Setup the anomaly detection module for high frequency, non cyclical data

[LR_9796586]

I have a dataset with 500 cycles of an industrial process, these samples have a duration of 8 seconds at a sampling rate of 4 milliseconds, that accounts for 2048 samples for each cycle. 

The sampling events are unevenly distributed in time, as shown in the image below where he green blocks mark the start of a sampling event:

In this scenario, E.G.  On Monday a cycle of 8 seconds is taken at 1:00 and at 5:00, but on Tuesday there is only one sample at 3:00.  

Does it make sense to concatenate several signals  from different dates (nine in this case) to train the anomaly detection algorithm? 

In that case, which can be proper values for the following parameters: 

Outbound Anomaly Rate:   To start with, it makes sense to me for this value to be 8 seconds, which is the same duration of the sampled cycles.

Sampling rate: The sampling rate of the system in production is 4 milliseconds, does this rate have to be the same?

[nsampat · Accepted solution]

@LR_9796586 ,

Thank you for posting your question to the PTC Community.

Unfortunately ThingWorx Analytics Anomaly Detection does not support non-cyclical data, as it requires to train on cycling predictable data trend to identify anomalous data when encountered.

More details around Anomaly Detection can be found on our HelpCenter Topic: ThingWorx Analytics 9.0/9.1 - Anomaly Detection

Regards,

Neel

[LR_9796586]

in this case the process itself is cyclical, but the sampling is not. Is it recommended to use  ThingWorx Analytics Anomaly Detection? 

[nsampat]

@LR_9796586 ,

Based on this statement from your post:

"The sampling events are unevenly distributed in time, as shown in the image below where he green blocks mark the start of a sampling event"

This would not be a good use case for ThingWorx Anomaly Detection. Intervals need to be consistent to enable the process to learn and train.  If your data is not sent in a cyclical rate, it will not be able to train correctly.

Regards,

Neel

[slangley]

Hi @LR_9796586.

If you feel your question has been answered, please mark the appropriate response as the Accepted Solution for the benefit of others with the same question.

Regards.

--Sharon