A tree of shapes, matched as readings arrive

How Patternode reads a historian archive: the segments the historian kept become the leaves of a tree of shapes, and each new segment is matched against trees already seen, so a shape unlike anything in normal operation stands out while the values are still in range. An illustration on made-up data, not a measured result.

Press play to watch segments arrive and be compared with normal operation.
Left, a signal as a historian keeps it, with the kept points joined into segments, and above it a tree that groups the segments into larger shapes. Right, a column for each arriving segment showing how far it is from the nearest segment seen in normal operation, against an alert line. The steady cycle stays low; the slow drift climbs over the line while the values are still in their usual range.

    An illustration on made-up data, not a result: no detection rate, lead time or diagnosis accuracy is claimed. Patternode is testing this method on a public benchmark (Tennessee Eastman) first, then on a utility's own historian data.