How to follow up anomalies

Anomalies are deviations Properate finds by itself, without anyone having set a threshold.

Anomalies are deviations Properate finds by itself, without anyone having set a threshold. The algorithm learns what's normal for a timeseries and flags what breaks the pattern.

That's the difference from an alarm: an alarm fires on a limit you decided; an anomaly fires on something that doesn't look normal.

How to open anomalies

  • Go to Anomalies in the sidebar.
  • Set the Period: — Half year, for example.
  • Use Search to find a particular system.

The columns in the table

Column

Contents

System

the installation

Health

the system's current health

Anomalies in chosen period

number of findings in the period

Monitored timeseries

how many series are being followed

Affected timeseries

which series the findings concern

Algorithm

which algorithm found them

Analysis

a link onwards to the analysis

How to read the table

Health is the entry point: a system in poor health has had persistent deviations, not just a single spike.

The combination with the two timeseries columns is where the insight sits:

  • High Anomalies in chosen period with Affected timeseries = 1 points at one sensor or one component.
  • Many Affected timeseries points at something covering the whole installation — an operational change, a rebuild, or a genuine fault.
  • Low Monitored timeseries means the basis is thin: few series give few chances to detect anything.
Note: An anomaly is an observation, not a conclusion. A rebuild, a new tenant or changed operating hours produce perfectly legitimate anomalies — the data is unusual, and the algorithm is right to say so. If the cause is known, add a note on that period so the next person doesn't troubleshoot it again. See How to add a note.

How to get from a finding to a cause

  • See which Affected timeseries the finding concerns.
  • Use Analysis to open the data.
  • If the deviation is real and persistent, set up an alarm so you catch it next time without having to look. See How to create an alarm.

That last point is the purpose of the module: anomalies find what you didn't know to look for, and after that an alarm is the right tool for monitoring it going forward.

Frequently asked questions

Do anomalies notify me?

The anomaly list is an overview you look at. Notification about events goes through alarms and alert configurations — see How to create an alert configuration.

Why does a system have no anomalies?

Either operation is stable, or Monitored timeseries is low. If the number is zero, nothing is being monitored for that system.

Can I choose which algorithm is used?

The Algorithm column shows which one produced the finding. Detection runs in the background on the monitored series.

Which period should I look at?

Half year gives enough data for the patterns to be clear. Shorter periods make normal seasonal variation look like deviation.

Tip!

Work through the anomaly list when you take over a building you don't know. It points at the installations behaving differently from how they should — and that's a far faster route into an unfamiliar building than reading through timeseries system by system.