Stacey Higginbotham‘s Stacey on IoT site had a live event last week, How to make machine learning at the edge work. In one of the sessions, Which Edge and Why?, Emerson’s Peter Zornio joined a panel hosted by Stacey along with John Deere’s Julian Sanchez and Tufts University’s Karen Panetta to discuss the benefits and challenges in this information architecture.
After opening introductions, Stacey asked Julian to define the edge. For John Deere tractors, one example is the the sub-second decisions being made by the controllers on the planter to optimize the planting of the crop for maximum yield.
Stacey turned next to Peter to ask where Emerson’s edge is. Peter explained that if tractors are the edge for John Deere, then manufacturing equipment is the edge for users of Emerson automation technology. More than just automating manufacturing facilities, these edge devices are increasingly analyzing and diagnosing what’s happening in these processes.
Manufacturing has been using “the edge” for a very long time with controllers running logic to maintain safe & efficient operations. More and more, the sensors are providing data for non-control functions such as predictive maintenance and energy optimization.
Peter noted that people have different ideas on the cloud from third party hosting on the internet, to on-premise data centers and even servers in the manufacturing area. It really boils down to doing the data processing, computing and analysis where it makes most sense based on many factors.
Karen explained that edge computing is really determining the best way to distribute computing to make the data and processes always available in a fast, efficient and secure way. This often means capturing, processing and analyzing it closest to where it is generated. For applications requiring immediate processing such as visual processing, the edge needs to be at the sensor.
Stacey asked Peter about how Emerson thinks about the cloud in relationship to the edge computing. Peter shared that the cloud is for those bigger problems. The data collected and analyzed at edge device boil up results to cloud applications to understand overall manufacturing performance or comparative analysis across manufacturing sites. This higher-level analysis can also be used to fine-tune the models running in the edge devices.
Peter described operational analytics at the edge as often being based on first principles kind of models, such as diagnosing what’s happening inside a pump by understanding the flow curves, hydraulic models and key process variables to create the diagnostic. Machine learning technology is used where there is not enough data available for good first principle models.
An example he offered is a wireless acoustic sensor that can be using for many things—determining leaks in a pipe or if a steam trap is working correctly or if a pressure relief valve has opened. Machine learning is used to train the models for the sensor to recognize these very different applications.
Watch this video for this very engaging discussion among these experts. You can also learn more about the Industrial Internet of Things, edge and cloud computing in the Industrial IoT and Plantweb digital ecosystem sections on Emerson.com.