The "Alert" feature allows users to schedule automatic, regular, and retrospective data analyses on specific data segments. This functionality enables the establishment of alarm trigger criteria based on the results of these analyses.
Creating an alert relies on three main components:
- The frequency of the analysis,
- The data segment to be analyzed,
- The conditions for triggering an alarm.
The alert performs an analysis at a specified frequency, defined by the value entered in the "exceedance count" option under "frequency."
A daily alert performs the analysis every day, while a monthly alert executes an analysis monthly.
During a control phase, the alert processes the time series data (the data segment) from the associated sensors to determine whether an alarm should be triggered. The time series analyzed depends on several criteria:
Duration: This is based on the frequency of the alert analysis (as explained above). For the duration, we follow the logic: a daily alert retrieves data from the past 24 hours; a monthly alert retrieves data from the past 30 days.
Verification: You can choose to start the alert analysis from its creation or with a delay. The start and end of the analyses depend on the duration and delay. The rule to determine the end date of the analysis is: Execution Date - delay. From there, you can deduce the start date of the analysis using the duration (e.g., Start Date of Analysis = End Date - 24 hours for a daily alert).
Aggregation: This setting allows you to choose whether the time series data provided to the alert for analysis is aggregated data.
Thresholds: The lower threshold is exclusive, and the upper threshold is inclusive. In the example below, the value 20 triggers an alarm, but the value 10 does not.
Example:
My temperature analysis alert is daily. It triggers an alarm for temperatures between 24°C and 30°C.
It performs a daily analysis at 2:50 PM.
The associated sensor has data at a 10-minute time step but receives it all at once every day at 5:00 PM.
Being a daily alert, the analysis retrieves 24 hours' worth of data. However, since the data is received at 5:00 PM, it is not desirable for the alert to check the last 24 hours as the data is not yet complete. Using a 24-hour delay ensures a complete dataset for the analysis period.
On February 24 at 2:50 PM, the alarm retrieves the time series starting on February 22 at 2:50 PM and ending on February 23 at 2:50 PM. (End Date = Execution Date - delay, which is 24/02 - 24 hours, and Start Date = End Date - frequency).
Wishing to verify the average temperature and not the peaks, I choose an hourly aggregation. This means a single value of 24°C will not trigger an alarm, but an average value above 24°C over an hour will.