Ingenieurwissenschaften und zugeordnete Tätigkeiten
Filtern
Erscheinungsjahr
- 2021 (2) (entfernen)
Dokumenttyp
Sprache
- Englisch (2)
Schlagworte
- Sequential learning (2) (entfernen)
Organisationseinheit der BAM
Environmentally friendly alternatives to cement are created through the synthesis of numerous base materials. The variation of their proportions alone leads to millions of materials candidates. Identifying suitable materials is very laborious; traditional systematic research in the laboratory consumes a lot of time and effort.
Sequential learning (SL) potentially speeds up the materials research process despite limited but highly complex available information. SL does not make direct predictions of material properties but ranks possible experiments according to their utility. The most promising experiments are prioritized over dead-end experiments and experiments whose outcome is already known.
Our work has shown that SL seems to be promising for cement research. So far, research has mainly focused on materials whose synthesis is faster and whose material properties require less time for development or characterization (allowing many successive experiments). Contrarily, in the case of binders, SL is only useful if few experiments lead to the desired goal, as for example, the determination of the compressive strength alone typically requires 28 days.
In research practice, experimental designs and the availability of resources often determine which data can be used - for example, when some laboratory resources are not available or deemed irrelevant to a task. As a result, new research scenarios are constantly emerging, each of which requires to demonstrate SL’s performance.
We are presenting the SLAMD app to facilitate the exploration of SL methods in numerous research scenarios. The app provides flexible and low-threshold access to AI methods via intuitive and interactive user interfaces. We deliberately pursue a software-based research approach (as opposed to code-, or script-based). On the one hand, the results are more comprehensible since we refer to a common (code) basis (’reproducible science’). On the other hand, the methods are easily accessible to all which accelerates the knowledge transfer into laboratory practice.
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.