A recent conversation with John Murphy, the former chief mechanical officer at CSX Railroad has actually made me significantly re-think the content of this blog, and this will be considered part I - where I state that digital twin technology is still "hype" and may never realize full value for some industries. However, John gave me reason to write part II about the substance of this topic and this topic may not just be hype for the industries I mention below.
Many of you probably haven't heard the term "Digital Twin." My apologies to those that have. For those of you that haven't, here are a few articles to get you up to speed:
Here is also a 16 minute video from GE Minds & Machines:
In just watching the first 2 minutes of the video, it is hard not to be impressed with what you see. I know I was. Toward the end of the video, you could tell that wasn't a truly "live" demo as things were popping up on the screen before the speaker asked for them. So, I am not sure I believe that the technology is THAT far along yet. Still impressive, nonetheless.
However, I have been doing a fair amount of thinking about this topic and talking with some really smart people to see if this is really something that will pick up steam in the near future and if so, where and when?
The answer I have come up with is yes and no and very soon and maybe never, respectively. That is a solid "wishy-washy" answer huh? Well, let me explain. I say "Yes and soon" because I think there are places where digital twins will be excellent for diagnosing issues on machines. I say "No and never" because I think other applications will require so much effort to build and maintain the models that they will not provide the value needed.
To clarify that statement, if you read the post above from Andy Bane of Element Analytics, then you will see that he believes there are several types of digital twins. My thinking really involves what he describes as "Simulation Twins," as to me, as I believe this is really what a digital twin should be. However, his point is duly noted and I will be reading parts II and III of his blog on the topic.
If you do much research on this topic, GE will almost always come up in the top of the search engine. Is this because of good marketing and Google AdWords? Maybe, but I think a lot of it is that they are pioneering this research and this approach for IIoT applications and it makes perfect sense that they are pioneers. They are manufacturers of large, capital intensive equipment that gets made according to certain specifications and are shipped across the world to be run under different conditions. By that, I mean they make things like:
I think all of these types of assets are excellent candidates for a digital twin. The reason being is that design specifications are well known and easy to get to. The OEM also happens to be in the data collection and data analytics space, these pieces of equipment are often sent out by the multitudes, so the cost of building the digital twin is relatively low, and the fact that many are built makes getting data from lots of like assets is relatively easy to do. Not only that, the OEM is providing the correct sensors and instrumentation required to do a good job of monitoring the equipment.
Now, let's switch to more complex industrial assets that are not as repetitive, except in name only. Let's take a whole power plant, a paper mill, a specialty chemical plant, a steel mill, or any other heavy industry. Heck, let's just take one asset like:
No two of these pieces of equipment is built exactly the same. There is a wide range of available instrumentation, control systems, and data collection systems on these pieces of equipment. These types of assets are of all different vintages and some have been upgraded mechanically or electrically, have become more fully instrumented, etc. So, there is likely no other like asset in the world like it, like there would be for locomotive engines, turbines, jet engines, etc. Those types of assets tend to stay more standardized, although I have been educated to the fact that no two locomotives are exactly alike either, but they are much more similar than a paper machine.
So, what you are left with is there will be a lot of work to build and maintain a digital twin for a lot of "one off" industrial assets and likely no one will be able to leverage the learning of a fleet, since these are "one off" assets. Also, the impetus of the digital twin build out and maintenance falls on the end user versus the OEM. Who is going to build and maintain a digital twin model of a 35 year old paper machine that has constantly been tweaked and improved over the years to adjust to the market?
I spoke with a representative from a company that I am working with who does some modeling for industrial customers in heavy industries. She told me that customers seem to be perfectly OK with paying for modeling on a greenfield engineering project, but if you look at a "brownfield" upgrade, they are much less inclined to do so. Why is that? Neither of us were sure. I am a believer in modeling and simulation can definitely be extremely important to industrial companies, even in "brownfield" applications. I even wrote about it back in 2012 while working for a different company and different type of service business.
So, if customers don't want to spend money for simulation on existing assets when they are doing process or control system upgrades and if they aren't even looking at the data produced by these assets proactively, then how are they going to take the time or pay someone to build and maintain a digital model of their process? Who is going to do it? The engineers working for manufacturing companies that I know with the technical talent to build a model like this don't have time to do so. I will be honest, it is a struggle at times to get people to proactively use data, which IS a strategic asset for these industrial companies. So, now we are expecting people who struggle with so many distractions in their day, who have more to do than they can do, and who often struggle to step back and take a proactive and strategic look at their processes to either build themselves or pay someone else to build a digital twin for them? No offense, but I don't see that happening.
When I mentioned this concept to John Murphy, he gave me a rebuttal that is making me rethink this position and I will cover that position in part II of this blog.
So, digital twins certainly have their place. If you have an asset that is likely to have a number of "like" assets elsewhere in the world, then you might be a candidate for using a digital twin on that asset. If you have a lot of "one off" assets, I think you will struggle to find the value of a digital twin. I would love to have others people's thoughts on this topic ahead of part II of this blog.