly applied in such settings. The proposed method can handle any number of constraints and nonsmooth regularizers via projection and proximal operators. We prove almost-sure convergence of the iterates to a solution and a convergence rate result for the expected residual, and close with numerical experiments on a distributionally robust sparse logistic regression problem.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2013

Center for Renewable Energy Science and Technology

The CREST research team conducted research that optimized catalysts used for the conversion of southwestern lignite into synthetic crude oil that can be shipped to nearby Texas refineries and power plants for development of transportation fuels and power generation. Research was also undertaken to convert any potential by-products of this process such as CO2 to useful chemicals and gases which could be recycled and used as feedstock to the synthetic fuel process. These CO2 conversion processes used light energy to drive the endogonic reduction reactions involved. The project was divided into two tasks: A CO2 Conversion Task, and a Catalyst Optimization Task. The CO2 Conversion task was aimed at developing molecular and solid state catalysts for the thermal, electro- and photocatalytic reduction of CO2 to reduced products such as simple feedstock compounds (e.g. CO, H2, CHOOH, CH2O, CH3OH and CH4). For example, the research team recycled CO that was developed from this Task and used it as a feedstock for the production of synthetic crude in the Catalyst Optimization Task. In the Catalyst Optimization Task, the research team conducted bench-scale experiments with the goal of reducing overall catalyst cost in support of several synthetic crude processes that had earlier been developed. This was accomplished by increasing the catalyst reactivity thus reducing required concentrations or by using less expensive metals. In this task the team performed parametric experiments in small scale batch reactors in an effort to improve catalyst reactivity and to lower cost. They also investigated catalyst robustness by testing lignite feedstocks that vary in moisture, h, and volatile content.

Billo, Richard↗

DOE OSTI2019

Roll-to-Roll Integrated Substrates for OLED Lighting (Final Report)

OLED lighting has immense potential as aesthetically pleasing, energy-efficient general illumination. Unlike other light sources, such as incandescents, fluorescents, and inorganic LEDs, OLEDs naturally emit over a large-area surface. They are glare free, do not need to be shaded, and are cool to the touch, requiring no heatsink. The best efficiencies and lifetimes reported are on par with or better than current forms of illumination. However, the cost for OLED lighting remains high – so much so that these products are not market competitive and there is very low consumer demand. We believe that flexible, plastic-based devices will highlight the advantages of aesthetically-pleasing OLED lighting systems while paving the way for lowering both materials and manufacturing costs. These flexible devices require new development in substrate and support technology, which was the focus of the work reported here. Sinovia Technologies, as part of a collaborative team, has developed integrated plastic substrates to serve as supports for flexible OLED lighting. These substrates include a barrier to protect the OLED from moisture and oxygen-related degradation, a smooth, highly conductive transparent anode to define the OLED area, and a light scattering layer to improve emission efficiency. To ensure low cost, the anode pattern must be defined using high throughput, continuous processing, and any pattern shape should be possible. No current patterning technique meets all of these requirements. This SBIR work demonstrated the use of flexographic printing as an effective method of patterning for integrated substrates. Flexography is a letterpress process commonly used for many types of everyday items including boxes, retail bags, milk cartons, food packaging, disposable cups, adhesive tapes, envelopes, wallpaper, and newspaper. The printing process can go as fast as 600 meters per minute with feature resolution around 10 microns, showing promise for low-cost with high throughput. In this case, Sinovia has successfully used this technique to print transparent anode patterns roll-to-roll as part of our integrated substrate for OLED lighting. We have characterized the printed films and tailored the edges of the printed areas for OLED compatibility. We followed this by making white OLEDs using these anodes. To ensure that our anodes were compatible with different OLED stack structures, we fabricated solution-processed polymer white OLEDs, as well as collaborated with OLEDWorks to fabricate evaporated small-molecule devices. Comparing to results from our previous studies, we find that the flexography process does not have any detrimental effects to our overall substrate. We were able to show that our flexographically printed integrated substrate improves OLED efficiency by 24 – 35%, similar to previous results demonstrated using batch-patterning techniques. The success of this work represents two substantial opportunities for OLED lighting manufacturers. The first to first facilitate flexible lighting panels at scale, and the second to lower costs for by enabling roll-to-roll manufacture of the entire device stack. When realized, both of these will help OLED to become market competitive with other forms of general illumination.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

DOE OSTI2020

Recovery of Rare Earth Elements (REEs) from Coal Mine Drainage, Phase 2

Phase 1 research identified a two-step refining train, acid leaching of AMD precipitate feed followed by solvent extraction (SX), to produce a 2% TREE product. For the Phase 2 bench scale system, the objective was to get the acid leaching step to operate in batch-mode with sufficient storage capacity to feed an aqueous, enriched REE liquor to a pilot SX plant operating in continuous-mode at a feed rate capable of producing 3 g/hr 2% TREE. Important process variables included feed rate, extractant makeup, flowsheet configuration, and stripping conditions. Operating parameters were optimized through a feedback-capable SCADA system configured for the SX bench scale plant. These conditions were met and exceeded. TREE grades approaching 90% were produced while our Techno-Economic Assessment indicated highly favorable economic returns. The project led to the design of a pilot facility for producing roughly 500 kg/TREE/year. That plant is now in the procurement stage with constructed expected to begin in late 2020.

01 COAL, LIGNITE, AND PEAT↗

DOE OSTI2020

Recovery of Rare Earth Elements (REEs) from Coal Mine Drainage, Phase 2

01 COAL, LIGNITE, AND PEAT↗

DOE OSTI2023

Evaluation of a Sulfate Solubility Limit Greater Than 0.65 Weight Percent in Sludge Batch 10 Glasses

Previous laboratory-scale crucible testing with batch chemicals confirmed that the sulfate (SO 4 2- ) limit for Sludge Batch 10 (SB10) was 0.65 weight percent (wt.%) in glass. This limit signifies that 0.65 wt.% SO 4 2- can be retained in the glass without the formation of a sulfate phase. The Defense Waste Processing Facility imposes this constraint in the Material Tracking Program. Based on preliminary calculations to support the Material Tracking Program, it was anticipated that transfer volumes of the monosodium titanate/sludge solids (MST/SS) stream from the Salt Waste Processing Facility (SWPF) may need to be reduced to maintain projected sulfate concentrations below 0.65 wt.% in glass. Savannah River Mission Completion requested that the Savannah River National Laboratory perform additional sulfate testing to determine whether a sulfate solubility limit greater than 0.65 wt.% is feasible for SB10, which could allow for higher transfer volumes of the MST/SS stream. This report documents the results of the testing at higher sulfate concentrations for the glass composition region defined by the most recent SB10 projection (November 2022) and Frits 473 and 625. Frit 473 was recommended for SB10 and Frit 625 was used during SB9 processing and the SB9 to SB10 transition. A total of twenty-one glass compositions were developed based on the expected compositional variables, which include sludge-only (SO) and coupled processing with the SWPF, waste loading (WL), and frit composition. The target sulfate concentrations were varied from 0.65-0.85 wt.% at 32 and 40% WL. Each glass was prepared from reagent grade chemicals and melted at 1150 °C. Visual observations were used to confirm the presence of a sulfate salt phase on the cooled glass surfaces. Representative samples of each glass were submitted for chemical composition analysis by inductively coupled plasma-optical emission spectroscopy and Cs analysis by inductively coupled plasma-mass spectrometry. Overall the majority of mean measured values are consistent with the target values for each major oxide of interest with less than 5% error. The percent errors for the measured SO 4 2- concentrations are generally less than 10%, which is comparable with previous sulfate solubility study measurements and acceptable. Only the SO glasses based on Frit 625 formed a sulfate phase at a 0.80 wt.% SO 4 2- target concentration at both 32 and 40% WL. The remainder of the glasses did not form a sulfate layer. Due to the formation of the sulfate phase, the limit is conservatively set at 0.70 wt.% based on the measured sulfate concentrations of 0.71 wt.% and 0.75 wt.% for these two glasses. None of the SO or coupled operation glasses based on Frit 473 formed a sulfate salt phase, which supports a sulfate limit of 0.80 wt.%. The following SO 4 2- concentration limits are recommended during SO and coupled SB10 processing: (1) 0.70 wt.% during processing with Frit 625, and (2) 0.80 wt.% during processing with Frit 473.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

DOE OSTI2013

Use of photovoltaics for waste heat recovery

A device for recovering waste heat in the form of radiated light, e.g. red visible light and/or infrared light includes a housing having a viewing window, and a photovoltaic cell mounted in the housing in a relationship to the viewing window, wherein rays of radiated light pass through the viewing window and impinge on surface of the photovoltaic cell. The housing and/or the cell are cooled so that the device can be used with a furnace for an industrial process, e.g. mounting the device with a view of the interior of the heating chamber of a glass making furnace. In this manner, the rays of the radiated light generated during the melting of glass batch materials in the heating chamber pass through the viewing window and impinge on the surface of the photovoltaic cells to generate electric current which is passed onto an electric load.

14 SOLAR ENERGY↗

DOE OSTI2022

Recursive Gaussian Process over graphs for Integrating Multi-timescale Measurements in Low-Observable Distribution Systems

The transition to a smarter grid is empowered by enhanced sensor deployments and smart metering infrastructure in the distribution system. Measurements from these sensors and meters can be used for many applications, including distribution system state estimation (DSSE). However, these measurements are typically sampled at different rates and could be intermittent due to losses during the aggregation process. These multi timescale measurements should be reconciled in real-time to perform accurate grid monitoring. This paper tackles this problem by formulating a recursive multi-task Gaussian process (RGP-G) approach that sequentially aggregates sensor measurements. Specifically, we formulate a recursive multi-task GP with and without network connectivity information to reconcile the multi time-scale measurements in distribution systems. Here, the proposed framework is capable of aggregating the multi-time scale measurements batch-wise or in real-time. Following the aggregation of the multi time-scale measurements, the spatial states of the consistent time-series are estimated using matrix completion based DSSE approach. Simulation results on IEEE 37 and IEEE 123 bus test systems illustrate the efficiency of the proposed methods from the standpoint of both multi time-scale data aggregation and DSSE.

42 ENGINEERING↗

DOE OSTI2023

Bayesian sequential optimal experimental design for nonlinear models using policy gradient reinforcement learning

We present a mathematical framework and computational methods for optimally designing a finite sequence of experiments. This sequential optimal experimental design (sOED) problem is formulated as a finite-horizon partially observable Markov decision process (POMDP) under a Bayesian setting and with information-theoretic utilities. The formulation is general and may accommodate continuous random variables, non-Gaussian posteriors, and nonlinear forward models. The sOED design policy incorporates elements of feedback and lookahead simultaneously, and we show it to generalize the commonly-used batch and greedy design strategies. We solve for the sOED policy using the policy gradient (PG) method from reinforcement learning, and provide a derivation for the PG expression in the sOED context. Adopting an actor-critic approach, the policy and value functions are parameterized using deep neural networks and improved via PG estimates produced from simulated episodes of designs and observations. The new PG-sOED algorithm is first validated on a linear-Gaussian benchmark, and then compared against other design baselines on a sensor movement problem for contaminant source inversion in a convection-diffusion field. As a result, we provide explanation for the policy behaviors using knowledge of the underlying physical process.

97 MATHEMATICS AND COMPUTING↗