Layers of Analytics: Edge to Cloud for Better Results

#News(IoTStack) [ via IoTForIndiaGroup ]

Cloud or Edge Analytics
Cloud analytics refers to analytics performed through cloud computing; by software running in the cloud, meaning on a virtual machine on a server in a datacenter somewhere, typically by a cloud service provider like Microsoft Azure.

Edge analytics refers to analytics performed in software running on a server at site in the plant, on premise “on-prem”, or as firmware in hardware such as in a transmitter. This would include for instance analytics software tools you already have for the turbomachinery in your plant.
The first level of analytics is often done in the transmitter on a single variable, close to the sensor, in real-time. This first level of analytics is on a single variable such as vibration, pressure, corrosion, or flow etc. As such, it is more of component-level analytics for a single bearing or a flame; not the whole pump, burner, or furnace. Analytics of whole pieces of equipment involving multiple variables is covered higher up in the architecture.

That is, the results from first level point measurement analytics on the plant floor propagate up to the next level of analytics of a whole machine as a single simple piece of information, not a whole blob of data, which in turn percolates up to plant unit analytics, that then goes further up the chain into analytics for the plant, and ultimately enterprise dashboards in the boardroom. As raw Data propagates up it becomes Information, Knowledge, and Wisdom as per the DIKW pyramid model. A massive amount of low-level detail is step-by-step distilled into increasingly higher level overview suitable for the decision makers at each level. This is entropy of information. The maintenance manager can see which equipment needs service. The commodity trader can tell if the plant is in a condition to accept high TAN opportunity crude or shale oil. Yet a reliability engineer can see vibration spectrum with the finest level of detail. The same result cannot be achieved simply by trying to find correlation between every variable and record in the plant.

Read More..