A short step between reading and meaning
An experimental system for wireless sensor networks generated small, machine-readable descriptions of sensor measurements quickly: most micro-ontology generation took place within 500 microseconds, although some outliers exceeded 1 millisecond. The reported generation times were not significantly different across OWL serialization formats, which are different ways of writing the same ontology data.
The work, called LEMONS, sets out to apply Semantic Web technology to large-scale wireless sensor network configuration and management without requiring expert knowledge. Its architecture separates WSN modeling and setup from runtime processing, dividing the task of preparing the network from the later handling of measurements.
The illustrative setup used a Waffle TurtleBot with a Nordic Thingy:52 sensor platform. MQTT linked a Raspberry Pi gateway with generic applications and the RDFier component, while BLE connected the gateway to the sensor nodes. The arrangement supplied the concrete setting for the semantic-enrichment test.
At runtime, the system was reported to generate and serialize semantically enriched measurement data. The modeling environment generated gateway code, a static ontology and the RDFier. In this use case, gateway code generation reached 100 percent.
The modeling claim extends beyond code. The authors report that a person without expert knowledge could create the LEMONS model in the modeling environment, and that models could be stored and loaded for network configuration and management. The report does not present this as a formal usability assessment.
What the measurements captured
To examine timing, the authors ran three measurements on an Intel CPU clocked at 2.67 GHz with 8 GB of RAM. The first run lasted 5 minutes and collected over 179,000 data points; the second ran for 60 minutes and collected over 266,000; the third lasted 4 minutes and collected over 17,000. Each run continued until an adequate number of data points had been acquired.
The analysis compared execution times for individual processing phases across OWL serialization formats. It also examined how GZIP compression changed micro-ontology size and related processing time to generated size for different property types. Timing results were presented using mean and exact execution values, keeping the comparison focused on the generated software and data.
Compression produced a split result. The compression ratio depended strongly on the serialization format, meaning the format changed how much the original data could be reduced. Yet the compressed micro-ontology sizes themselves showed negligible variance between formats. In this evaluation, format choice mattered more for the ratio than for the final compressed size.
Across property types, all RDFier-generated micro-ontology instances were smaller than 1 KB, and their size variance was reported as non-significant. That result gives an upper bound for the instances produced in the evaluation, although it does not provide a full distribution of sizes.
The model still has gaps
The authors interpret the timing in practical terms: they say the overhead is likely to be dominated by network transmission and describe a trade-off that favors a non-intrusive design over the lowest possible latency. They also interpret the approach as a route toward lower setup effort and management complexity, broader interoperability and smaller MQTT payloads through compression.
The approach has a stated boundary. Its current WSN metamodel, the template used to represent a network, does not cover the entire SSN ontology. The authors flag missing specialized-vector definitions in QUDT and differences in what SSN and EMF can express as continuing challenges.
The reported evidence stays within the illustrated setup and three benchmark runs on one CPU configuration. It covers generated software, semantic serialization, compression and property-type comparisons, so the numbers describe that test arrangement.
The supplied manuscript identifies itself as an accepted manuscript of a 2020 SEAA conference paper and states that a version of record is available. It reports partial funding from the BMBF COMPACT project, grant 01|S17028C, and the BMWi InsightProducts project, grant 228EN/2.
Paper data and sources
Original title: LEMONS: Leveraging Model-Based Techniques to Enable Non-Intrusive Semantic Enrichment in Wireless Sensor Networks
Authors: Jan Novacek, Arthur Kühlwein, Sebastian Reiter et al.
Journal/Repository: 2020 46th Euromicro Conference on Software Engineering and Advanced Applications (SEAA), Portoroz, Slovenia, 2020, pp. 561-568
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: 10.1109/seaa51224.2020.00092
Original paper · Full text