How Much Control Goes to the Cloud?

None. That’s what any reasonable automation engineer will tell you.  Or at least, that’s what he or she would have said a few years back.  Today, with growing interest and acceptance of Industrial IoT and Industrie 4.0, the message is starting to change.  People are talking about the possibility of doing supervisory control through cloud-based systems.  For example, the cover theme of last month’s issue of Control Engineering was “Optimize controls via cloud software“, and it included articles from MESA, Honeywell, and Skkynet related to cloud-based control of manufacturing.

Our contribution was a short article titled: “Control in the cloud: How much?”  In it we point out how users and suppliers are becoming more sophisticated in their understanding, and are starting to look at edge computing as an alternative to cloud computing.  We encourage plant engineers and managers to get the best of both approaches by putting computing power where it is needed.

We identify four areas where real-time processing can take place:

Device: Adding compute power to sensors and other equipment can reduce the amount of data sent to the plant and cloud, and also simplify upstream processing.

Plant: This is where most industrial computing has taken place traditionally, and where new computing tasks can support IIoT to improve efficiency.

Gateway: Processing and filtering data at the gateway can support installed SCADA and HMI systems by serving as an intelligent interface to the cloud.

Cloud: By reducing, managing, and enhancing the quality of the data at or near the source, cloud computing resources become more effective.

As you might expect, what you gain from using cloud services for industrial control depends on how you manage the data you send to the cloud and what you need to get back in return. The article explains how choosing the right level for each computing task can reduce costs and generate a quicker round trip time for any data or analytics that come back to the plant.

Balancing the data load at each step in the process seems to be the key to a successful implementation, and adding edge computing where it is needed looks to be the thing that pulls it all together.