Visual Analysis and Anomaly Detection of Material Flow in Manufacturing

Abstract The automated tracking of objects in factories via real-time locating systems (RTLS) is gaining increased attention due to its improved availability, technical sophistication, and most of all, its plethora of applications. The tracking of workpieces through their production process, for example, unlocks a detailed understanding of timings, patterns, and bottlenecks. While research mostly focuses on technological advancements, the analysis of the generated data is often left unclear. We propose a visual analysis framework based on ultra-wide-band (UWB) RTLS tracking data of material flow for this purpose. With this, we present an analysis and define a practical approach for how factory-level data can be analyzed. Advanced algorithms adapted from non-adjacent research domains are used to process and detect anomalies in the data, which would otherwise be hidden behind oversimplified analysis methods. Our approach considers different levels of granularity for the analysis in its visualization and, therefore, scales with increasing data sizes effortlessly. We also generated a ground truth dataset of RTLS UWB data with labeled anomaly cases. Combined, we provide a full, end-to-end, efficient processing and multi-visualization analysis pipeline for self-contained yet generalizable UWB RTLS data.