SigSPARQL: Signals as a First-Class Citizen When Querying Knowledge Graphs

TL;DR

This work tackles the challenge of leveraging sensor-rich cyber-physical systems by embedding signals, i.e., time-evolving sensor values, as first-class citizens in knowledge-graph queries. It introduces SigSPARQL, a SPARQL extension that supports a signal-annotated RDF data model and FRP-inspired semantics, enabling continuous, time-aware queries that compute on signals and trigger events. The authors present a data model, syntax, and semantics for signals, a feasibility-focused prototype, and a proof-of-concept monitoring use case, demonstrating that graph-based context plus runtime signals can be queried in a single framework. The approach promises more scalable, domain-agnostic CPS monitoring, enabling single queries to span multiple system instances and yielding actionable trigger events, with future work on richer signal processing and large-scale deployments.

Abstract

Purpose: Cyber-Physical Systems (CPSs) integrate computation and physical processes, producing time series data from thousands of sensors. Knowledge graphs can contextualize these data, yet current approaches that are applicably to monitoring CPS rely on observation-based approaches. This limits the ability to express computations on sensor data, especially when no assumptions can be made about sampling synchronicity or sampling rates. Methodology: We propose an approach for integrating knowledge graphs with signals that model run-time sensor data as functions from time to data. To demonstrate this approach, we introduce SigSPARQL, a query language that can combine RDF data and signals. We assess its technical feasibility with a prototype and demonstrate its use in a typical CPS monitoring use case. Findings: Our approach enables queries to combine graph-based knowledge with signals, overcoming some key limits of observation-based methods. The developed prototype successfully demonstrated feasibility and applicability. Value: This work presents a query-based approach for CPS monitoring that integrates knowledge graphs and signals, alleviating problems of observation-based approaches. By leveraging system knowledge, it enables operators to run a single query across different system instances within the same domain. Future work will extend SigSPARQL with additional signal functions and evaluate it in large-scale CPS deployments.

Figures (4)

• Figure 1: charging garage with multiple sensors (circles) at different devices (AP = Active Power, SoC = State of Charge, OE = Operating Envelope Limit)
• Figure 2: Modeling the Garage with the extended data model from Section \ref{['sec:datamodel']}. Green nodes represent the signals while the dashed lines represent a mapping from graph nodes and signal properties to the actual signals.
• Figure 3: Two power sensors (red and blue), and their sum (brown) based on different models for values between observations.
• Figure 4: Illustrates a that compute the active power for three garages: <garage1> (blue), <garage2> (red), <garage3> (brown). The graph shows the temporal evolution of the resulting signals, the center table depicts the with bound signals, and the right table shows the result of evaluating the at $\tau$.

Theorems & Definitions (2)

• Definition 7.1
• Definition 7.2