Time-series based simulations of industrial processes are instrumental to optimizing a variety of industrial settings. In this paper, we describe a use case, developed together with In neon Technologies Austria AG. Monitoring data stored in relational databases was used to build process models of industrial chillers. Optimization algorithms were then applied to nd optimal strategies for operating the chillers. Even though the results from this approach were convincing, the access to the necessary data was a labor-intensive and error-prone task. Therefore, in this paper, we investigate how Semantic Web technologies can help to improve data access for time-series data and under which circumstances they would be helpful for the domain experts performing the simulation.
This use case is an actual semiconductor plant operated by In neon Technologies Austria AG in Villach, Austria. In total, twelve chillers with electrical power input rates between 300 and 500kW were investigated. The optimization of the operation schedule of these chillers was the goal of our use case. Simulation models were created using historic monitoring data. The resulting simulation model was combined with an optimization module. Accessing the data necessary to create those models is an error prone and tedious task. Therefore, we propose the extension of a simulation software with Semantic Web technologies which allows the de nition of semantic mappings to the input sources of time-series data in line with ontology-based data access (OBDA) principles.
We developed a Semantic Web prototype in the context of a concrete use
case from industry. This use case originally was de ned in the research project
Balanced Manufacturing (BaMa) [
Recent studies [
OBDA systems were developed in the course of the EU funded Optique
project3. One of the results is the Ontop framework which we also make use
of in this work [
Facilitating Access to Simulation Data with Semantic Web Technologies In the given application area, the data access tasks vary and are hardly ever repeatedly executed. Therefore providing prede ned SQL queries is not a suitable solution as these would have to be adapted quite often.
We applied a simpli ed version of existing ontology development methods in order to develop the ontologies for our OBDA prototype. The primary goal was to evaluate the general technical applicability and motivate further development. The development process started with example expert questions. These, and other similar, questions had to be answered throughout the model parametrization process. Some example questions are: { Which chillers are located in building XY? { Show all chillers located in country YW. { Which chillers are connected with recooling plant YZ?
Based on those questions, two ontologies were created: the Top Level Ontology and the Chiller Ontology. The decision to create two rather than a single ontology was made in order to illustrate that it is possible to combine di erent conceptualizations of a domain in one prototype tool.
The Top Level ontology, depicted in Figure 1a, represents the main domain concepts and their relations. This ontology could be used in order to combine di erent, more specialized ontologies, such as the one depicted in Figure 1b. The concept Machine from the top-level ontology is further specialized in the Chiller ontology in terms of various chiller types. A di erent way to expand the ontology might be to describe the Building domain in more detail. Such extensions, however, were not in the scope of the prototype.
We used Ontop's Protege plugin to implement our prototype. This plugin aids the mapping process. A mapping combines results from SQL queries with ontology elements such as concepts and predicates. This can be seen as a virtual T-box. The de ned mappings are then used to translate SPARQL queries 3 http://optique-project.eu/
Improving industrial optimization with Semantic Web technologies (constructed with terms from the ontology) into SQL queries (executed on the underlying databases).
An example query can be seen in (Listing 1.1), some technological capabilities of the Semantic Web are used to identify sensors based on knowledge such as machine types and machine connections. This information, without the use of the Semantic Web, would have to be extracted from factory plans or wiring diagrams. Necessary meta-information, which is stored in the database, can be used directly by domain experts for their queries. In the given example, only sensors generating data from machines which are connected to other machines are returned. Then, the results are ltered so that only those where machine 2 was a speci c kind of recooling plant (identi ed via its name) are returned to the user. Another aspect that is covered in this example is reasoning. Only the predicate isMeasuredBy was explicitly mapped to data from the underlying dataset. Nevertheless, the query makes use of the measures predicate (the inverse of isMeasuredBy) which is possible thanks to the reasoner being aware of OWL semantics. The presented prototype was shown to project partners familiar with the use case. Then, user interviews were conducted. Based on the results we are condent, that the Semantic Web can improve data access for domain experts and therefore improve adaption of simulation-based approaches to optimization tasks. The general approach is seen positively by the involved stakeholders. User interaction and implementation with legacy systems, however, need to be addressed in order to make the system actually useful in a production environment.
We would like to thank our project partner, In neon Technologies Austria AG for the constructive cooperation both throughout the BaMa project4 and the writing of this article. 4 http://bama.ift.tuwien.ac.at/