By Pat Dixon, PE, PMP
President of DPAS, (DPAS-INC.com)
"When I use a word, it means whatever I want it to mean."
This quotation of Humpty Dumpty from Lewis Caroll's "Through the Looking-Glass" is referenced in some articles I recently perused. The first article is "How to tell the difference between a model and a digital twin" by Wright and Davidson (https://doi.org/10.1186/s40323-020-00147-4). This article begins with:
"Digital twin is a term that is being used for a wide range of things across a wide range of applications, from high value manufacturing and personalised medicines to oil refinery management and risk identification and mitigation for city planning. For some of the definitions, the reason why "twin" is used has been lost. The danger of this variety and vagueness is that a poor definition and explanation of a digital twin may lead people to reject it as just hype, so that once the hype and the inevitable backlash are over the final level of interest and use may fall well below the maximum potential of the technology."
A particular example of a poor definition cited in this article in this one by LMS research in which they state:
"While we see many definitions of "Digital Twin," LMS Research keeps it simple: A Digital Twin is an executable virtual model of a physical thing or system."
Wright and Davidson reply:
"This definition merely renames technology that has existed for many years, leading many engineers wondering why they are being told that a practice they have been successfully employing for decades is a new and vibrant thing."
This is precisely what I ran into when I first heard the term digital twin. How is this different from what I have been doing my whole career?
When I wrote my article "Digital Twin" in this newsletter in December 2019, I described this problem. I proposed some hypothetical definitions of digital twin, realizing that amidst all the uncertainty it is hard to know what is right. It is quite a bit like the imaginary world of "Alice in Wonderland".
If you ask IBM, they say the definition of a digital twin actually is vague, and should be. I have spoken with Lisa Seacat DeLuca (Director Digital IoT Transformation & Digital Twin at IBM) about this and she says the problem is that a digital twin is literally any digital representation of a physical object. If you have an HMI representation in your control system of a Kamyr digester, a dataset of that digester from a historian, or a process model of that digester, they are all digital twins. Lisa says that to be precise, you have to use more direct terms. She would call the process model a simulation twin, the HMI an operational twin, and the dataset a data twin. It is a bit like the difference between the words "vehicle" and "jeep".
Fortunately my friend Dr. Russell Rhinehart has 2 articles "Understanding the digital twin" parts 1&2 in the online publication Control (see links below) that gives further definition. Specifically, Russ describes digital twin as a simulation twin, which is the most common understanding of how we are using this technology. The difference between process models that we have been using forever and a simulation twin is that the model is being adapted with online data. The idea is that to keep the model coefficients accurate, we can use online data to detect changes in the behavior of the process. This is risky. When you are using online data to adapt a model, you need to be careful that the data is valid. Noise or measurement problems will take a model that is reasonably accurate and turn it into nonsense. If this happens in an application that is critical to operation, you can be in trouble.
As we wander through the wonderland of Industry 4.0, we need to understand that terms like digital twin can mean whatever someone wants them to mean.
https://www.controlglobal.com/articles/2021/understanding-the-digital-twin-part-1/
https://www.controlglobal.com/articles/2021/understanding-the-digital-twin-part-2/
