Ingenieurwissenschaften und zugeordnete Tätigkeiten
Filtern
Dokumenttyp
- Vortrag (27)
- Forschungsdatensatz (17)
- Posterpräsentation (6)
- Zeitschriftenartikel (2)
- Beitrag zu einem Tagungsband (2)
Referierte Publikation
- nein (54) (entfernen)
Schlagworte
- Automation (54) (entfernen)
Organisationseinheit der BAM
- 6 Materialchemie (45)
- 6.0 Abteilungsleitung und andere (39)
- VP Vizepräsident (12)
- VP.1 eScience (12)
- 6.5 Synthese und Streuverfahren nanostrukturierter Materialien (5)
- 1 Analytische Chemie; Referenzmaterialien (3)
- 1.4 Prozessanalytik (2)
- 5 Werkstofftechnik (2)
- 8 Zerstörungsfreie Prüfung (2)
- 8.6 Faseroptische Sensorik (2)
In recent years, many protocols in computational materials science have been automated and made available within software packages (primarily Python-based). This ranges from the automation of simple heuristics (oxidation states, coordination environments) to the automation of protocols, including multiple DFT and post-processing tools such as (an)harmonic phonon computations or bonding analysis. Such developments also shorten the time frames of projects after such developments have been made available and open new possibilities. For example, we can now easily make data-driven tests of well-known rules and heuristics or develop quantum chemistry-based materials descriptors for machine learning approaches. These tests and descriptors can have applications related to magnetic ground state predictions of materials relevant for spintronic applications or for predicting thermal properties relevant for thermal management in electronics. Combining high-throughput ab initio computations with fitting, fine-tuning machine learning models and predictions of such models within complex workflows is also possible and promises further acceleration in the field. In this talk, I will show our latest efforts to link automation with data-driven chemistry and materials science.
Bonds and local atomic environments are crucial descriptors of material properties. They have been used to create design rules and heuristics and as features in machine learning of materials properties. Implementations and algorithms (e.g., ChemEnv and LobsterEnv) for identifying local atomic environments based on geometrical characteristics and quantum-chemical bonding analysis are nowadays available. Fully automatic workflows and analysis tools have been developed to use quantum-chemical bonding analysis on a large scale. The lecture will demonstrate how our tools, that assess local atomic environments and perform automatic bonding analysis, help to develop new machine learning models and a new intuitive understanding of materials. Furthermore, the general trend toward automation in density functional-based materials science and some of our recent contributions will be discussed.
By automatically recording as much information as possible in automated laboratory setups, reproducibility and traceability of experiments are vastly improved. This presentation shows what such an approach means for the quality of experiments in an X-ray scattering laboratory and an automated synthesis set-up.
The LOBSTER (Deringer et al., 2011;Maintz et al., 2013 ,2016 ;Nelson et al., 2020 ) software aids in extracting quantum-chemical bonding information from materials by projecting the plane-wave based wave functions from density functional theory (DFT) onto an atomic orbital basis. LobsterEnv, a module implemented in pymatgen (Ong et al., 2013) by some of the authors of this package, facilitates the use of quantum-chemical bonding information obtained from LOBSTER calculations to identify neighbors and coordination environments. LobsterPy is a Python package that offers a set of convenient tools to further analyze and summarize the LobsterEnv outputs in the form of JSONs that are easy to interpret and process. These tools enable the estimation of (anti) bonding contributions, generation of textual descriptions, and visualization of LOBSTER computation results. Since its first release, both LobsterPy and LobsterEnv capabilities have been extended significantly. Unlike earlier versions, which could only automatically analyze Crystal Orbital Hamilton Populations (COHPs) (Dronskowski & Blöchl, 1993), both can now also analyze Crystal Orbital Overlap Populations (COOP) (Hughbanks & Hoffmann, 1983) and Crystal Orbital Bond Index (COBI) (Müller et al., 2021). Extracting the information about the most important orbitals contributing to the bonds is optional, and users can enable it as needed. Additionally, bonding-based features for machinelearning (ML) studies can be engineered via the sub-packages “featurize” and “structuregraphs”. Alongside its Python interface, it also provides an easy-to-use command line interface (CLI) that runs automatic analysis of the computations and generates a summary of results and publication-ready figures. LobsterPy has been used to produce the results in Ngo et al. (2023), Chen et al. (2024), Naik et al. (2023), and it is also part of Atomate2 (2023) bonding analysis workflow for generating bonding analysis data in a format compatible with the Materials Project (Jain et al., 2013) API.
Bonds and local atomic environments are crucial descriptors of material properties. They have been used to create design rules and heuristics and as features in machine learning of materials properties. Implementations and algorithms (e.g., ChemEnv and LobsterEnv) for identifying local atomic environments based on geometrical characteristics and quantum-chemical bonding analysis are nowadays available. Fully automatic workflows and analysis tools have been developed to use quantum-chemical bonding analysis on a large scale. The lecture will demonstrate how our tools, that assess local atomic environments and perform automatic bonding analysis, help to develop new machine learning models and a new intuitive understanding of materials.[5,6] Furthermore, the general trend toward automation in density functional-based materials science and some of our recent contributions will be discussed.
Jobflow is a free, open-source library for writing and executing workflows. Complex workflows can be defined using simple python functions and executed locally or on arbitrary computing resources using the FireWorks workflow manager.
Some features that distinguish jobflow are dynamic workflows, easy compositing and connecting of workflows, and the ability to store workflow outputs across multiple databases.
We present Jobflow, a domain-agnostic Python package for writing computational workflows tailored for high-throughput computing applications. With its simple decorator-based approach, functions and class methods can be transformed into compute jobs that can be stitched together into complex workflows. Jobflow fully supports dynamic workflows where the full acyclic graph of compute jobs is not known until runtime, such as compute jobs that launch
other jobs based on the results of previous steps in the workflow. The results of all Jobflow compute jobs can be easily stored in a variety of filesystem- and cloud-based databases without the data storage process being part of the underlying workflow logic itself. Jobflow has been intentionally designed to be fully independent of the choice of workflow manager used to dispatch the calculations on remote computing resources. At the time of writing, Jobflow
workflows can be executed either locally or across distributed compute environments via an adapter to the FireWorks package, and Jobflow fully supports the integration of additional workflow execution adapters in the future.
Bonds and local atomic environments are crucial descriptors of material properties. They have been used to create design rules and heuristics for materials. More and more frequently, they are used as features in machine learning. Implementations and algorithms (e.g., ChemEnv and LobsterEnv) for identifying these local atomic environments based on geometrical characteristics and quantum-chemical bonding analysis are nowadays available. Fully automatic workflows and analysis tools have been developed to use quantum-chemical bonding analysis on a large scale and for machine-learning approaches. The latter relates to a general trend toward automation in density functional-based materials science. The lecture will demonstrate how our tools, that assess local atomic environments, helped to test and develop heuristics and design rules and an intuitive understanding of materials.
Bonds and local atomic environments are crucial descriptors of material properties. They have been used to create design rules and heuristics and as features in machine learning of materials properties. Implementations and algorithms (e.g., ChemEnv and LobsterEnv) for identifying local atomic environments based on geometrical characteristics and quantum-chemical bonding analysis are nowadays available. Fully automatic workflows and analysis tools have been developed to use quantum-chemical bonding analysis on a large scale. The lecture will demonstrate how our tools, that assess local atomic environments and perform automatic bonding analysis, help to develop new machine learning models and a new intuitive understanding of materials. Furthermore, other recent workflow contributions to the Materials Project software infrastructure (pymatgen, atomate2) related to phonons and machine-learning potentials will be discussed.
Many microstructural features exhibit non-trivial geometries, which can only be derived to a limited extent from two-dimensional images. E.g., graphite arrangements in lamellar gray cast iron have complex geometries, and the same is true for additively manufactured materials and three-dimensional conductive path structures. Some can be visualized using tomographic methods, but some cannot be due to weak contrast and/or lack of resolution when analyzing macroscopic objects. Classic metallography can help but must be expanded to the third dimension. The method of reconstructing three-dimensional structures from serial metallographic sections surely is not new. However, the effort required to manually assemble many individual sections into image stacks is very high and stands in the way of frequent application. For this reason, an automated, robot-supported 3D metallography system is being developed at BAM, which carries out the steps of repeated preparation and image acquisition on polished specimen.
Preparation includes grinding, polishing and optionally etching of the polished surface. Image acquisition comprises autofocused light microscopic imaging at several magnification levels. The image stacks obtained are then pre-processed, segmented, and converted into 3D models, which in the result appear like microtomographic models, but with high resolution at large volume. Contrasting by classical chemical etching reveals structures that cannot be resolved using tomographic methods. The integration of further imaging and measuring methods into this system is underway. Some examples will be discussed in the presentation.
The PMD Core Ontology (PMDco) is a comprehensive set of building blocks produced via consensus building. The ontological building blocks provide a framework representing knowledge about fundamental concepts used in Materials Science and Engineering (MSE) today. The PMDco is a mid-level ontology that establishes connections between narrower MSE application ontologies and domain neutral concepts used in already established broader (top-level) ontologies. The primary goal of the PMDco design is to enable interoperability between various other MSE-related ontologies and other common ontologies.
PMDco’s class structure is both comprehensive and extensible, rendering it an efficient tool to structure MSE knowledge. The PMDco serves as a semantic middle-layer unifying common MSE concepts via semantic mapping to other semantic representations using well-known key terms used in the MSE domain. The PMDco enables straight-forward documentation and tracking of science data generation and in consequence enables high-quality FAIR data that allows for precise reproducibility of scientific experiments.
The design of PMDco is based on the W3C Provenance Ontology (PROV-O), which provides a standard framework for capturing the production, derivation, and attribution of resources. Via this foundation, the PMDco enables the integration of data from various data origins and the representation of complex workflows.
In summary, the PMDco is a valuable advancement for researchers and practitioners in MSE domains. It provides a common MSE vocabulary to represent and share knowledge, allowing for efficient collaboration and promoting interoperability between diverse domains. Its design allows for the systematic integration of data and metadata, enabling seamless tracing of science data. Overall, the PMDco is a crucial step towards a unified and comprehensive understanding of the MSE domain in general.
Understanding and Machine Learning of Materials Properties with Quantum-Chemical Bonding Analysis
(2023)
Bonds and local atomic environments are crucial descriptors for material properties. They have been used to create design rules for materials and are used as features in machine learning of material properties. This talk will show how our recently developed tools, that automatically perform quantum chemical bond analysis and enable the study of chemical bonds and local atomic environments, accelerate and improve the development of such heuristics and machine-learned models for materials properties.
Understanding the chemistry and nature of individual chemical bonds is essential for materials design. Bonding analysis via the LOBSTER software package has provided valuable insights into the properties of materials for thermoelectric and catalysis applications. Thus, the data generated from
bonding analysis becomes an invaluable asset that could be utilized as features in large-scale data analysis and machine learning of material properties. However, no systematic studies exist that conducted high-throughput materials simulations to curate and validate bonding data obtained from LOBSTER. Here we present an approach to constructing such a large database consisting of quantum-chemical bonding information.
Bonds and local atomic environments are crucial descriptors of material properties. They have been used to create design rules and heuristics for materials. More and more frequently, they are used as features in machine learning. Implementations and algorithms (e.g., ChemEnv and LobsterEnv) for identifying these local atomic environments based on geometrical characteristics and quantum-chemical bonding analysis are nowadays available. Fully automatic workflows and analysis tools have been developed to use quantum-chemical bonding analysis on a large scale and for machine-learning approaches. The latter relates to a general trend toward automation in density functional-based materials science. The lecture will demonstrate how our tools, that assess local atomic environments, helped to test and develop heuristics and design rules and an intuitive understanding of materials.
While the synthesis of Metal-Organic Framework (MOF) particles can be as easy as adding two solutions together, reproducibly obtaining the same particles, time and time again, is a lot harder. As laboratory-independent reproducibility is a cornerstone of the scientific method, we must put effort into finding and controlling all necessary parameters to achieve this.
An open-source Python/EPICS-controlled robotic platform (see picture) was adapted to systematically explore this for a 20 ml MOF synthesis of the Zeolitic Imidazole Framework-8 (ZIF-8) chemistry in methanol. Parameters that were explored included: 1) addition sequence, 2) addition speeds, 3) reaction times, 4) source chemicals, 5) stirring speeds, 6) stirring bar choice, 7) starting concentrations, and 8) workup methodologies. It was found that, by controlling these parameters, highly reproducible syntheses are obtained. Secondly, the variation of these parameters alone led to a dramatic difference in volume-weighted particle size means, which exceeds an order of magnitude as investigated by our in-house X-ray scattering instrument [1].
The syntheses are thoroughly documented in an automated fashion, and the synthesis libraries as well as analyses libraries will become available in batches soon. With this library, it will be possible to extract previously unknown correlations, and other laboratories can produce specific particles by following the exact procedures of the particles of their choice.
This database consists of bonding data computed using Lobster for 1520 solid-state compounds consisting of insulators and semiconductors. The files are named as per ID numbers in the materials project database.
Here we provide the larger computational data JSON files for the rest of the 820 compounds. This file consists of all important LOBSTER computation output files data stored as a dictionary.
This database consists of bonding data computed using Lobster for 1520 solid-state compounds consisting of insulators and semiconductors. It consists of two kinds of json files. Smaller lightweight JSONS consists of summarized bonding information for each of the compounds. The files are named as per ID numbers in the materials project database.
Here we provide also the larger computational data json files for 700 compounds. This files consists of all important LOBSTER computation output files data stored as dictionary.
Talk about my recent research on data-driven chemical understanding with geometrical and quantum-chemical bonding analysis.
A deep insight into the chemistry and nature of individual chemical bonds is essential for understanding materials. Bonding analysis is expected to provide important features for large-scale data analysis and machine learning of material properties. Such information on chemical bonds can be calculated using the LOBSTER (www.cohp.de) software package, which post-processes data from modern density functional theory computations by projecting plane wave-based wave functions onto a local atomic orbital basis. We have performed bonding analysis on 1520 compounds (insulators and semiconductors) using a fully automated workflow combining the VASP and LOBSTER software packages. We then automatically evaluated the data with LobsterPy (https://github.com/jageo/lobsterpy) and provide results as a database. The projected densities of states and bonding indicators are benchmarked on VASP projections and available heuristics, respectively. Lastly, we illustrate the predictive power of bonding descriptors by constructing a machine-learning model for phononic properties, which shows an increase in prediction accuracies by 27 % (mean absolute errors) compared to a benchmark model differing only by not relying on any quantum-chemical bonding features.
Chemical bonding and coordination environments are crucial descriptors of material properties. They have previously been applied to creating chemical design guidelines and chemical heuristics. They are currently being used as features in machine learning more and more frequently. I will discuss implementations and algorithms (ChemEnv and LobsterEnv) for identifying these coordination environments based on geometrical characteristics and chemical bond quantum chemical analysis. I will demonstrate how these techniques helped in testing chemical heuristics like the Pauling rule and thereby improved our understanding of chemistry. I will also show how these tools can be used to create new design guidelines and a new understanding of chemistry. To use quantum-chemical bonding analysis on a large-scale and for machine-learning approaches, fully automatic workflows and analysis tools have been developed. After presenting the capabilities of these tools, I will also point out how these developments relate to the general trend towards automation in the field of density functional based materials science.
McSAS3
(2023)
McSAS3 is a refactored version of the original McSAS (see DOI 10.1107/S1600576715007347). This software fits scattering patterns to obtain size distributions without assumptions on the size distribution form. The refactored version has some neat features:
- Multiprocessing is included, spread out over as many cores as number of repetitions!
- Full state of the optimization is stored in an organized HDF5 state file.
- Histogramming is separate from optimization and a result can be re-histogrammed as many times as desired.
- SasModels allow a wide range of models to be used
- If SasModels does not work (e.g. because of gcc compiler issues on Windows or Mac), an internal sphere model is supplied
- Simulated data of the scattering of a special shape can also be used as a McSAS fitting model. Your models are infinite!
- 2D fitting also works.
Chemical bonding and coordination environments are crucial descriptors of material properties. They have previously been applied to creating chemical design guidelines and chemical heuristics. They are currently being used as features in machine learning more and more frequently. I will discuss implementations and algorithms (ChemEnv and LobsterEnv) for identifying these coordination environments based on geometrical characteristics and chemical bond quantum chemical analysis. I'll demonstrate how these techniques helped in testing chemical heuristics like the Pauling rule and thereby improved our understanding of chemistry. I'll also show how these tools can be used to create new design guidelines and a new understanding of chemistry. To use quantum-chemical bonding analysis on a large-scale and for machine-learning approaches, fully automatic workflows and analysis tools have been developed. After presenting the capabilities of these tools, I will also point out how these developments relate to the general trend towards automation in the field of density functional based materials science.
Measuring an X-ray scattering pattern is relatively easy, but measuring a steady stream of high-quality, useful patterns requires significant effort and good laboratory organization.
Such laboratory organization can help address the reproducibility crisis in science, and easily multiply the scientific output of a laboratory, while greatly elevating the quality of the measurements. We have demonstrated this for small- and wide-angle X-ray scattering in the MOUSE project (Methodology Optimization for Ultrafine Structure Exploration).
With the MOUSE, we have combined a comprehensive and highly automated laboratory workflow with a heavily modified X-ray scattering instrument. This combination allows us to collect fully traceable scattering data, within a well-documented, FAIR-compliant data flow (akin to what is found at the more automated synchrotron beamlines). With two full-time researchers, our lab collects and interprets thousands of datasets, on hundreds of samples, for dozens of projects per year, supporting many users along the entire process from sample selection and preparation, to the analysis of the resulting data.
This talk will briefly introduce the foundations of X-ray scattering, present the MOUSE project, and will highlight the proven utility of the methodology for materials science. Upgrades to the methodology will also be discussed, as well as possible avenues for transferring this holistic methodology to other instruments
We created a workflow that fully automates bonding analysis using Crystal Orbital Hamilton Populations, which are bond-weighted densities of states. This enables understanding of crystalline material properties based on chemical bonding information. To facilitate data analysis and machine-learning research, our tools include automatic plots, automated text output, and output in machine-readable format.
High-throughput computations are nowadays an established way to suggest new candidate materials for applications to experimentalists. Due to new packages for automation and access to databases of computed materials properties, these studies became more and more complex over the last years. Besides suggesting new candidate materials for applications, they also offer a way to understanding the materials properties based on chemical bonds. For example, we have recently used orbital-based bonding analysis to understand the results of high-throughput studies for spintronic materials, ferroelectric materials and photovoltaic materials in detail. To do so, we have developed Python tools for high-throughput bonding analysis with the programs VASP and Lobster (see www.cohp.de). They are based on the Python packages pymatgen, atomate, and custodian. This implementation will be discussed within the talk. We also expect that these tools offer possibilities to arrive at new descriptors based on chemical bonds for materials properties.
The talk „Automation in computational materials science“ deals with the current state of automation in the field of computational materials science. It illustrates how automation can, for example, be used to speed up the search for new ferroelectric materials and spintronic materials. Furthermore, it lists current tools for automation and challenges in the field.
The Meticulous Approach: Fully traceable X-ray scattering data via a comprehensive lab methodology
(2021)
To find out if experimental findings are real, you need to be able to repeat them. For a long time, however, papers and datasets could not necessarily include sufficient details to accurately repeat experiments, leading to a reproducibility crisis. It is here, that the MOUSE project (Methodology Optimization for Ultrafine Structure Exploration) tries to implement change – at least for small- and wide-angle X-ray scattering (SAXS/WAXS).
In the MOUSE project, we have combined: a) a comprehensive laboratory workflow with b) a heavily modified, highly automated Xenocs Xeuss 2.0 instrumental component. This combination allows us to collect fully traceable scattering data, with a well-documented data flow (akin to what is found at the more automated beamlines). With two full-time researchers, the lab collects and interprets thousands of datasets, on hundreds of samples for dozens of projects per year, supporting many users along the entire process from sample selection and preparation, to the analysis of the resulting data.
While these numbers do not light a candle to those achieved by our hardworking compatriots at the synchrotron beamlines, the laboratory approach does allow us to continually modify and fine-tune the integral methodology. So for the last three years, we have incorporated e.g. FAIR principles, traceability, automated processing, data curation strategies, as well as a host of good scattering practices into the MOUSE system. We have concomitantly expanded our purview as specialists to include an increased responsibility for the entire scattering aspect of the resultant publications, to ensure full exploitation of the data quality, whilst avoiding common pitfalls.
This talk will discuss the MOUSE project1 as implemented to date, and will introduce foreseeable upgrades and changes. These upgrades include better pre-experiment sample scattering predictions to filter projects on the basis of their suitability, exploitation of the measurement database for detecting long-term changes and automated flagging of datasets, and enhancing MC fitting with sample scattering simulations for better matching of odd-shaped scatterers.
Data set of low-field NMR spectra of continuous synthesis of nitro-4’-methyldiphenylamine (MNDPA). 1H spectra (43 MHz) were recorded as single scans.
Two different approaches for the generation of artificial neural networks training data for the prediction of reactant concentrations were used: (i) Training data based on combinations of measured pure component spectra and (ii) Training data based on a spectral model.
Synthetic low-field NMR spectra
First 4 columns in MAT-files represent component areas of each reactant within the synthetic mixture spectrum.
Xi (“pure component spectra dataset”)
Xii (“spectral model dataset”)
Experimental low-field NMR spectra from MNDPA-Synthesis
This data set represents low-field NMR-spectra recorded during continuous synthesis of nitro-4’-methyldiphenylamine (MNDPA). Reference values from high-field NMR results are included.
The presentation describes a novel approach to dynamically adjusting the weaving motion of the electrode in narrow gap GMAW.
An event driven arc sensor is used to dynamically adjust the weaving angle to variations in gap width by detecting each groove sidewall independently and in real-time. The approach presented requires only minimal user configuration for spray-arc or pulsed-arc transfer modes and can effectively be used in double- and single-sided weaving applications. Furthermore displacements of the welding torch with regards to the groove center line or contact-tip to workpiece distance are compensated.