What comes after DataConnect?

[drichter]

Hi,

I implement the whole tutorial of using data connect (How to use DataConnect when sending property values from the ThingWorx platform to ThingWorx Analytics.) and this works fine. The data will send to analytics, analytices make its job and then? I'm not realy sure what happens next? Some Questions I have:

• How to work with he results of analytics?
• How to implement a predictive maintenance like (xx days to next failure) like in the pump flow demo?
• How to predict an error from the current data/properties?
• How often shoule I send the data to analytics? Every new record?
• How to bind the results of analytics to a mashup?

I hope someone can help.

[maglietti]

Hi David,

I suggest that you work your way through the "Analyze" guides on the Developer Portal. The Anomaly Detection How-To Guide will get you going in the direction of the questions that you asked.

-- Michael

[drichter]

Anomaly Detection and predict the future with current data is not the same or I'm wrong?

For better explanation. I build a simulator which sends two double values (value1, value2) + an error value which is true or false. Value1 and Value2 random values between 14 and 16. After some minutes value2 increase with random values between 16 and 20. Some minutes later (for example 10 minutes) the error value switch to true.

My idea:

• thingworx (analytics or what ever) learn this behavier (value2 is too high ==> in 10 minutes you get a failure)
• thingworx gives me a notification like "failure is upcomming in 10 minutes" when value2 is to high

[cmorfin]

Hi David

I will try to answer some of your initial points about what to do once data has been collected with DataConnect

To predict a future outcome, we will need to build a predictive model  first and then score new data against this model in order to predict the outcome.

To build a predictive model, ThingWorx Analytics needs to receive as much historic data as possible to learn the data and the relation between the different features in this data.

When using ThingWorx Platform, you need to use DataConnect to send historic data from the platform to Analytics Server.
Once you have a dataset created in Analytics Server, you will use REST API or, easier, ThingWorx Analytics Builder to build a predictive model.

Once you have tested this model and are happy with its performance you can use it to score production data.

Again in the case of the ThingWorx platform, if you want to be able to score real time, then you will need to publish the predictive model from Analytics Server to ThingPredictor, then you can predict the outcome for new incoming values.

The following article lists the different steps necessary to implement a model in ThingPredictor with some videos, so it probably should be useful: https://www.ptc.com/en/support/article?n=CS262212  .

The following article gives a short summary of the different solutions that exist withing ThingWorx Analytics, which may help to understand where all fits: Article Viewer | PTC

Kind regards

Christophe