Visual Analytics for Root Cause Analysis in Self-Organizing Industrial Systems

Abstract

Root cause analysis (RCA) is a central task for quality assurance in manufacturing plants. By tracing back anomalies to its actual trigger, recurrent misbehavior can be eliminated, which improves the system’s future performance. In self-organizing industrial systems (SOIS), however, where the system adapts its behavior to the current circumstances and requests, new challenges arise for RCA. For example, the system decides dynamically at runtime how to route the work-pieces through the factory. This high degree of freedom of the system causes a state space explosion, which makes it difficult to formalize explicit connections. In addition, there are new dependency relationships resulting from the online decision making process and its influencing factors, which have to be taken into account for RCA. Accordingly, in this paper, we present first of all a taxonomy of possible root causes in such SOIS. Thereby, we focus in particular on possible error sources resulting from the online decision making process. Based on this, corresponding backtracking approaches are presented, whereby automatable and non-automatable procedures are distinguished. The latter becomes relevant in case that a component of the online decision making system is not evaluable automatably due to the state space explosion. To trace back anomalies anyway, we propose here a visual analytics solution. A corresponding proof of concept which implements the necessary functions for an expert-based assessment is presented in this paper.