Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
cancel
Showing results for 
Search instead for 
Did you mean: 
0 Kudos

Will all IOT / Industrie 4.0 happen in the cloud?


SAP has launched the HCP as the IOT platform - in the cloud. Siemens announced to build their industrial IOT cloud on HCP (Cloud for Industry - Industry Services - Siemens) - in the cloud.

 

But I also still hear Quentin Clark's answer when asked about IOT in the cloud during a panel discussion at SAPPHIRE NOW 2015.  He said he expected to see as much intelligence pushed to the edge as possible.

 

Thinking back to Hannover Fair I agree strongly. Several sensors manufacturers have integrated a lot of intelligence in their sensors - to collect, filter and some even plan to "predict" locally.

 

To the edge, or to the cloud then?

 

Is this a contradiction? I do not think so, and I would like to use a few examples from the mill industries to suggest guiding principles whether cloud or edge is more appropriate.

 

Where belongs the big data - in the cloud or on the shop floor?


Howard Baldwin from Forbes wrote a great summary Does big data belong up there or down here? - Thoughts On Cloud where he re-iterates the idea of "data gravity" and the suggestion "to do the appropriate analytics wherever the bulk of the data has landed". The moment you want to bring in data from somewhere else - like weather data, ERP data - the cloud may be the better way to do this.

 

Indicators for cloud-based IOT scenarios



  1. The analytics & predictions happen at a location different to the one where the data is generated.
    Examples:
    Central monitoring of remote asset health, OEE, yield or quality data.
    Monitoring of fleet sensor data
    Any remote locations involved (from wind turbines to pipelines)

  2. Data is analyzed centrally from multiple locations.
    Example:
    Central data science team looking for root causes, correlations and predictive models

  3. The data source is on-site at another company.
    Examples:
    The Siemens scenario above, and any machine manufacturer's predictive service & maintenance offering.
    Any smart meters involved

  4. Multiple parties are involved
    Examples:
    Connected logistics examples with own and external fleet
    Connected home where multiple parties communicate (windows' rain sensor, heating, ventilation or air condition, shutters, lighting)

  5. Additional external data sources required
    Example:
    Weather data like temperature & humidity
    CRM or ERP data  like complaints, warranty claims etc


 

Indicators for IOT scenarios on premise, close to the edge & sensor world



  1. Data is analyzed and acted upon locally
    Example:
    Short interval control type of scenarios where the machine operators review real-time production data to take improvement decisions locally within the same shift e.g. to manually adjust run rate if an asset or quality deviates.

  2. Immediate action required
    Example:
    In-line quality analysis where an error needs to be recognized and reacted upon e.g. in range of 1 sec

  3. Machine-to-machine
    Example:
    Sensor data from one machine (e.g. thickness of paper) directly affects other machines (e.g. controlling run speed, tension).


Further considerations  - like brown field implementations, fast POCs, and different IOT scenarios in one place


The decision between cloud and edge (or shop floor) depends also on a number of other border conditions:

Cloud is a great accelerator to get the hardware and software ramped up very quickly and cost efficiently - especially for POCs

 

You may already have SAP MII running and connected to the operational technology and historian, and you already run analytics and KPIs for the shop floor operators in SAP MII on-premise. It may be a lot smarter then to go for on-premise SAP MII enhanced with ESP and HANA.

 

If you consider to run predictive quality and predictive maintenance in parallel, it obvious make sense to do this on the same infrastructure - especially if the same sensor readings like e.g. vibrations are relevant for both. So there may be multiple indicators to take into consideration.

 

In the end, "it depends". For some scenarios it's 100% clear, for others it requires more evaluation.

Would you have other "clear candidates". or "tricky - could be both" to add to the collection?