SEARCH · Engineering Papers
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.
This 20-minute presentation highlights uncertainty and technical risk reduction capabilities in CCSI2, especially intrusive methods: model-based design of experiment and robust optimization.
Power converters are pervasive in modern electronic component design. They can be found in all electronic devices from household appliances and cellphone chargers to vehicles. Currently, designing new circuit topologies is hard because it requires human expertise based on experience and is difficult to automate. However, artificial-intelligence-assisted design can significantly facilitate the development of new power converters and/or improve the final result. Intelligently designed highly efficient power converters can have a significant effect on many important attributes, such as power efficiency, layout size, cost, heat dissemination, energy requirements, etc. We propose Autonomous Topology Generator (AutoTG), a reinforcement-learning-based framework that generates power converter topology candidates based on user specifications, optimized for user preferences. By modeling power converter design as a symbolic optimization problem, we sequentially sample components in an autoregressive manner until new topologies are formed, providing both the topology specification and the sizing (magnitude of each component parameter) of the proposed power converter. Here, we provide an empirical evaluation and show that AutoTG is able to generate varied high-efficiency topologies within component restrictions based on user input and show that previously unknown topologies can be found for further evaluation.
With decreasing computational costs, improvement in algorithms, and the aggregation of large industrial and commercial datasets, machine learning is becoming a ubiquitous tool for process and business innovations. Machine learning is still lacking applications in the field of dynamic optimization for real-time control. This work presents a novel framework for performing constrained dynamic optimization using a recurrent neural network model combined with a metaheuristic optimizer. The framework is designed to augment an existing control system and is purely data-driven, like most industrial Model Predictive Control applications. Several recurrent neural network models are compared as well as several metaheuristic optimizers. Hyperparameters and optimizer parameters are tuned with parameter sweeps, and the resulting values are reported. Further, the best parameters for each optimizer and model combination are demonstrated in closed-loop control of a dynamic simulation, and several recommendations are made for generalizing this framework to other systems. Up to 0.953% improvement is realized over the non-optimized case for a simulated coal-fired boiler. While this is not a large improvement in percentage, the total economic impact is $991,000 per year, and this study builds a foundation for future machine learning with dynamic optimization.
Focal Area 3: Explainable AI Our confidence in the projections made by Earth System Models (ESMs) depends on understanding them to be, in some important respects, faithful representations of the Earth system. Here we present an “explainable Artificial Intelligence (AI)” method that allows us to uncover the dynamical structure of the observed and modeled Earth system, discover hidden links across wide spatiotemporal scales, target model development efforts at poorly-represented dynamics, and optimize observed or modeled data collection to maximize predictive information. Science Challenge: Dynamical system science for the Earth system poses unique challenges given the large degree of internal climate variability. Thus, tools that help us understand how ESMs succeed and fail at representing these dynamics are crucial, particularly in relation to the observed system. Furthermore, the computational and memory constraints on ESM data output motivate in situ analysis of ESM dynamics, including automatic detection of dynamical shifts. Also, of key importance are procedures that leverage ESMs to optimize observational campaigns for improving process representation, reducing structural uncertainty and improving model skill.
Background: Drug discovery is a multi-stage process that comprises two costly major steps: pre-clinical research and clinical trials. Among its stages, lead optimization easily consumes more than half of the pre-clinical budget. We propose a combined machine learning and molecular modeling approach that partially automates lead optimization workflow in silico, providing suggestions for modification hot spots. Results: The initial data collection is achieved with physics-based molecular dynamics simulation. Contact matrices are calculated as the preliminary features extracted from the simulations. To take advantage of the temporal information from the simulations, we enhanced contact matrices data with temporal dynamism representation, which are then modeled with unsupervised convolutional variational autoencoder (CVAE). Finally, conventional and CVAE-based clustering methods are compared with metrics to rank the submolecular structures and propose potential candidates for lead optimization. Conclusion: With no need for extensive structure-activity data, our method provides new hints for drug modification hotspots which can be used to improve drug potency and reduce the lead optimization time. It can potentially become a valuable tool for medicinal chemists.
Abstract Data-driven design shows the promise of accelerating materials discovery but is challenging due to the prohibitive cost of searching the vast design space of chemistry, structure, and synthesis methods. Bayesian optimization (BO) employs uncertainty-aware machine learning models to select promising designs to evaluate, hence reducing the cost. However, BO with mixed numerical and categorical variables, which is of particular interest in materials design, has not been well studied. In this work, we survey frequentist and Bayesian approaches to uncertainty quantification of machine learning with mixed variables. We then conduct a systematic comparative study of their performances in BO using a popular representative model from each group, the random forest-based Lolo model (frequentist) and the latent variable Gaussian process model (Bayesian). We examine the efficacy of the two models in the optimization of mathematical functions, as well as properties of structural and functional materials, where we observe performance differences as related to problem dimensionality and complexity. By investigating the machine learning models’ predictive and uncertainty estimation capabilities, we provide interpretations of the observed performance differences. Our results provide practical guidance on choosing between frequentist and Bayesian uncertainty-aware machine learning models for mixed-variable BO in materials design.
This technical report summarized the contribution of the DADAIST project funded by the Data Model Convergence Initiative via the Laboratory Directed Research and Development (LDRD) investments at Pacific Northwest National Laboratory (PNNL). Specifically, we report the development of the NeuroMANCER (Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations), a new open-source Scientific Machine Learning library for formulating and solving parametric constrained optimization problems, physics-informed system identification, and parametric optimal control problems. NeuroMANCER is using differentiable programming to combine modern data-driven models and optimization modeling language into a coherent algorithmic and software framework. NeuroMANCER is a Pytorch-based framework and adopts much of its philosophy focused on research and development, rapid prototyping, and streamlined deployment. Strong emphasis is given to extensibility, interoperability with the PyTorch ecosystem, and quick adaptability to custom domain problems. Neuromancer repository contains a comprehensive library of differentiable modules, including custom activation functions, matrix factorizations, deep learning architectures, neural differential equations, differential equation solvers, implicit layers such as iterative solvers, high-level API for symbolic expressions, API for modeling and control of dynamical systems, and extensive set of tutorial code examples in the form of python scripts and jupyter notebooks.
CMIP model explores various charging infrastructure network designs to serve a free-floating car-sharing fleet and determine the charging downtime experienced by the fleet for each design. Development of the CMIP model had two major steps: (1) describing modeling assumptions and (2) developing an integer program (IP) that jointly optimizes decisions about locations to install DC fast chargers and EV-to-charger assignments. The CMIP model integrates an EV charging model, EV energy consumption model, and heterogeneous, real-world vehicle use data with an integer programming optimization model to identify optimal location of new charging stations and calculate vehicle downtime for charging. The CMIP model can be applied to understand: (a) the reduction of EV fleet downtime if an additional fast-charging station is added to the current infrastructure and (b) to what extent total vehicle downtime would be sensitive to additional charging infrastructure.
Critical minerals and rare earth elements play an important role in our climate change initiatives, particularly in applications related with energy storage. Here, we use discrete optimization approaches to design a process for the recovery of Lithium and Cobalt from battery recycling, through membrane separation. Our contribution involves proposing a Generalized Disjunctive Programming (GDP) model for the optimal design of a multistage diafiltration cascade for Li-Co separation. By solving the resulting nonconvex mixed-integer nonlinear program model to global optimality, we investigated scalability and solution quality variations with changes in the number of stages and elements per stage. Results demonstrate the computational tractability of the nonlinear GDP formulation for design of membrane separation processes while opening the door for decom-position strategies for multicomponent separation cascades. Future work aims to extend the GDP formulation to account for stage installation and explore various decomposition techniques to enhance solution efficiency.
Recent studies have shown that fossil generators equipped with post-combustion carbon capture (PCC) systems are needed to reduce the cost of deep decarbonization. Such generators need to be flexible and responsive to grid conditions, particularly in a high variable renewable energy (VRE) environment. In this work, we evaluate the net present value (NPV) of retrofitting an existing natural gas combined cycle (NGCC) unit with a flexible PCC system while incorporating market signals from a high VRE grid. We use our industrial partner’s NGCC configuration as representative of existing NGCC units and Svante’s rapid-temperature swing adsorption (TSA) for PCC. Because of its ability to rapidly startup/shutdown and ramp-up/ramp-down, the chosen capture technology is very attractive for load-following operations. For a given set of market signals, we formulate a two-stage stochastic multi-period optimization problem, under the price-taker assumption, to simultaneously optimize the design of the capture system and operation of the entire plant. Rigorous models for the NGCC unit, PCC system, and compression system are developed using commercial process simulators and validated with either plant or vendor data. For computational tractability, we develop surrogate/reduced-order models for use in the optimization problem. The surrogate model for the NGCC plant is constructed by linearizing the rigorous dynamic model at 75% load, while data-driven nonlinear surrogate models for the capture and compression systems are constructed using simulation data from the rigorous models. The optimization problem, formulated as a mixed integer bilinear program, is implemented in the IDAES® integrated platform and solved to global optimality using Gurobi 9.5. Using this formulation, we determine the profitability of retrofitting an existing NGCC unit with the chosen capture system for multiple regions in the U.S. under two scenarios with different carbon prices. Importantly, the results show that the optimal decision strongly depends on the region and on the carbon price, thereby demonstrating the importance of the inclusion of market signals in the design process.
River sediment microbial respiration is a key indicator of ecosystem functioning and the biogeochemical fluxes across this critical zone link surface and subsurface waters. As such, there is tremendous interest in measuring and mapping these respiration rates. Respiration observations are expensive and labor intensive; there is limited data available to the community. An open science, collaborative initiative is collecting samples for respiration rate analysis and multi-scale metadata; this evolving data set is being used for making machine learning (ML) predictions at unsampled sites to help inform continued community engagement. However, it is a challenge to find an optimum configuration for ML models to work with this feature-rich (i.e., 100+ possible input variables) data set. Here, we present results from a two-tiered approach to managing the analysis of this complex data set: (a) a stacked ensemble of models that automatically optimizes hyperparameters and manages the training of many models and (b) feature permutation importance to detect the most important features in the models. The major elements of this workflow are modular, portable, open, and cloud-based thus making this implementation a potential template for other applications. The models developed here predict that sediment organic matter chemistry is one of the most important features for predicting sediment respiration rate. Other larger-scale, important features fall into the categories of climatic, ecological, geological, and fluvial settings. Leveraging these larger-scale features to generate data-driven estimates of river sediment respiration rates reveals spatially consistent but heterogeneous patterns across the river network of the Columbia River Basin.
The purpose of this project is to examine the energy impact of urban-scale traffic for the Los Angeles Basin by developing and implementing a scalable traffic assignment model. An energy optimization function will be posed and when integrated into the optimization code for travel assignment it can be mathematically proven to converge. The energy optimization function can then be compared to the typical travel time optimization that is traditionally used in traffic assignment models. The analysis will begin with static traffic assignment models with the routing for all origin and destinations computed in parallel on high performance computing facilities. Convergence of the numerical methods rely on the solution of convex programs (or extensions of these). This step will mostly consist of demonstrating the ability to parallelize the Frank Wolfe algorithm on various platforms.
The purpose of this project is to examine the energy impact of urban-scale traffic for the Los Angeles Basin by developing and implementing a scalable traffic assignment model. An energy optimization function will be posed and when integrated into the optimization code for travel assignment it can be mathematically proven to converge. The energy optimization function can then be compared to the typical travel time optimization that is traditionally used in traffic assignment models. The analysis will begin with static traffic assignment models with the routing for all origin and destinations computed in parallel on high performance computing facilities. Convergence of the numerical methods rely on the solution of convex programs (or extensions of these). This step will mostly consist of demonstrating the ability to parallelize the Frank Wolfe algorithm on various platforms. This work will contribute to LBNL’s efforts to develop new processes, analytical tools, program designs, and business models to advance the state of the art in next-generation sustainable transportation solutions.
To address the rising electricity demand and greenhouse gas concentration in the environment, considerable effort is being carried out across the globe on installing and operating renewable energy sources. However, the renewable energy production is affected by diurnal and seasonal variability. To ensure that the electric grid remains reliable and resilient even for the high penetration of renewables into the grid, various types of energy storage systems are being investigated. In this paper, a compressed-air energy storage (CAES) system integrated with a natural gas combined-cycle (NGCC) power plant is investigated where air is extracted from the gas turbine compressor or injected back into the gas turbine combustor when it is optimal to do so. First-principles dynamic models of the NGCC plant and CAES are developed along with the development of an economic model. The dynamic optimization of the integrated system is undertaken in the Python/Pyomo platform for maximizing the net present value (NPV). NPV optimization is undertaken for 14 regions/cases considering year-long locational marginal price (LMP) data with a 1 h interval. Design variables such as the storage capacity and storage pressure, as well as the operating variables such as the power plant load, air injection rate, and air extraction rate, are optimized. Results show that the integrated CAES system has a higher NPV than the NGCC-only system for all 14 regions, thus indicating the potential deployment of the integrated system under the assumption of the availability of caverns in close proximity to the NGCC plant. The levelized cost of storage is found to be in the range of 136–145 $/MWh. Roundtrip efficiency is found to be between 74.6–82.5%. A sensitivity study with respect to LMP shows that the LMP profile has a significant impact on the extent of air injection/extraction while capital expenditure reduction has a negligible effect.
This work presents a detailed case study on using Generative AI (GenAI) to develop AI surrogates for simulation models in fusion energy research. The scope includes the methodology, implementation, and results of using GenAI to assist in model development and optimization, comparing these results with previous manually developed models.
Wavelength-dependent mid-infrared (400–1,800 cm -1 ; 5.56–25 μm) optical constants (real and imaginary indices of refraction; n and k) are determined using reflectance spectra at a spectral sampling of 4 cm -1 for several silicate glasses of varying SiO 2 wt% which include (a) basaltic volcanic glass from Kilauea, Hawaii, (b) synthetic andesite, (c) two synthetic dacites, (d) obsidian volcanic glass from Mount Lassen, California, (e) synthetic rhyodacite, and (f) rhyolitic volcanic glass from Mexico. Because glasses are optically isotropic, no specific orientation was required for spectral measurements, and polished glass samples were measured at random orientations using micro-FTIR spectrometer. Lorentz-Lorenz dispersion theory and Fresnel reflectance model for high symmetry materials were used to model the optical constants by optimizing oscillator parameters in modeled spectra to match laboratory spectra. A genetic algorithm (GA) approach automatically finds the natural oscillators and their parameters for each spectrum, and then uses these parameters in a non-linear least squares optimization routine. The study compared spectral parameters such as Christiansen feature, reststrahlen bands (RBs), and the peaks centered around 860–1,100 (peak 1) and 400–480 cm -1 (peak 2) of n and k to their respective SiO 2 wt% of the glasses. CF, RBs, peak 1 of n and k, and peak 2 of n shift to higher wavenumbers with increased SiO 2 wt%, whereas peak 2 of k shifts to lower wavenumbers with increased SiO 2 wt%. Derived optical constants of these glasses will improve quantitative abundance mapping of volcanic materials on the surfaces of silicate targets in the Solar System.
The computer code uses a parallel simulated annealing framework with embedded machine learning components to solve multi-constrained optimization problems. The software automatically balances the execution of low and high fidelity physics models within the optimization procedure. The low fidelity model is used to rapidly explore the design space while the high fidelity physics model is executed sparingly to account for complex design constraints that are not resolved by the quickly executing low fidelity model.
The Hybrid Optimization Software Suite (HOSS) modeling framework and capabilities are described.