Planning TSI Environment/Architecture for Plant Feedback Telemetry

Category: azure time series insights


alivebeef on Sat, 21 Mar 2020 14:24:34

I’m a university student looking to start using the Azure platform in one of my projects namely for plant performance feedback.

The plant will eventually have more sensors added to begin research into autonomous functionality and Industry 4.0 however I’m starting with basic telemetry.   I’m looking to stream data into Azure from an end device with a multitude of sensors using JSON via MQTT into IoT Hub which will then feed into Time Series Insights.

Further analytics will hopefully occur using Power BI alongside the TSI Explorer

If I’m reading the documentation correctly this should work?

The plant goes through a defined process of operations after which some parameters may be changed, and the plant will do another run.

Would TSI be able to split the data up between the different runs of the process? i.e. Configuration A of the plant does several runs, then we make a change to Configuration B to do another several runs.

The process has several different operations and we would like to compare between different configurations across different runs. (Configuration A might make the first 10 operations more efficient vs Configuration B for example)

Some configurations can be fixed for the process run, and other configurations directly affect the process data (maximum velocity for an actuator for example) so some could be stored as metadata?

In an ideal world we hope to be able to change parameters on the plant, input this via a PowerApp that can interface with this architecture (or record as metadata for the process run)

Eventually we would like to start using Machine Learning and Analytics within Azure to help us decide what configuration changes to make to optimise the process (and potentially train the ML/AI required to do autonomous operation)

JSON Example

  • Sensor_Array
  • Velocity_Array
  • Temp_Array
  • Position_Array

PowerApp Data Input Example

  • Physical_Setting A
  • Physical_Setting B
  • Physical_Setting C
  • Physical_Setting D


lyranah on Tue, 24 Mar 2020 18:09:21

Hello, yes this an exact use-case for Azure Time Series Insights, that sounds like such a cool project! It would help to have the actual JSON sample--I see in your bullet points that you have arrays and so I'd like to understand the shape of the payload. I'm unclear if the JSON example above is the payload for one event, or if multiple events are being sent in each array?

This is a great example of how the Time Series Model can be leveraged to model data and analyze results from different settings. There are a few ways that you can model your set-up in TSI, but based on your desired PowerApp input you might consider doing it this way:

Create unique IDs for each sensor in each configuration (as if there were actually four different plants with one or many sensors in each plant), and that value (sensorId or whatever you refer to it as) then becomes your environment Times Series ID. Alternatively, you could also set up a composite TS ID comprised of two values: sensorId and physcialSettingId. In that case, those two values together become the unique identifier for your time series events. You'd have to be sure to update the value of physcialSettingId when switching configurations and send that key value pair in each message along with the sensorId.

Create a hierarchy with a Level "Physical Setting" (or whatever you prefer)

Add variables for velocity, temp, and position to the defaultType, or create your own Custom type(s). 

Create time series instances corresponding to each sensor in each physical setting. You can edit these and upload in bulk. 

You might want to go through this lab which shows TSM model creation to get more familiar with it.

With this model, you'll then be able to make comparisons by plotting the time series in the explorer, similar to the image shown that is from our demo environment


AshokPeddakotla-MSFT on Fri, 27 Mar 2020 09:07:12


Have you had a chance to see the above response? Do let us know if you have further queries.

AshokPeddakotla-MSFT on Wed, 08 Apr 2020 18:51:00

Hello alivebeef,

I am marking the above response as answer. Please feel free to respond if you have further questions.