Data aggregation computes summary values from multiple records. For time-series data, readings are grouped into time windows, then reduced to statistics such as a mean, minimum, maximum, or sum. A digital twin can use those summaries to show trends without repeatedly processing every sensor reading. section 4.3.3 section 6.3.2
From readings to a useful chart
A temperature sensor produced 6,375 readings over 45 days. Calculating each day’s low, high, and mean turns them into 45 daily records. Plotting those records makes the spring warming trend and unusual patterns easier to see than either the raw readings or the summary table alone. section 6.2.1
timestamp,min,max,mean 1754784000,13.8,18.4,16.31 1754870400,12.8,17.6,15.57 1754956800,13.3,17.4,15.99 1755043200,15.8,19.3,18.14 . . 40 readings . 1758585600,19.9,22.7,21.64
The daily records show the time bucket alongside its minimum, maximum, and mean. Each row summarizes many readings rather than representing one sensor measurement. section 6.2.1
Choose windows to match the question. Minute-level summaries can support close inspection; hourly, daily, or monthly summaries can support longer views. Precomputing these resolutions avoids recalculating statistics for every dashboard request and lets a chart switch resolution as its zoom level or viewport changes. In a data warehouse, these resolutions are called grains. section 4.3.3 section 6.3.2
Aggregation is not the same as decimation. Aggregation calculates new values for time buckets and suits long-term trend dashboards. Decimation removes selected points while retaining the curve’s shape and suits zoomed-in, high-frequency views. A mean alone can obscure a brief extreme; storing the minimum and maximum alongside it retains evidence of peaks within each window. section 6.3.2
Aggregation can also happen as data arrives. A 30-second sliding average of one-second temperature readings drops the oldest reading as each new one enters, smoothing noise without waiting for a batch. The same windowing approach can track the share of late train movements over five minutes for a timely operational measure. section 4.4.2
For a stored dashboard dataset, aggregate during transformation instead. One Apache Spark example reads minute-by-minute power CSV data, groups records by the week of their local timestamp, sums watt-hours into kilowatt-hours, and sums cost. Validate readings and standardize units before relying on derived figures; retain transformation lineage so results can be traced back to their inputs. section 4.4.4
