Recently, I had the pleasure of attending OSIsoft's annual User’s Conference. It is always a great event and there is so much to take in and I must say I always come home with brain overload. I will probably write several blogs based on observations, conversations, and/or presentations from the event over the coming months.
The first one that really struck me though, was Pat Kennedy addressing what he believes the IIoT will look like. He spends some significant time describing how things will “stratify.” I definitely agree with him that the IoT is not likely to become “100 trillion sensors spurting data up to the cloud.”
My belief is that there will be no one single architecture and no single system that will provide all of the capability that an industrial company will want. Sometimes, a combination of Edge, on-premise, centralized, and cloud systems will be required for a company to be successful.
Let’s take a transformer as an example. Let’s say this transformer is getting dangerously close to failure. If I am responsible for this transformer, I want to know as soon as possible of imminent failure and want up to the second (potentially) information. So, I would be looking for a solution either at the edge, on premise, and maybe even a central system, depending on data throughput of the system. However, the closer to the device, the better. The closer the analytics are to real-time, the better also.
In most cloud systems, the goal is to use the tremendous computing power of a distributed system to crunch extremely large data sets. Often, the goal is to look at more long term trends. Maybe I want to look at aggregate health of all transformers of the same make/model/type/size and understand how behavior of customers of those whose power is drawn from that type of transformer affects the longevity of the transformer. Maybe there is some other type of analytic that I need to run that needs me to look wider across the data set (what is attached to the transformer and how it is loaded, customer behavior, weather, location, etc.) and look much deeper (I want to look at this data for the previous 5 years and compare it to actual maintenance performed on the transformers). The reality is, the edge and on premise system won’t give me all of the context I need; nor will it likely have the computing power I want. A central system might, but I would expect someone would leverage a cloud system to crunch this type of data. Speed of needing to make decisions isn’t critical, because I am looking long range, not short range, as in the case of the imminent transformer failure.
I think one also has to consider “what happens if the cloud system goes down?” like Amazon Cloud Services did recently. It is one thing if it goes down when doing long range planning, but what if I am running analytics in the cloud that I am using to keep my plant or my enterprise running? That is a whole different story. I would also have to consider the price point versus the risk of the system going down and how much control I would want to have in getting the system back up and going. Sometimes, cheaper isn’t better, but sometimes it is. Everyone will have to consider this.
This is why that I too agree that the IoT and IIoT will “stratify” based on needs. Having a ton of sensors just start connecting to the cloud for analytics purposes in my opinion is not only impractical, but I also don’t believe it is desired because of the various business needs.
It is going to be fun to watch for sure.