Below is a comprehensive list of articles, events, projects, references and research related content that is specific to the term described above. Use the filter to narrow the results further. To explore additional science and technology topics that Argonne researchers and engineers may be working on please visit our Research Index.
|
Invention Opportunity & Solution Benefits
Applications, Industries
|
A flexible user scripting interface for the definitions of robot-performed experimental procedures
Autonomous experiments require orchestrating robotic movements and functions of hardware modules to perform a sequence of experimental steps designed and specified by a user who is familiar with the type of experiments (peptide synthesis, nanostructure characterization, etc.), but not necessarily accustomed to robotics hardware or computer programming. Our workflow scripting interface utilizes the Python programming language to simplify the scripting of hardware control sequences, and organizes them into logical entries that represent experimental steps. The entire experimental procedure is encompassed within a single script, allowing easy archival. The design enables sophisticated backend processing of the workflow script to enable seamless integration with machine learning/optimization frameworks and the robotics system control. Since the experimental procedure is presented in a form that resembles the original hardware control sequences, a wide range of experimental workflows and hardware configurations can be programmed with ease due to the readability and flexibility of the Python scripting language.
A ML-based scheduler for concurrent sample processing in autonomous experiments
A scheduler is as a software component that orchestrates the execution of steps (e.g., robot arm movement, solution stirring, temperature annealing, etc.) for performing an autonomous experiment. It is challenging to build a scheduler that enables seamless integration of non-vendor specific hardware and can efficiently optimize the execution order of steps in an experiment involving multiple hardware modules and concurrent processing of samples. Our software comprises a new scheduler using the widely used Python programming language. Since it is built using a general language that is increasingly common in the robotics and scientific fields, one can easily couple it with script-based hardware APIs and utilize machine learning to provide automatic data-driven tuning and hardware-aware scheduling of experimental steps. The ease of integration with existing software codes enables faster progress to be made across laboratories and reduces the cost of autonomous systems thus making them available to less well-funded laboratories.
Data structure and data workflow for handling material samples in high-throughput experiments
High-throughput experiments enabled by robotics generate large volumes of heterogeneous data, coming from multiple hardware modules in different data formats. There needs to be an easy way to standardize data and store them in a universal format such that data and metadata of individual material samples are well organized. This software consists of a material sample data class/object that utilizes JSON file in the backend for data storage. The sample data class is implemented using the Python programming language. It couples the flexibility of JSON files with rigid schema to make data files compatible with database handling. There are also functions for smart creation of unique sample ID based on workflow and input parameters, as well as handling the underlying data writing and conversion. Instead of the traditional approach of specifying file paths within an experimental workflow, we developed an object/data-oriented approach that initiates actions from the sample data class, providing an elegant way of handling the file path related issues in concurrent/parallel processing of material samples.
Invention:
Multi-fidelity scale bridging between various flavors of molecular dynamics (i.e. ab-initio, classical and coarse-grained models) has remained a long-standing challenge. BLAST (Bridging Length/time scales via Atomistic Simulation Toolkit) is a framework that leverages machine learning principles to address this challenge.
Opportunity and Solution
BLAST provides users with the capabilities to train and develop their own classical atomistic and coarse-grained interatomic potentials (i.e., force fields) for molecular simulations. BLAST is designed to address several long-standing problems in the molecular simulation community, such as unintended misuse of existing force fields due to a knowledge gap between developers and users, bottlenecks in traditional force field development approaches, and other issues relating to the accuracy, efficiency, and transferability of force fields. The BLAST architecture consists of a web user-friendly interface, front-end and back-end web services, and machine learning algorithms that run on high-performance computing (HPC) clusters.
Cybersecurity issues are a day-to-day struggle for businesses and organizations. Keeping information secure can be a herculean task. SMORE-MTD, developed by Argonne’s Joshua Lyle and Nate Evans with laboratory funding, defends against cybersecurity attacks by using software-defined networking to manipulate network paths that service user requests.
By randomly selecting which server and service will respond to a given user’s request, SMORE-MTD makes it more difficult for an attacker to identify which services to attack. SMORE-MTD also increases organizations’ resilience by preventing an attacker exploit from being routed to the vulnerable software, forcing attackers into repeated attacks that are more likely to be noticed. SMORE-MTD also eliminates the need to install and maintain configuration software on each host in rotation, which reduces complexity and increases the amount of software available for use.
SVTRIP (Stochastic Vehicle Trip Prediction) generates a naturalistic vehicle speed profile (vehicle speed as a function of time at 1 Hz or more) for a given trip or route. The trip provided as an input is defined by the attributes of its sub-segments, such as travel time, distance and speed limit. The inputs can be provided directly by the user, extracted from digital maps, or generated from macroscopic or mesoscopic traffic flow simulators. The generated speed profiles can be used to predict vehicle energy consumption and operations for trips with low-resolution information. The algorithm used for generation relies on Markov Chains, making the generated speed profiles stochastic.
SPARTA is a mobile supported venue assessment tool created to assess both cyber and physical vulnerabilities in stadiums and arenas.
Features:
Technical Requirements:
ChemNODE: A Chemical Kinetics Solver Framework Based on Neural Ordinary Differential Equations (ANL-SF-20-154)
Accurate and fast chemical kinetic models are key to the development of cleaner and more efficient combustion engines with reduced emissions and enhanced efficiency. Argonne’s software technology, ChemNODE, is a data-driven framework for learning reduced, yet accurate, chemical kinetic representations from high-fidelity data. ChemNODE works by replacing the chemical source terms derived from the law of mass action with artificial neural networks trained to correctly calculate the source terms. As opposed to conventional approaches that minimize the errors in the predicted source terms, ChemNODE uses neural ordinary differential equations (NODEs), combining the power of machine learning and numerical methods, to directly minimize the loss based on the chemical species’ profiles. ChemNODE employs forward-mode automatic differentiation and the Levenberg-Marquardt algorithm to adjust the neural network parameters, such that the discrepancies between the actual and predicted species mass fractions at different points in time are minimized.
As a proof-of-concept, the accuracy and efficiency of ChemNODE was demonstrated for hydrogen-air combustion in a homogenous reactor at various initial temperatures and equivalence ratios. For more information on ChemNODE: ChemNODE: A Neural Ordinary Differential Equations Approach for Chemical Kinetics Solvers
Technical Contacts:
Ope Owoyele, Postdoctoral Researcher, Transportation and Power Systems Division
(oowoyele@anl.gov; 630-252-2132)
Pinaki Pal, Research Scientist, Transportation and Power Systems Division
(pal@anl.gov; 630-252-1361)
CIERA: Cavitation-Induced Erosion Risk Assessment Model (ANL-SF-19-118)
Argonne has developed an approach for modeling cavitation-induced erosion for use with any multiphase computational fluid dynamics (CFD) code (e.g. CONVERGE, ANSYS Fluent, etc.). Argonne’s CIERA model provides a framework for linking multiphase flow predictions with the response of the solid material. This link is represented via an energy balance at the fluid-solid interface, which considers the cumulative energy absorbed by the solid material from repeated hydrodynamic impacts. In contrast to existing methods, Argonne’s CIERA model provides more reliable prediction of erosion severity by considering both impact load and duration.
Application of CIERA to simulations of multiphase flow systems can guide design studies in improving system durability and reducing maintenance costs. To date, CIERA’s erosion prediction capabilities have been demonstrated in simulations of pressurized fuel through aluminum channels and a heavy-duty fuel injector, and validated against available experimental data. For more information on the CIERA model: Evaluation of a new cavitation erosion metric based on fluid-solid energy transfer in channel flow simulations; Linking cavitation collapse energy with the erosion incubation period
Technical Contact:
Gina Magnotti, Research Scientist, Transportation and Power Systems Division
(gmagnotti@anl.gov; 630-252-8554)
LESI: Lagrangian-Eulerian Spark Ignition Model (ANL-SF-18-030)
Argonne has developed an approach for spark-ignition modeling of complex engine conditions, for use within industry CFD solver packages, such as Convergent Science’s CONVERGE framework. Argonne’s LESI model allows for enhanced accuracy in spark-ignition modelling of internal combustion engines and extends current capabilities to more challenging real-world conditions. This is an important upgrade for the automotive industry, as spark-ignition engine technologies move toward unconventional boosted and dilute operation that impact a wider range of performance factors, such as flame propagation, cycle-to-cycle variation (CCV), and spark-plug durability. Additionally, compression ignition strategies are also increasingly reliant on ignition systems to control combustion behavior. Predictive models coupled with high-performance computing (HPC) can evaluate advanced combustion concepts and accelerate the development of high-efficiency engines.
Argonne’s LESI model leverages previous findings that have expanded the use and improved the accuracy of Eulerian-type energy deposition models. The Eulerian energy deposition is coupled at any computational time-step with a Lagrangian-type evolution of the spark channel. Typical features such as spark channel elongation, stretch, and attachment to the electrodes are properly described to deliver realistic energy deposition along the channel during the entire ignition process. This is a decisive factor to accurately describe ignition processes in a highly-dilute and highly-turbulent environment.
Please visit the LESI Model web page for more information.
Journal article: Development of a Hybrid Lagrangian-Eulerian Model to Describe Spark-Ignition Processes at Engine-Like Turbulent Flow Conditions
Technical Contact:
Riccardo Scarcelli, Research Scientist, Transportation and Power Systems
(rscarcelli@anl.gov; 630-252-6940)
ML-GA: Machine-Learning Genetic Algorithm (ANL-SF-18-098; ANL-SF-19-073)
Argonne’s ML-GA software provides a unique capability for rapid design optimization by combining machine learning (ML) and genetic algorithm (GA) techniques. It employs ML (either one or multiple ML algorithms can be incorporated) to predict the quality (merit) of a design from the input parameters. Then, a stochastic global optimization genetic algorithm (GA) is used with the machine learning model as the objective function to optimize the input parameters based on the merit function. ML-GA is scalable to high-performance computing platforms such as supercomputers, enabling optimization to be performed in significantly short time frames (of the order of a few days).
As a proof-of-concept, the potential of the ML-GA approach coupled with computational fluid dynamics (CFD) was demonstrated for optimization of a heavy-duty internal combustion engine operating under medium load conditions. For more information on this application of ML-GA: A Machine Learning-Genetic Algorithm (ML-GA) Approach for Rapid Optimization Using High-Performance Computing
Technical Contact:
Pinaki Pal, Research Scientist, Transportation and Power Systems
(pal@anl.gov; 630-252-1361)
PPM4CCV: Parallel Perturbation Model for Cycle-to-Cycle Variability (ANL-SF-17-030)
PPM4CCV is a pre-processing approach to modelling cyclic variability in spark ignition (SI) engines that can be coupled with any major engine CFD platform (e.g., CONVERGE CFD, AVL-Fire or STAR-CD). The parallel perturbation method overcomes several challenges associated with predicting cyclic variability, resulting in up to a 10x speed-up in computation times compared to conventional approaches. By implementing PPM4CCV, engine developers can not only free up computational resources, but also accelerate the engine design process.
Cycle-to-cycle variability (CCV) is known to be detrimental to SI engine operation resulting in partial burn and knock, and an overall reduction in the reliability of the engine. Modelling CCV in SI engines is challenging because computationally intensive high-fidelity methods are required; and CCV is experienced over long timescales requiring simulations to be performed over hundreds of consecutive cycles. The PPM4CCV approach is to perform multiple parallel simulations, each of which encompasses multiple cycles, by perturbing simulation parameters such as the initial and boundary conditions. More information: Parallel Methodology to Capture Cyclic Variability in Motored Engines
Technical Contact:
Muhsin Ameen, Research Scientist, Transportation and Power Systems
(mameen@anl.gov; 630-252-5784)
Business & Licensing Contact (for all software listed above):
Eric Tyo, Business Development Executive, Science & Technology Partnerships and Outreach
(etyo@anl.gov; 630-252-4924)
TF-MoE: Tabulated Flamelet – Mixture of Experts (ANL-SF-19-174)
Argonne has developed a deep learning driven approach for modeling turbulent combustion that provides a framework for incorporating high-dimensional datasets in computational fluid dynamics (CFD) simulations in a tractable and efficient manner, with 2-5 times speed-up over traditional methods. This open source software enhances predictive modeling via machine learning. Argonne’s TF-MoE software employs a mixture of experts (MoE) approach to bifurcate high-dimensional tabulated flamelet (TF) data into simpler manifolds in a physically intuitive manner. It employs a divide-and-conquer competitive approach, where different zones in the manifold are assigned to various neural networks for inference. It consists of two classes of neural networks, namely, a gating network which is a neural network classifier, and a number of experts which are neural network regressors. The software bifurcates the manifold by having different neural networks compete for each input signal. The gating network rewards the best predictors with stronger signals during subsequent training episodes and feeds poor-performing networks with weaker signals. The gating network and experts are trained using a standard backpropagation approach.
As a proof-of-concept, the accuracy and efficiency of TF-MoE was demonstrated in an a priori study of high-dimensional tabulation for modeling of non-premixed combustion using the flamelet approach. For more information on TF-MoE: Efficient bifurcation and parameterization of multi-dimensional combustion manifolds using deep mixture of experts: an a priori study
TF-MoE is available through GitHub: https://github.com/owoyeleope/TFM-MoE
Technical Contacts:
Ope Owoyele, Postdoctoral Researcher, Transportation and Power Systems Division
(oowoyele@anl.gov; 630-252-2132)
Pinaki Pal, Research Scientist, Transportation and Power Systems Division
(pal@anl.gov; 630-252-1361)