The Materials Science and Engineering (MSE) field is an interdisciplinary branch of engineering characterized by high volumes of heterogeneous data, which also serve as inputs for other fields reliant on materials. While semantic technologies are already utilized in various research areas to address data management challenges, their adoption in MSE is still in its early stages. This paper provides an overview of data management issues in MSE, existing semantic technologies, and the potential application of these technologies to address those issues. This leads to a roadmap for the implementation of semantic technologies in MSE.
Science generates vast amounts of data. Before investing efort and money in the generation of new datasets (e.g. by conducting experiments), it would be useful to find existing datasets that already address the scientific question at hand. However, the sheer volume of data makes this a daunting task, akin to searching for a needle in a haystack. Furthermore, the use of diferent schemas to encode and describe data complicates the search process.
To address this, the FAIR principles (Findable, Accessible, Interoperable, Reusable) were
proposed to enhance data sustainability [
MSE operates within a data-centric sphere, generating diverse datasets aimed at advancing manufacturing technologies and detailing material structures along with their relevant parameters. The sharing of this data, however, remains a challenge due to a lack of documentation, of standardized usage of Research Data Management (RDM) practices in the field, and others. Despite the rapid evolution of technology, the process of collecting new data in MSE still heavily relies on loosely defined analogue processes, even if the data products are, in the end, digital. Moreover, the absence of proper infrastructure, such as centralized databases for both data products and metadata, hampers presesrvation of this data. Scientific findings are often disseminated without establishing a clear link to the underlying data or software utilized in their generation, and there is a notable lack of integration with well-defined licenses or policies governing their usage. The absence of formalized processes and detailed descriptions poses significant challenges in conducting follow-up research, reproducing results, identifying potential sources of errors and workflow bottlenecks, and validating research findings.
Descriptive, process- and domain-specific metadata are typically encapsulated either in only human-readable resources or digital artifacts that are not machine-actionable. The multidisciplinary nature of MSE compounds the issue of a lack of common and shared representations for material knowledge, e.g., by ontologies, which are still insuficient in this field.
Laboratory metadata in MSE manifests in various forms, ranging from handwritten notes in laboratory notebooks or paper protocols to semi-digital lists and fully digital, yet not necessarily machine-readable or -actionable, data files. The adoption of electronic laboratory notebooks remains minimal. These factors collectively make it exceedingly challenging to analyze scientific laboratory workflows in MSE or employ process records in automated data analysis processes.
In this paper, we will outline how the majority of the MSE community’s data management challenges can be addressed through the incorporation of existing semantic technologies.
Currently, numerous national and international initiatives and consortia in MSE aim to establish
semantic technologies, such as shared vocabularies, metadata schemas, and ontologies. Notable
examples include the European Materials Modelling Council (EMMC) [
The first comprehensive review of ontologies in MSE, focusing on Domain-Level Ontologies
(DLOs), was recently published by De Baas et al. [
Currently, there is only one repository focused on MSE ontologies called MatPortal [
Ontologies are often used as the schema (T-box) component of a Knowledge Graph (KG),
which is then populated with domain-specific data, such as from experiments. Documents can
also contain relevant data or metadata from experiments. One way to populate KG with
information extracted from documents, are Large Language Models (LLMs). Ligabue et al. address
open information extraction on textual documents by creating sub-graphs from documents
and merging them into a larger KG [
Domain experts in MSE often lack the knowledge and experience required for developing usable and semantically correct ontologies. Since properly educated ontology developers are not readily available, the domain experts in MSE are often solely responsible for developing, implementing, and testing semantic technologies for their everyday work. The missing knowledge itself might already result in sub optimal semantic artifacts with issues such as ambiguity, semantic conflicts, and low interoperability. Furthermore, taking simultaneously a domain expert and an ontology developer role might cause conflicts of interest between the roles. In these cases the domain expert role is often dominant, due to the higher amount of experience in it, which might lead to decisions that are sound from the domain perspective, but semantically incorrect. Furthermore, having little experience in the ontology developer role, the domain expert is often not aware of these consequences.
This raises the question why the required experience and knowledge is lacking. First, the
adoption of ontologies in MSE is still in its early stages, explaining the lack of experience.
Second, acquiring the required knowledge is not straightforward. Directly learning from
examples of other domains that have progressed further in adopting ontologies, e.g. biomedical
domains, requires a good understanding of the respective domains. Third, domain-agnostic
ontology development methodologies are not well standardized, which poses a challenge for
the inexperienced ontology developer. While there are multiple ontology development
methodologies available and in use, they all share common roots and core processes. However, most
common frameworks (such as METHONTOLOGY [
There is an understanding of the necessity for and usefulness of interoperability of ontologies
in the MSE domain. However, finding suitable reusable ontologies, evaluating their quality, and
assessing their applicability are tasks typically within the expertise of semantic technology
specialists such as ontology developers. Despite the abundance of existing ontologies, locating a
specific one that fulfills the requirements to be reused for a given use case can be challenging. In
heterogeneous fields like MSE, domain ontologies, even when developed as part of an initiative or
consortium, are frequently created independently of other initiatives. This lack of coordination
between the initiatives makes these ontologies challenging to discover or harmonize, which is
especially important in case of intersecting domains. While ontology repositories that aid in
ifnding MSE ontologies exist, they are not well connected with each other and provide little
information about vital aspects like scope and addressed topics of the ontologies, which are
important to consider when reusing ontologies. Consequently, the concepts of ontologies are
ifndable, but it requires efort and expertise to assess how these concepts are related to the topics
one wants to address with ontology reuse. Moreover, owing to the often necessarily pragmatic
approach used in ontology development in MSE, the developed ontologies are often not rich
or explicit in their semantics, ofering only taxonomic relationships, and lack documentation
of the developed concepts [
Finding reusable ontologies is often further complicated, because it may not be possible to find
a single ontology that meets all given requirements. One option is to reuse multiple ontologies
to cover the requirements while maintaining high interoperability. However, this leads to
multiple challenges regarding their semantic compatibility, which possibly results in the need
to repeatedly align the ontologies during their integration. TLOs ofer a semantic framework
for harmonization and integration of DLOs [
Furthermore, it might not be possible to find required subdomains for a particular use case, or
the available ontologies of the subdomain are not interoperable with other required subdomains.
While the former aspect may be attributed to generally incomplete and heterogeneous coverage
of MSE subdomains, the latter is a consequence of the lack of harmonization endeavors between
subdomains [
The last group of challenges regarding the compatibility of semantic artefacts between diferent DLOs is caused by the inconsistent use of terminology and semantics. For example, two concepts from diferent DLOs might have the same label, but actually refer to diferent semantic concepts - or vice versa. This example becomes especially challenging, when the semantic concepts also lack an elucidating definition and descriptions, which may lead to ambiguous semantic concepts. While the structure of the DLOs can help decrease ambiguity of these concepts, the pragmatically structured DLOs in MSE often do not contain the required semantic richness to do so.
Laboratory work documentation is essential to ensure accurate replication, reproducibility, and traceability of experiments. At the same time, its preparation is a time-consuming and tedious task. Meticulous records of experimental procedures facilitate eficient data analysis and collaboration by enabling researchers to build on existing work. It is also necessary to meet regulatory requirements in the laboratory environment. However, diversity in documentation formats and limited use of Electronic Laboratory Notebooks (ELNs) complicate standardization eforts and the digitalization of laboratory (meta)data in MSE. Non-unified documentation practices and often manual (meta)data acquisition produce digital artifacts that have limited reusability, even for their original authors. Since metadata is not always connected to their related data, the establishment of meaningful relationships between datasets is hindered as well. This especially applies to process-specific records, where, without a semantic model of laboratory processes in place, the result is only artifacts of workflow runs lacking data provenance or data lineage for specific data products. The systematic use of Persistent Identifiers ( PIDs) for samples, equipment, and other elements of laboratory processes is also insuficient, itself not preventing poor metadata quality associated with these elements. Moreover, if MSE input data does not align with FAIR principles, it becomes dificult to produce FAIR data outputs. Manual, unstructured documentation is time-consuming and ineficient, yet researchers and laboratory technicians often resist adopting new technologies due to perceived complexity and the significant time required for implementation. The high investment in time, money, and expertise needed to implement semantic technologies presents a considerable barrier for many laboratories. Furthermore, the complexity of MSE data, which is often highly specialized, complicates mapping to existing semantic schemas. Establishing and maintaining relevant ontologies is resource-intensive, posing additional challenges, as noted in Sections 3.1 and 3.2.
Adhering to a well-defined methodology in ontology development ensures high quality and
reusability of the final product by providing structured guidelines for consistent design, rigorous
validation to catch errors early, and providing systematic documentation to support maintenance
and interoperability with other ontologies. This aspect is particularly relevant for domain
experts who are new to ontology development, as predicting potential issues in ontology
development can be challenging for them. The Linked Open Terms (LOT) methodology [
LOT has been proven efective across a selection of diverse projects, covering a range of
applications and domains [
Meanwhile, to enhance the interoperability of DLOs in MSE, concepts from a common TLO
should be reused to provide a base harmonisation of the DLO concepts. If DLOs use diferent
TLOs, it is suggested to engineer bridge concepts [
Besides manual ontology development, there are approaches to enhance ontologies
(semi-)automatically. Ontology Learning (OL) is a field of research in Artificial Intelligence ( AI) and
knowledge engineering. LLM-based OL techniques have shown to perform better than
traditional ones on unstructured text to extract significant and domain relevant concepts and
their inter-dependencies [
KGs are, unlike ontologies, meant to contain also entity-level data. Modelling KGs is even more
time-consuming than modeling ontologies. Information Extraction (IE) focuses on extracting
information from documents, e.g., from experiment protocols or from data sheets. KGs can
subsequently be enriched by this extracted information to reduce manual efort and increase
scalability. To enrich KGs with IE, triples must be generated and incorporated into the KG, a
process known as KG fusion or KG population. A typical IE task, comprising of sub-tasks like
entity extraction and relation extraction, has shown to achieve better performance using
out-ofthe-box LLMs with correct prompting techniques than with model finetuning [
Despite the significant variations among subdomains within MSE, there are noticeable parallels
in laboratory processes across the domain. Therefore, establishing unified frameworks for
collecting records of these processes can be done with a relatively low efort. Introducing
a process ontology, with a particular focus on gathering process provenance records in a
uniform format, can formalize these procedures. Utilizing PROV-O [
The semantic description of laboratory processes aligns well with the utilization of ELNs. To
overcome resistance to developing and implementing new technologies in the community, one
should be considerate when choosing a suitable ELN. First, it should reduce the efort necessary
for eficient and easy manual forms population while reducing the number of errors. This can
be achieved by field restrictions, such as drop-down menus and other forms of assistance in
free text entries, such as semantic annotation of the content. Second, the implementation of
ontologies is necessary to semantically annotate the entries or cross-reference sub-elements
of a given form, which can be especially useful for automating the ELN use. ELN Finder [
We have given an overview, which and how semantic technologies can be used to tackle data management challenges in MSE, including the domain specific dificulties in doing so. To show actual results based on this theoretical roadmap, we started a project called "Ontology-based Data Integration and eXploration" (ODIX). In ODIX, we re-use and create ontologies in the MSE domain, based on the LOT methodology. These ontologies are then used to guide LLMs in information extraction from documents. At the same time, the ontologies build the T-Box of a KG, which gets enriched with the information extracted from documents. The information in the KG is consequently well described to be used as input for further applications.
A key application in ODIX is a visual exploration tool that allows domain experts to compare information across diferent process steps of the same material probe or between diferent runs of the same process — all formerly only stored in documents. The project involves collaboration between computer scientists (primarily semantic experts) and domain experts (mostly in the area of MSE). Evaluation of the results will be conducted by the domain experts with outcomes to be published as the project advances. We look forward to seeing further practical applications of the methodologies and processes outlined in this paper.