Poster: Towards Explaining the Effects of Contextual Influences on Cyber-Physical Systems

The increasing complexity of Cyber-Physical Systems (CPS) increases
the difficulty for users to understand their behavior. Using
existing Explainable Artificial Intelligence (XAI) methods, CPS can
explain their behavior to the users. However, the input-output correlations
used in XAI methods are not capable of explaining certain
anomalies in CPS behavior caused by contextual influences (CIs)
since they do not consider the context of the CPS. Some well-known
techniques used for understanding such CIs on CPS are test chambers
and the analysis of logged CPS data. However, test chambers
are typically only available to the manufacturer of a CPS, thus
not useful for understanding CIs on the shop floors. Data analysis
methods focus on data correlations, which are insufficient to explain
causal relationships without using expert (human) knowledge.
Hence, we propose a context-aware log-based explanation system
to explain the causal relationship between CIs and the behavior of
a CPS. The proposed solution employs semantic technologies to
access the context of the CPS. It demonstrates the causal relationship
between the CPS and CIs through counterfactual explanation
and abductive reasoning methods. The contextual explanations offered
by the proposed system will assist users in visualizing diverse
scenarios in order to improve the CPS’ behavior accordingly.