Ingenieurwissenschaften und zugeordnete Tätigkeiten
Filtern
Dokumenttyp
- Vortrag (11)
- Zeitschriftenartikel (4)
- Posterpräsentation (3)
Sprache
- Englisch (18) (entfernen)
Schlagworte
- Machine learning (18) (entfernen)
Organisationseinheit der BAM
- 6 Materialchemie (12)
- 6.0 Abteilungsleitung und andere (9)
- 8 Zerstörungsfreie Prüfung (3)
- 1 Analytische Chemie; Referenzmaterialien (2)
- 1.6 Anorganische Referenzmaterialien (2)
- 6.1 Oberflächen- und Dünnschichtanalyse (2)
- 8.0 Abteilungsleitung und andere (2)
- 8.5 Röntgenbildgebung (2)
- 5 Werkstofftechnik (1)
- 5.2 Metallische Hochtemperaturwerkstoffe (1)
Eingeladener Vortrag
- nein (11)
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.
Knowing phonon properties is beneficial for predicting low thermal conductivity thermoelectric materials. Employing DFT consumes lots of computational resources. Using ML-driven interatomic potentials (MLIP, e.g., GAP) opens up a faster route, but most potentials are specifically tailored to a certain compound. We aim to generalize the MLIP generation in a Python code-based workflow, combining automatic DFT runs with automated GAP fits. Automation enables easier tests, benchmarks, and validation.
Investigating the phononic properties is beneficial for predicting low thermal conductivity thermoelectric materials.1–3 Employing density functional theory4 takes many calculation steps and consumes a lot of computational resources.5,6 Using machine learning driven interatomic potentials (MLIP, e.g., Gaussian approximation potential,8 GAP) opens up a faster route to phonons7 but in most cases, the potentials are specifically tailored for a certain compound. In this work, we automate the generation of such MLIPs in a Python code-based workflow, based on the automation tools atomate29 and pymatgen10 which combines the automatic DFT computations with the automated fitting of GAPs. Automation enables easier testing, benchmarking and validation.11 We aim to provide the workflow-generated potentials for storage in databases.
The prediction of porosity is a crucial task for metal based additive manufacturing techniques such as laser powder bed fusion. Short wave infrared thermography as an in-situ monitoring tool enables the measurement of the surface radiosity during the laser exposure. Based on the thermogram data, the thermal history of the component can be reconstructed which is closely related to the resulting mechanical properties and to the formation of porosity in the part. In this study, we present a novel framework for the local prediction of porosity based on extracted features from thermogram data. The framework consists of a data pre-processing workflow and a supervised deep learning classifier architecture. The data pre-processing
workflow generates samples from thermogram feature data by including feature information from multiple subsequent layers.
Thereby, the prediction of the occurrence of complex process phenomena such as keyhole pores is enabled. A custom convolutional neural network model is used for classification. Themodel is trained and tested on a dataset from thermographic in-situ monitoring of the manufacturing of an AISI 316L stainless steel test component. The impact of the pre-processing parameters and the local void distribution on the classification performance is studied in detail. The presented model achieves an accuracy of 0.96 and an f1-Score of 0.86 for predicting keyhole porosity in small sub-volumes with a dimension of (700 × 700 × 50) μm3. Furthermore, we show that pre-processing parameters such as the porosity threshold for sample
labeling and the number of included subsequent layers are influential for the model performance. Moreover, the model prediction is shown to be sensitive to local porosity changes although it is trained on binary labeled data that disregards the actual sample porosity.
Talk about my recent research on data-driven chemical understanding with geometrical and quantum-chemical bonding analysis.
Vibrational properties play a key role in determining the stability and thermal conductivity behaviour of materials. The quasi- harmonic approximation gives insight into the phononic properties of a compound, but in the established way, i.e. density functional theory based methods, it takes many calculation steps and consumes a lot of resources to arrive at the desired results. Machine learning (ML) trained interatomic potentials (e.g. Gaussian approximation potential, GAP) pose an alternative to the traditional computation way of phonons. We develop a Python code based workflow which combines automation tools like atomate2 with ML to ease providing interactomic potentials for (quantum chemical) computations and databases.
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.
Additively manufactured (AM) triply periodic metallic minimum surface structures (TPMSS, from the English Triply Periodic Minimum Surface Structures) fulfill several requirements in both biomedical and engineering fields: tunable mechanical properties, low sensitivity to manufacturing defects, mechanical stability, and high energy absorption. However, they also present some quality control challenges that may prevent their successful application. In fact, optimization of the AM process is impossible without considering structural features such as manufacturing accuracy, internal defects, and surface topography and roughness. In this study, quantitative nondestructive analysis of Ti-6Al-4V alloy TPMSS was performed using X-ray computed tomography (XCT). Several new image analysis workflows are presented to evaluate the effects of buildup direction on wall thickness distribution, wall degradation, and surface roughness reduction due to chemical etching of TPMSS. It is shown that the fabrication accuracy is different for the structural elements printed parallel and orthogonal to the fabricated layers. Different strategies for chemical etching showed different powder removal capabilities and thus a gradient in wall thickness. This affected the mechanical performance under compression by reducing the yield stress. A positive effect of chemical etching is the reduction of surface roughness, which can potentially improve the fatigue properties of the components. Finally, XCT was used to correlate the amount of powder retained with the pore size of the TPMSS, which can further improve the manufacturing process.
Chemical heuristics are essential to understanding molecules and materials in chemistry. The periodic table, atomic radii, and electronegativities are only a few examples. Initially, they have been developed by a combination of physical insight and a limited amount of data. It is now possible to test these heuristics and generate new ones using automation based on Materials Informatic tools like pymatgen and greater amounts of data from databases such as a Materials Project. In this session, I'll speak about heuristics and design rules based on coordination environments and the concept of chemical bonding. For example, we have tested the Pauling rules which describe the stability of materials based on coordination environments and their connections on 5000 oxides from the Materials Project. In addition, we have created automated processes for analyzing the chemical bonding situation in crystalline materials with Lobster (www.cohp.de) in order to discover new heuristics and design rules.
Grain boundaries (GBs) are planar lattice defects that govern the properties of many types of polycrystalline materials. Hence, their structures have been investigated in great detail. However, much less is known about their chemical features, owing to the experimental difficulties to probe these features at the atomic length scale inside bulk material specimens. Atom probe tomography (APT) is a tool capable of accomplishing this task, with an ability to quantify chemical characteristics at near-atomic scale. Using APT data sets, we present here a machine-learning-based approach for the automated quantification of chemical features of GBs. We trained a convolutional neural network (CNN) using twenty thousand synthesized images of grain interiors, GBs, or triple junctions. Such a trained CNN automatically detects the locations of GBs from APT data. Those GBs are then subjected to compositional mapping and analysis, including revealing their in-plane chemical decoration patterns. We applied this approach to experimentally obtained APT data sets pertaining to three case studies, namely, Ni-P, Pt-Au, and Al-Zn-Mg-Cu alloys. In the first case, we extracted GB specific segregation features as a function of misorientation and coincidence site lattice character. Secondly, we revealed interfacial excesses and in-plane chemical features that could not have been found by standard compositional analyses. Lastly, we tracked the temporal evolution of chemical decoration from early-stage solute GB segregation in the dilute limit to interfacial phase separation, characterized by the evolution of complex composition patterns. This machine-learning-based approach provides quantitative, unbiased, and automated access to GB chemical analyses, serving as an enabling tool for new discoveries related to interface thermodynamics, kinetics, and the associated chemistry-structure-property relations.
Alkali-activated binders (AAB) can provide a clean alternative to conventional cement in terms of CO2 emissions. However, as yet there are no sufficiently accurate material models to effectively predict the AAB properties, thus making optimal mix design highly costly and reducing the attractiveness of such binders. This work adopts sequential learning (SL) in high-dimensional material spaces (consisting of composition and processing data) to find AABs that exhibit desired properties. The SL approach combines machine learning models and feedback from real experiments. For this purpose, 131 data points were collected from different publications. The data sources are described in detail, and the differences between the binders are discussed. The sought-after target property is the compressive strength of the binders after 28 days. The success is benchmarked in terms of the number of experiments required to find materials with the desired strength. The influence of some constraints was systematically analyzed, e.g., the possibility to parallelize the experiments, the influence of the chosen algorithm and the size of the training data set. The results show the advantage of SL, i.e., the amount of data required can potentially be reduced by at least one order of magnitude compared to traditional machine learning models, while at the same time exploiting highly complex information. This brings applications in laboratory practice within reach.
Developments in density functional theory (DFT) calculations, their automation and therefore easier access to materials data have enabled ab initio high-throughput searches for new materials for numerous applications. These studies open up exciting opportunities to find new materials in a much faster way than based on experimental work alone. However, performing density functional theory calculations for several thousand materials can still be very time consuming. The use of, for example, faster chemical heuristics and machine-learned interatomic potentials would allow to consider a much larger number of candidate materials. In addition to DFT based high-throughput searches, the seminar will discuss two possible ways to accelerate high-throughput searches.
Using data analysis on the structures and coordination environments of 5000 oxides, we were able to investigate a chemical heuristic – the famous Pauling rules – regarding its usefulness for the fast prediction of stable materials.
We have also investigated how machine-learned interatomic potentials can be used to accelerate the prediction of (dynamically) stable materials. The use of these potentials makes vibrational properties accessible in a much faster way than based on DFT. Our results based on a newly developed potential for silicon allotropes showed excellent agreement with DFT reference data (agreement of the frequencies within 0.1-0.2 THz).
In addition, we have successfully used high-throughput calculations in the search for new candidate materials for spintronic applications and ferroelectrics
The main outcomes of the EMPIR project nPSize are presented and the suitability of the new capabilities (e.g. reference materials, measurement procedures, inter-laboratory comparisons) as pertinent contributions to normative projects within CEN/TC 352 Nanotechnologies are discussed. E.g. the first technical report of nPSize on full algorithm sequences for nanoparticle detection and size measurement as developed on both a physical basis (SEM, TSEM, AFM and SAXS) and by machine learning is put at disposal.
A brief introduction to the efforts we have done in our lab towards AI/ML analysis of SAXS data. For this, we need to extend the data with an extensive, structured hierarchy of metadata and associated data. A practical look into the information stored in our files, and the organization of the files in a data catalog is presented.
An alternative method for lithium isotope amount ratio analysis is proposed by combining atomic absorption spectrometry with spectra data analysis by machine leaning. It is based on the well-known isotope shift of around 15 pm for the electronic transition at wavelength 670.7845 nm which can be measured by a high-resolution continuum source atomic absorption spectrometer (HR-CS-AAS). For isotope amount ratio analysis, a scalable three boosting machine learning algorithm (XGBoost) was employed and calibrated with a set of samples with a 6Li isotope amount fraction ranging from 99% to 6%. The absolute Li isotope amount fractions of these calibration samples were previously measured by multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) and used as ab-initio data for the machine learning algorithm. Validation of the machine leaning model was performed with two standard reference materials (LSVEC and IRMM-016). The procedure was employed for the isotope amount ratio determination of a set of stock chemicals (Li2CO3, LiNO3, LiCl, LiOH, and LiF) as well as a BAM candidate LiMNC cathode reference material. Achieved uncertainties are one order of magnitude higher than those obtained by MC-ICP-MS. This precision and accuracy is nonetheless sufficient to resolve natural occurring variations in Lithium isotope ratios. Also, the LiMNC material was analyzed by HR-CS-AAS with and without matrix purification. The results are comparable within statistical error.
An alternative method for lithium isotope amount ratio analysis is proposed by combining atomic absorption spectrometry with spectra data analysis by machine leaning. It is based on the well-known isotope shift of around 15 pm for the electronic transition at wavelength 670.7845 nm which can be measured by a high-resolution continuum source atomic absorption spectrometer (HR-CS-AAS). For isotope amount ratio analysis, a scalable three boosting machine learning algorithm (XGBoost) was employed and calibrated with a set of samples with a 6Li isotope amount fraction ranging from 99% to 6%. The absolute Li isotope amount fractions of these calibration samples were previously measured by multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) and used as ab-initio data for the machine learning algorithm. Validation of the machine leaning model was performed with two standard reference materials (LSVEC and IRMM-016). The procedure was employed for the isotope amount ratio determination of a set of stock chemicals (Li2CO3, LiNO3, LiCl, LiOH, and LiF) as well as a BAM candidate LiMNC cathode reference material. Achieved uncertainties are one order of magnitude higher than those obtained by MC-ICP-MS. This precision and accuracy is nonetheless sufficient to resolve natural occurring variations in Lithium isotope ratios. Also, the LiMNC material was analyzed by HR-CS-AAS with and without matrix purification. The results are comparable within statistical error.
In the present work a series of design rules are developed in order to tune the morphology of TiO2 nanoparticles through hydrothermal process. Through a careful experimental design, the influence of relevant process parameters on the synthesis outcome are studied, reaching to the develop predictive models by using Machine Learning methods. The models, after the validation and training, are able to predict with high accuracy the synthesis outcome in terms of nanoparticle size, polydispersity and aspect ratio. Furthermore, they are implemented by reverse engineering approach to do the inverse process, i.e. obtain the optimal synthesis parameters given a specific product characteristic. For the first time, it is presented a synthesis method that allows continuous and precise control of NPs morphology with the possibility to tune the aspect ratio over a large range from 1.4 (perfect truncated bipyramids) to 6 (elongated nanoparticles) and the length from 20 to 140 nm.