Real-Time Visualization of Large-Scale Sensor Data

The proliferation of sensors in underground mining and heavy industry has createddata streams of unprecedented scale and complexity. A single hard-rock mine canproduce millions of measurements per shift, while modern smart rock bolts report submillimeter deformations to cloud servers in real time. Operators tasked with maintainingproduction and safety must sift through thousands of parallel time-series, identifyemerging anomalies and take rapid corrective action. Conventional dashboards, composed of a grid of small line charts, quickly succumb to clutter and performancelimitations as the number of sensors grows. Furthermore, the psychological constraintsof human working memory limit how many simultaneous signals can be tracked; userstypically cannot attend to more than seven items at once. The resulting informationoverload increases the risk of “alert fatigue” and decision errors. This thesis proposes and evaluates a suite of visualization techniques designed tomake large-scale sensor networks legible at a glance. The approach combines hierarchical aggregation with real-time status summaries, drawing on the visual informationseeking mantra - overview first, zoom and filter, then details on demand. A radialsunburst diagram represents nested groups of sensors, allowing operators to collapsethousands of channels into a few colored wedges without losing context. The radialextent of each wedge encodes the relative severity of alerts in its sub-tree, therebyguiding attention towards high-risk areas. Complementary stacked area charts displaythe temporal evolution of status counts (“fine,” “warning,” “alert”) for the currentselection, while linked line graphs provide precise sensor-level readings when needed.Direct manipulation - dragging and dropping sensors to restructure groups - enablesthe interface to adapt to evolving mine layouts and user mental models. The design isimplemented as a web-native prototype using a simulated data pipeline and a responsive database that pre-aggregates status metrics. Only changed values are streamedto the client, ensuring smooth performance even with thousands of live sensors.This aligns with best practices in real-time dashboards and hierarchical visualizationresearch. To evaluate the system, the work combines quantitative performance experimentswith qualitative feedback from domain experts. Synthetic datasets model realisticdeformation patterns using Perlin noise processes, while stress tests measure framerates under varying sensor counts. Interviews with engineers and geotechnical specialists from ThingWave AB and partner mining companies assess perceived usability,situational awareness and trust in the visualizations. The results demonstrate thatthe hierarchical-plus-temporal view reduces time-to-detection compared to traditionaldashboards and is preferred by operators for rapid triage. The sunburst representationin particular was reported to be intuitive and insightful, confirming findings from previous studies comparing radial and rectangular layouts. Users also valued the ability tocreate ad-hoc sensor groups, reflecting dynamic production processes.