News and Views, Volume 55 | Digital Twins – Concept, Uses, and Adoption in Industry

By:  Sarbajit Ghosal, Dick de Rover, and Abbas Emami-Naeini



Digital Twins are dynamically synchronized digital representations of physical equipment or systems. This technology is emerging in the power generation industry and assists with early detection of potential failures, failure accommodation, optimized maintenance schedules, development of next-generation equipment, and workforce training.


SC Solutions has decades of experience with technology that powers devices in your pocket and on your desk and continues to be an industry leader in providing process control solutions to the semiconductor industry. SI Solutions brings together the combination of SC’s controls expertise with that of Structural Integrity’s, modeling expertise and highly capable AIMS platform cyberinfrastructure, cultivating the total package to handle the development of digital twins in critical infrastructure.

Origin of the Digital Twin Concept

On April 13, 1970, while 210,000 miles from Earth, the three astronauts in Apollo 13 were startled by a loud bang that shook their tiny spacecraft. Astronaut John Swigert immediately messaged the NASA Mission Control Center: “Houston, we’ve had a problem here.” One of the two oxygen tanks had exploded catastrophically, damaging the other tank and thus putting the astronauts in extreme danger. The mission had to be aborted.

NASA engineers and scientists in Houston worked feverishly around the clock to devise a way to bring the astronauts back safely. They were assisted by 15 simulators used to train astronauts and mission controllers in every aspect of the mission, including multiple failure scenarios [1]. These simulators, made up of high-fidelity models, had been developed at NASA in the 1960s as “living models” of the mission [2]. They were controlled by several networked computers, e.g., four computers for the command module simulator and three for the lunar module simulator [1]. By utilizing these simulators and real-time sensor data from the spacecraft, Mission Control devised a successful strategy to guide the astronauts back to Earth safely.

While the term “digital twin” was coined later, the Apollo 13 mission is widely recognized as the first application of this technology, where a digital version of a physical system was updated with sensor data which was then used to run simulations to test potential solutions to troubleshoot a complex, high-stakes problem in real-time.

What Exactly is a Digital Twin?

Claims of using digital twins to solve various problems and marketing supposed digital twin products have proliferated over the past seven years. The term’s use to describe virtual representations of all sorts of assets, ranging from cities to racing cars, has led to considerable confusion. Experts from academia, industry, government agencies, and standards organizations have published definitions describing the key features of digital twins to mitigate confusion [3]-[5].

Since the definition often gets bogged down in semantics, it is preferable to identify the three primary parts that constitute a digital twin (DT). They are the physical object or process and its physical environment, the digital representation of this object or process, and the communication channel between these two that helps maintain state concurrence of the digital representation even as the state of the object or process changes dynamically. This communication channel transmits sensor data and state information and is called the digital thread. It is noted that a static model of a system or process cannot be a DT. A dynamic model whose parameters are not updated to reflect changes in the physical counterpart of the model also cannot be a DT.

The International Organization for Standardization (ISO) adopted a concise yet complete definition of a digital twin in 2021. The standards document on the digital twin framework for manufacturing (ISO 23247 [5]) defines a digital twin as a “fit-for-purpose digital representation of an observable manufacturing element with synchronization between the element and its digital representation” [6].

Whether maximizing machine performance or preventive maintenance, a clear goal for the twin is necessary for selecting the states of interest and a corresponding model of sufficient fidelity.

A digital twin of the next-generation machine, a digital twin prototype (DTP), incorporates its physical twin’s design specifications and engineering requirements. The DTP is valid in the design phase before investing resources to build a hardware prototype. DTP simulations help designers decide whether the eventual prototype would meet performance specifications. Once the prototype is fabricated and operational, the corresponding DT, now updated with sensor data, is called the digital twin instance (DTI). A collection of DTIs with a standard function is called a digital twin aggregate (DTA). DTA’s may be a collection of digital twins of the same equipment, e.g., several nominally identical pumps in a hydroelectric power station, or different equipment with a common purpose, e.g., robots, conveyors, and quality inspection stations in a material handling system of a factory.

Additionally, a simulation with a DT does not necessarily have to be performed in real-time—it would depend on the application. A DT used for real-time system control must run faster than real time. However, a high-fidelity DT used for design optimization may run simulations over many hours to sufficiently probe the parameter space in its underlying models.

Figure 1. Digital twin (DT) of a power generation equipment operating in parallel with its physical twin.

DTs have three key aspects: model, data, and services, i.e., services used or provided by DTs. The software that makes up the DT of a system has different functionalities that address these three aspects. We have divided the software into six broad classes:

Six Classes of Software used in a Digital Twin System

Software implementation of models: These may be physics-based models or gray box models (a combination of physical subsystem models and input-output heuristic models) of the physical components that may be integrated to create the system DT. The physics-based models are low-order versions of complex finite-element models that run simulations faster than in real-time. The gray box models combine known physical/mathematical relationships (the system model – the “white box” part) with phenomenological relationships or black-box models such as artificial intelligence/machine learning (AI/ML) that replace physics too complex to be modeled or overlooked. One example of gray box models is surrogate models such as Gaussian process models and physics-informed machine learning (PIML).

Sensor data-related software: This group includes software for signal processing and noise filtering of the sensor data. The data acquisition frequency may vary from milliseconds for real-time sensor data to hours for statistically sampled measurements of attributes of a manufactured product. Depending on the number of sensors and sampling rate, the volume of data may be substantial, especially in a manufacturing application. There is also software for interacting with external databases that would organize and store the data, make them available for updating the DT, and help perform prognostic tasks. This class of software would also include the implementation of sensor fusion algorithms and data compression algorithms.

Analytical and prognostic software: This class of software provides the DT’s “intelligence” and its benefits to the user. It includes the implementation of predictive maintenance algorithms, system performance optimization, decision support, and anomaly detection. Also included is software for updating models with sensor data by estimating new model parameters or re-training machine learning networks.

Software that enables user interaction: A well-designed user interface is key to digital twins gaining wider acceptance. A DT should include tools for customizing dashboards and interactive control interfaces, 3D graphics libraries for visualization of the physical twin at different levels of detail, and reporting tools for its prognostic and related functions. Some DTs may benefit from using augmented reality/virtual reality (AR/VR) tools.

Network communication and security software: This software is part of the so-called “digital thread” that involves all aspects of securely dealing with data streaming from hundreds, if not thousands, of sensors. Tasks performed by such software would include message queuing, protocol translation, connection monitoring, API management, and, very importantly, network security. For DTs to gain trust, the intellectual property (IP) embedded in the DT and in the data must be protected against all cyber threats.

Administrative software: This group includes “everything else”! It provides software and tools for configuration and change management, requirements tracking, documentation, access control, resource monitoring, and backup.

Drivers for Digital Twin Development

The confluence of advances in four technological factors has driven the development and adoption of digital twin technology over the past decade. These factors are:

  1. The decreasing cost of high-performance computing (HPC), both at the edge (i.e., in physical proximity to the end-user or the physical twin) and in the cloud. While problematic limitations imposed by physics and manufacturing costs have slowed Moore’s Law, computational power has continued to increase through a combination of heterogeneous integrated circuit (IC) architecture, such as 3D stacking and chiplets, and chips designed for a specific use, such as graphics processing units (GPUs).
  2. The proliferation of sensors and sensor networks (sometimes called the Internet of Things or IoT) enables individual sensors to acquire, flow and store data. Data analysis makes it possible to monitor a variety of system attributes, which, in turn, allows the digital twin to keep up with changes occurring in its physical twin.
  3. Availability of software tools enabling faster development of more complex models. Modeling techniques have been developed to use different models, including physics-based, data-driven, and machine learning (ML) models, to create a more comprehensive and accurate digital representation of the physical system. Merging various modeling approaches helps capture a more precise view of the physical system by leveraging the strengths of each model type. Commercial modeling software such as ANSYS also provides tools to develop surrogate models, proxies for high-fidelity physics-based models, and speed up model simulations [7].
  4. The arrival of large language models (LLMs) developed in the field of generative artificial intelligence. Before the release of ChatGPT to the public in November 2022, the role of AI in digital twins was primarily in using supervised machine learning for surrogate and data-based models. LLMs have advanced “embedding” capabilities, i.e., they can significantly compress data (both numeric and text) while retaining essential information. For example, in a manufacturing setting, LLMs can organize data from maintenance logs, equipment images, and operational videos and make them available in a DT.  Maintenance logs often have valuable information related to system failure diagnostics and health maintenance that would add to the DT’s capabilities. AI is expected to play an essential role in the future of digital twin technology.

The fast-paced progress made in the above technologies makes it possible to transform digital twins of complex systems from merely a nebulous concept to a valuable technology that can be implemented once a few hurdles (such as standardization and data sharing) are overcome. Figure 2 attempts to show how the digital twin concept has evolved from solid models and offline computations to the virtual representations of complex systems being developed.

Figure 2. Factors in the evolution of digital twins.

How Can Digital Twins Be Useful?

The holy grail of digital twin technology derives from its ability to monitor the health of its physical twin, and the benefits include the following:

Early detection of potential failures: While sensors in the physical twin can monitor the system’s local state in the proximity of the sensors, the digital twin’s states act as virtual sensors and effectively scan the state of the entire system and can detect anomalous behavior. When the DT incorporates reliable degradation models (e.g., heater degradation or crack propagation), it can predict potential failures. The process cycle may then be ended in an orderly manner to repair or replace the part without any damage to the system that may result from a catastrophic failure of the part.

Failure detection and accommodation: The digital twin can be a valuable tool in case of a component failure in equipment. There are different ways to perform such root cause analysis. One way is to use physically meaningful model parameters continually monitored by sensor data estimation. If a parameter value strays outside a specified range, the failure is related to the component associated with that parameter. A second method uses a bank of Kalman filters to detect anomalies. The second article in this series will examine failure detection for sensor and actuator failures in a rapid thermal processing (RTP) system in greater detail.

As an example of using DT for failure accommodation, if a temperature sensor fails in the RTP system, the DT’s estimate of the system’s temperature near the sensor (one of the DT’s states) can temporarily serve as a virtual sensor. The process can continue until the faulty sensor is replaced during regular maintenance.

Optimizing maintenance schedules: Currently, scheduled maintenance of equipment is more frequent than needed to avoid unplanned downtime. The ability to foresee some potential problems down the road allows a factory to implement predictive maintenance strategies to reduce cost by eliminating unnecessary maintenance.

Develop Next-Generation Equipment: The digital twin of an existing asset may be modified to help speed up the development of next-generation equipment. Simulations run with the latter are very helpful in determining whether the design would meet the desired performance goals. Design changes are fast and inexpensive to implement and test in virtual space, and they can help ensure that the prototype built would meet all the requirements. SC has used this approach with its equipment models, which are components of the equipment DT, to help its customers design and build next-generation equipment.

Workforce Training: Since the roots of digital twins go back to NASA’s simulators for training astronauts, it is not surprising that DTs are finding a role in the education and training of the industrial workforce. Here, DTs can provide an immersive learning experience and practice with virtual control of tools to run real-time simulations, often aided by virtual reality accessories. Like other digital educational tools, DTs have the advantage of offering customizable learning, distance learning, and a safe environment without any accidents resulting from incorrect operation. Finally, DTs can be used for scenario-based training dealing with various operational conditions, equipment failures, and emergency response training. While the prognostic applications of digital twins require very frequent updating with sensor data, the DTs for other applications need significantly less updating.

Potential Applications for Digital Twins in Power Generation and Other Critical Infrastructure

The digital twin paradigm offers promise in the energy industry where a DT is developed and maintained to identify changes in the system that helps detect anomalies, make maintenance decisions, or perform root cause analysis of failures.  A finite element (FEM) model of a structure with a crack which is periodically updated with measurements of the crack dimension may be considered to be a DT of the structure whose purpose is to monitor crack propagation. One may scale up such models to larger structures, e.g., large components of energy systems such as gas turbines [8].

The application of DT technology to combined gas turbine, wind turbine, solar, and nuclear power plants are expected to increase in the years ahead with several application areas in the nuclear industry [9]. These include design, licensing, plant construction, training simulators, autonomous operation and control, failure and degradation prediction, physical protection modeling and simulation, and safety/reliability analyses [10].

SI’s expertise in FEM modeling, material degradation, and lifetime prediction models, together with the AIMS development team’s expertise in cyberinfrastructure, is well suited to building and maintaining a DT of an energy system or some other critical infrastructure and using the DT for preventive maintenance and other applications.  DT is an evolving technology, and it may not be possible to fully automate the model updating process. Hence,  the software as a service (SaaS) model may become the norm for DT products. SI is optimistic about the technical aspects of DT Technology and the opportunities to leverage these tools in supporting our clients.

Figure 3. Digital twin concept for a nuclear power plant [10]

Power Gen Applications

With the emergence of Digital Twins in the power generation industry, our teams are able to use the synchronized digital representations of equipment to assist with early detection of potential failures, failure accommodation, optimized maintenance schedules, development of next-generation equipment, and workforce training.

Our staff are positioned to support digital twins’ development, coinciding with SI’s modeling expertise and highly capable AIMS platform cyberinfrastructure, cultivating the total package to handle any digital twins’ needs. The AIMS Digital Solutions platform is integral to our mission of providing the best-in-value, innovative, fully integrated asset lifecycle solutions. Digital products paired with our expertise in Engineering, inspections, and analytics help achieve a holistic asset management approach to our clients.

References

  1. S. Ferguson, Apollo 13: The First Digital Twin, April 14, 2020. Available at: https://blogs.sw.siemens.com/simcenter/apollo-13-the-first-digital-twin/
  2. B. D. Allen, Digital Twins and Living Models at NASA. Keynote presentation at ASME’s Digital Twin Summit, Langley, VA, November 3 2021 Available at: https://ntrs.nasa.gov/api/citations/20210023699/downloads/ASME%20Digital%20Twin%20Summit%20Keynote_final.pdf
  3. L. Wright and S. Davidson, “How to tell the difference between a model and a digital twin,” Adv. Model. and Simul. in Eng. Sci., 2020, 7:13.
  4. The Digital Twin, Ed. N. Crespi, A. T. Drobot and R. Minerva, Springer, 2023.
  5. ISO 23247-1: Automation Systems and Integration – Digital Twin Framework for Manufacturing – Part 1: Overview and general principles. International Organization for Standardization, Geneva, Switzerland, 2021.
  6. G. Shao, S. Frechette, and V. Srinivasan, An Analysis of the New ISO 23247 Series of Standards on Digital Twin Framework for Manufacturing, Proc. of the ASME 2023 Manuf. Sci. Eng. Conf., MSEC2023, June 12-16, 2023, New Brunswick, NJ, USA.
  7. M. Adams, et al., “Hybrid Digital Twins: A Primer on Combining Physics-Based and Data Analytics Approaches,” in IEEE Software, vol. 39, no. 2, pp. 47-52, March-April 2022.
  8. D. de Roover, Possibilities and Challenges in Developing a Digital Twin for Rapid Thermal Processing (RTP), APCSM Conference, Toronto, Canada, 2024.
  9. N. V. Zorchenko, et al., Technologies Used by General Electric to Create Digital Twins for Energy Industry. Power Technol. Eng., 58, 521–526, 2024.
  10. U.S. NRC, Digital Twins. Available at: https://www.nrc.gov/reactors/power/digital-twins.html#reports

News and Views, Volume 55 | Nonintrusive and Robotic Solutions for Tank Asset Management

By:  Jason Van Velsor

Nuclear plant aging management programs require periodic inspections of liquid storage tanks. Traditional inspection methods can be disruptive, requiring tanks to be drained to provide personnel access. SI has developed innovative solutions, including screening techniques that can identify degradation from the tank exterior, and submersible robotics that perform comprehensive NDE without draining. Although initially developed for nuclear applications, these technologies can be employed at conventional power generation, petrochemical, and municipal utility facilities.

Nuclear Industry Guidelines

The nuclear industry established guidelines for integrity management of underground piping and tanks in the early 2000s with the publication of NEI 09-14. More recently, additional constraints have been imposed for plants pursuing life extension, especially for sites applying for subsequent license renewal (SLR) to extend permitted operation from 60 to 80 years. These new requirements for outdoor and large atmospheric tanks are conveyed in the form of specific guidelines for aging management programs (AMPs) within NUREG-1801, Revision 2 and NUREG-2191, Revision 1.

The guidelines within the NUREG documents apply to:

  • All metallic outdoor tanks constructed on soil or concrete.
  • Indoor metallic storage tanks with capacities greater than 100,000 gallons, designed for internal pressures approximating atmospheric conditions, and exposed internally to water.
  • Other indoor metallic tanks that sit on, or are embedded in, concrete, where plant-specific operating experience reveals that the tank bottom (or sides for embedded tanks) to concrete interface is periodically exposed to moisture.
  • For utilities with tanks meeting the above criteria, license renewal commitments generally necessitate performing examinations under one of the following three categories:
    • Inspection of the bottom 20% of walls for wall loss and cracking.
    • Inspection of the outer two feet of the floor plates for pitting/cracking.
    • 100% inspection of the bottom floor plates.

Conducting floor inspections using traditional approaches that require draining and personnel entrance into the tank can be undesirable, given the potential impact on operations, as well as the possibility for certain tanks to contain radiological content. For this reason, many utilities have begun pursuing alternative solutions to safely and accurately perform tank inspections.

Traditional Approaches

Historically, utilities have emptied and entered tanks to conduct manual floor inspections. The scope of these examinations can range anywhere from visual to full volumetric inspection of the tank floor using electromagnetic or ultrasonic techniques. More recently, utilities have attempted to utilize robotically deployed examination methods to avoid emptying and entering tanks. Obtaining large-area NDE coverage of tank floors with robotic methods has usually fallen into one of two categories:

  1. Relatively simple robotic systems that deploy traditional NDE sensors for localized measurements, which may take an impractical amount of time to reliably obtain the required coverage, or…
  2. Relatively complex robotic systems that deploy a large quantity of traditional NDE sensors to obtain the required coverage more quickly, but that are large, complex, and often expensive to deploy. 

Deploying any robotics in tanks is a challenging endeavor, with many practical factors that affect the ease and success of implementation. Several of these factors include:

  • Accessibility – there are limited access points to the inside of a tank; they are often on top of the tank, and just large enough for a person to fit through.
  • Visibility – maneuvering within a liquid-filled tank often relies on optical methods, which are impaired or even ineffective in murky or opaque liquid.
  • Navigation – with poor visibility, continuously tracking the position of a robot within a tank (and hence, the location of acquired data) can be unreliable.
  • Cleanliness – sediment present in the tank can impair data acquisition and cause additional visibility issues if agitated by the robot.
  • Geometry – getting complete coverage at the tank edges and around internal features can be challenging or impossible given the size and limited maneuverability of some robotic systems.
  • Tank Size / Inspection Time – for large tanks, obtaining complete coverage of all in-scope surfaces may require extended scanning time, which can impact operations and delay return to service.
  • Liners/Coatings – tanks that have been lined with thick coatings such as fiberglass or carbon wrap can render traditional NDE methods useless.

Figure 1. GWPA from Tank Chime

SI’s Approach

SI offers a suite of engineering and inspection solutions that are designed to help our clients meet their examination commitments while minimizing the associated cost, time, and impact to operations. We work with our clients to determine the required scope of any necessary examinations, based on their Aging Management Program and license renewal commitments, and present a customized inspection approach. Where possible, inspection solutions that can be conducted from outside of the tank are prioritized over approaches that require deploying equipment in the tank. These approaches may include performing pulsed eddy current testing to examine tank walls through insulation or employing guided wave phased array (GWPA) from the tank chime plate to examine the outer annulus of the tank floor (Figure 1).

In situations where putting equipment inside the tank is unavoidable, SI has developed a robotic solution that uses a novel technological approach to provide rapid, 100% volumetric coverage with a range of NDE sensors deployed on a relatively basic robotic platform. Rapid, large-area volumetric coverage is obtained by mapping the tank floor with GWPA testing to identify critical areas (Figure 2). These critical areas can then be investigated using high-resolution, non-contact methods, such as electromagnetic acoustic transducers (EMATs) for UT thickness measurements or SI’s dynamic pulsed eddy current technology, SIPEC™, for dirty or lined tanks.

Figure 2. GW Phased Array Tank Floor Inspection

From a deployment perspective, SI’s robotic system is designed to fit through existing tank access points, can scale carbon steel walls, and can quickly switch between a range of sensor types. Additionally, SI has incorporated a proprietary acoustic vehicle positioning system, where sound pulses track the absolute position of the robot at all times, ensuring precision results and the ability to accurately relocate and rescan specific areas for follow-up activities or future inspections. The acoustic positioning system uses a transmitter placed on the robot and a series of receivers placed on the exterior tank wall to track the movement of the robot. With this approach, positioning is not reliant on visual or other optical methods that can be confounded by cloudy, murky, or otherwise opaque liquid and is not subject to encoder error or drift.

Engineering Support Services

Beyond inspection, SI provides integrated engineering support that couples directly with tank inspections. SI’s expertise includes disposition of findings, detailed evaluation of any anomalies, and optional integration with our piping and tank asset management database, MAPPro™. SI engineers are adept at employing detailed FEA models and/or fracture mechanics techniques to assess the acceptability of any observed flaws/defects. For time-critical inspections, engineering handbooks can be developed before the examinations to provide real-time disposition of any findings. Loading inspection results into MAPPro enables risk-ranking utilizing time-proven algorithms, thereby informing future programmatic actions.

Using the advanced deployment and inspection technologies highlighted herein, SI is able to comprehensively inspect and disposition findings from tanks of various sizes while reducing schedule risk and ensuring accurate results.

Tank Inspection NDE Technologies

  • Guided Wave Phased Array (GWPA)
    • Used to rapidly inspect large areas such as a tank floor.
    • Provides 100% volumetric coverage in a minimal amount of time and with minimal surface preparation.
    • Identifies critical areas that should be investigated further using more precise examination technology. This follow-up is completed by switching probe heads and conducting targeted quantitative exams.
    • Can also be applied from the chime plate on the tank exterior to examine the outer annulus of the tank floor.
  • Electromagnetic Acoustic Transducer (EMAT) Ultrasonic Thickness Testing
    • A non-contact (up to 0.25 inches of liftoff) volumetric examination that provides quantitative wall thickness for either carbon or stainless-steel tanks.
    • Electromagnetic sensors generate UT for thickness measurements and are electromagnetically coupled, thus eliminating the need for couplant or close contact with the test surface.
    • This type of testing is helpful for corrosion mapping where surface prep is not possible or costly or if there is a coating/liner present; additionally, it allows for remote robotic inspections, including those where the probe is submerged.
  • SIPEC™ Electromagnetic Thickness Testing
    • Proprietary non-contact volumetric examination that works through internal liners and sediment (up to several inches) targeting either carbon or stainless steel.
    • Employs a proprietary dynamic pulsed eddy current measurement technique for rapid scanning.
    • Higher liftoff, lower resolution option when compared to EMAT UT.

Download Volume 55

View Video Contact Us