Digital Twins: What is the hype all about?
What is a Digital Twin?
What is a digital twin? Just a fancy new buzzword? I have been trying to determine what the hype is all about!?
McKinsey & Company have provided this definition of a digital twin [McKinsey, 2024].?
A digital twin is?a digital replica of a physical object, person, system, or process, contextualized in a digital version of its environment. Digital twins can help many kinds of organizations simulate real situations and their outcomes, ultimately allowing them to make better decisions.?
Further, McKinsey provides an example: Google Maps is a digital twin of the Earth’s surface.?Another example they provide is digitally enabled supply chains that deploy digital-twin and AI technology to drive optimization and efficiency.?
?Some analysts categorize the value of a digital twin along the following four dimensions:
?A digital twin is a virtual model of a physical object, process, or end-to-end system that mimics its real-world counterpart.?Digital twins can be used for a variety of purposes, including
1.???Simulation of a real-world environment using discrete event or continuous event simulation. It enables scenario planning and what-if analysis.
2.???Predictive modeling capability to predict outcomes of intractable mathematical models.
3.???Monitoring a physical object or real-world process in real-time
4.???Maintenance capability to identify deficiencies in an existing physical object or process. Examples are identifying bottlenecks in a manufacturing facility for changing arrival patterns.
Hence the question I am posing is: is a digital twin a new over-hyped name for something process engineers and industrial engineers have been modeling for decades well before the age of AI and machine learning?
IBM goes to great lengths [IBM, 2024] to explain why a digital twin goes beyond a simulation. According to IBM
“The difference between a digital twin and a simulation is largely a matter of scale: While a simulation typically studies 1 particular process, a digital twin can?run any number of useful simulations to study multiple processes.”
?
Simulation Modeling
As a backdrop, I have a background in Operations Research, by education and training. I was pursuing my PhD in the 1980s at Purdue University. I spent three years at Purdue University in the School of Industrial Engineering. During that period, I also served as a teaching assistant (TA) to Professor Alan Pritsker in my second semester (Spring 1981) for the graduate course Simulation with SLAM (Simulation Language for Alternate Modeling) [Pritsker and Pegden, 1980].
领英推荐
I had prior experience using GPSS (General Purpose Simulation System) [Schriber, 1974] at Ohio University where I pursued my master’s degree. SLAM is a general purpose simulation language built upon discrete event simulation languages Q-GERT and GASP IV to model a wide variety of systems such as manufacturing, airport operations, transportation, communication, computer systems, military operations, health care and banking. As I recall, it consists of three major components – PROCESS view, DISCRETE event view, and CONTINUOUS event view.
The PROCESS view consists of branches that link graphical symbols called nodes such as queues and activities. Its appeal is that the development of a graphical network model of events and activities that replicates a system can be accomplished quickly. The DISCRETE event view, which is the most widely used component, consists of numerus built-in function calls as well as user generated function calls such as random number generation, event scheduling, queueing priority rules, etc. that are used to represent the real-world application. Unlike the network view this requires detailed knowledge of the programming language (FORTRAN). The flexibility to develop complex models is unlimited. The CONTINUOUS event view is a continuous simulation model consisting of a system of differential or difference equations.
The Simulation with SLAM graduate course started with simple models of manufacturing systems and toward the end of the semester the students will build complex simulation models taken from the real word. (Dr. Pritsker had a consulting company which was a source for some of these homework assignments).
The job of a TA was challenging, exhausting, and rewarding. As due dates approached for the homework assignments, there would be a long line of students outside my office seeking help to debug their computer programs. ?Alan Pritsker retired at the end of the semester, and I continued as a TA for Professor Joe Talavage in the Fall of 1981. After supporting students by debugging hundreds of computer programs, I became an expert on SLAM. SLAM and SLAM II have since been superseded by SIMAN from System Modeling Corporation, Today, there are advanced simulation models (some with 3D animation) in the commercial market such as Arena, Simio, FlexSim, and many more.
Digital Twin versus Simulation
So, when I came across this article from McKinsey and Company, I could not help but wonder how a digital twin was different from a detailed simulation that could provide answers to complex problems that were mathematically intractable? Many of the problems we simulated during my graduate days were mathematically intractable (no closed-form solutions exist) since we could relax the assumptions of the probability density functions (discrete or continuous) of arrival times, service times, complex queueing logic (e.g. FIFO, LIFO, priority queue, state-dependent routing, jockeying, etc.), and many more. ?Frequently, empirical distributions captured from real-world applications were used as input into the simulation models.
While this is not done today, simulation modeling frameworks can be enhanced to ingest data in real-time which seems to be the main difference from a digital twin.
According to McKinsey, creative product development leaders are increasingly enthusiastic about digital twins. Their analysis indicates the global market for digital-twin technology will grow about 60 percent annually over the next five years, reaching $73.5 billion by 2027! Perhaps McKinsey and IBM are promoting digital twins for their lucrative consulting business.
?
Airline Retailing and Personalization
In the world of Airline Retailing. promoted by IATA as part of the NDC initiative, personalization of offers and the entire end-to-end journey of a traveler across all touchpoints is much talked about at industry conferences. If a traveler can have a digital twin that predicts his or her actions this would be the holy grail of personalization. The human brain has over 86 billion neurons that form 100 trillion connections. But alas, replicating the human mind is a formidable task that will not be realized in our lifetimes.
?
References
1.?IBM (2024) What is a Digital Twin? IBM, https://www.ibm.com/topics/what-is-a-digital-twin#:~:text=The%20difference%20between%20a%20digital,simulations%20to%20study%20multiple%20processes.
2. McKinsey and Company (2024) What is digital-twin technology? McKinsey and Company, August 26,
3.?Pritsker, A.A.B., Pegden, C.D.(1980) Introduction to Simulation and SLAM, Wiley.
4.?Schriber, Thomas, J. (1974) Simulation with GPSS, Wiley, 1st Edition.
?
Global Cloud & Digital Transformation Leader | Product Development | Private Pilot | Kind Human
5 个月Here is a real world simulation use-case in airline industry. https://www.cio.com/article/213940/qantas-cloud-based-flight-sim-saving-millions-in-fuel.html. They use a unique way of sourcing 4000 spot CPUs that are excess capacity for a short period of time so the compute expense makes sense.
Data Analyst | Digital Twin | Electric Vehicle | Power System | Electrical engineer | Sustainability Engineer|
5 个月Exciting insights in your post! If you’re interested in further discussions on digital twin technology, I invite you to join our new LinkedIn group, Digital Twin in Engineering. It’s a great platform to explore innovations and share knowledge in this field. Would love to have you there! https://www.dhirubhai.net/groups/14546210/
Senior Mainframe Migration Architect
5 个月What this doesn't address is the fact that the digital twin is subject to the understanding of the creator. Anything which emulates or simulates another process or entity cannot in reality be expected to fully represent a real world item. There are always compromises and omissions. The example in the article of Google Maps representing a digital twin of the earth itself is a flawed comparison. Google Maps does not represent altitude normally. Therefore, it is compromising the representation of "true" distance between points. Instead, it settles for linear (or roughly linear given the Earth's shape) distance or if tasked, pathfinding using roads which also represent pre-established paths over roughly linear distances.
I see it as a useful metaphor that represents some key concepts to enable. First, a data model that encapsulates important data around a real world object. Second, the ability to measure changes of that real world object. Third, the ability to synchronize that change to the digital copy in near real time. Once you have this, you can now run both established and ad hoc analytics on aggregates of these twins, maybe establish trigger events for those real time incoming data elements, and of course potentially useful for predictive analytics/AI training. I won't disagree that we have done much of this in the past. I think the key element is the near real time linking of the real world object with the digital model. An example would be an airplane engine during their lifespan, all the cycles and measurable elements. One could query what the real status is, trigger an event over a threshold, or perform an analysis of all current engines of that class and their repair rates, etc. So "digital twin" maybe useful as a metaphor and maybe useful to sell stuff as a vendor .;-)
Data scientist / Systems architect
5 个月To me, "digital twin" is a marketing term, to sell more consulting. I've written a couple of large scale simulation systems from scratch, it's not like they've invented anything new here.