MIT-led project incorporates hard-to-measure data to develop digital twin for drilling risers

By Stephen Whitfield, Associate Editor

Vortices, or closed circular flows of air or water, are the primary cause of motion for marine drilling risers. Understanding when these vortices may occur, as well as the potential force of vortex-induced vibrations (VIVs), is important in predicting potential fatigue damage to the riser. However, VIVs are still difficult to anticipate due to the geometric complexity of the riser configuration, as well as the sheared and unsteady nature of ocean currents.

Recently, researchers from the Massachusetts Institute of Technology (MIT) and Brown University have been working with a consortium of operators, drilling contractors and service companies to create a digital twin applicable to marine risers, DigiMaR. The twin was trained using a variety of data like field sensor data, computational flow dynamic simulations and existing databases, and has semi-empirical codes that incorporate various assumptions about the data to predict VIVs. The result is an accurate reconstruction of riser motion and prediction of fatigue damage, which can help to optimize the placement of sensors on the risers and extend their field life. 

“Artificial intelligence has become an essential tool for the design and operation of complex systems, and digital twins provide an unprecedented capability to explore the parametric space,” said Michael Triantafyllou, Professor of Mechanical and Ocean Engineering at MIT. The digital twin utilizes a parametric representation of the data characterizing VIV behavior in a series of databases, which Dr Triantafyllou said would provide a comprehensive view of how VIVs affect riser motion. 

The twin takes streaming data from physical marine risers and their environment and feeds it into AI predictive models and neural networks. Once the data is assessed and interpreted, the twin outputs information characterizing the magnitude of parameters like the riser’s displacement, velocity and acceleration. Those outputs can then be used to predict the potential fatigue damage of the structure. Additional machine learning algorithms within the digital twin also predict optimal times for servicing of the riser. 

During the 2023 Offshore Technology Conference (OTC) in Houston on 2 May, the American Bureau of Shipping (ABS) announced it had joined the project to provide third-party classification services for the design of the digital twin, which is now being incorporated into the workflows of major operators. “As a safety-focused organization, we’re able to share the insight we have learned with MIT to help support the adoption of new tools and systems that can improve safety and performance,” said Patrick Ryan, ABS Senior VP and Chief Technology Officer. 

The consortium includes operators Petrobras, ExxonMobil and Shell, as well as drilling contractor Saipem. 

Digital twin design process

Algorithms used in the digital twin were developed under a three-pronged approach. First, active learning algorithms were used to enhance the capabilities of the codes typically used to build a model of a marine riser. Digital twins of marine risers usually rely on the data captured from the field that measure the motion of fluids and the forces, such as VIVs, acting on solid bodies immersed in fluid, also known as hydrodynamic data. This data is stored into a database, while a separate database houses data modeling the geometry of the riser. 

Because this method is not comprehensive enough to adequately measure riser motion, Dr Triantafyllou said, other parameters such as the Reynolds number (a ratio that helps predict fluid flow patterns in different situations), surface roughness of the riser and external turbulence were included into the database.

Additionally, long-term issues that can affect riser motion, such as biofouling and the impact of equipment age on structural integrity, are impossible to measure in a hydrodynamic database, making long-term prediction and monitoring even more challenging.

To handle this challenge of incorporating hard-to-measure data in the digital twin, the MIT-led team developed an active learning neural network that uses algorithms that can interact with a user to sort unlabeled data. By interacting with the user, the network learned how to associate certain unlabeled data with certain parameter labels, eventually taking over the task by itself. 

MIT also developed the Intelligent Towing Tank (ITT), a robot that is capable of learning about complex fluid-structure phenomena, to test the validity of the active learning network. The robot designed, performed and analyzed experiments, using field data, to explore and map the complex forces that govern VIVs. An “explore-and-exploit” methodology also helped to reduce the number of experiments needed, improving the efficiency of the active learning network. 

The second prong of the algorithm development involved creating a neural network (LSTM-ModNet) that could use sensor measurements to construct and analyze the motion of a riser in deepwater. The network allowed the researchers to combine different types of sensor measurements, such as strain and acceleration, to reconstruct the motion of the riser, as well as fatigue, over time. 

The third prong involved using another neural network, DeepONet, to map a separate set of input parameters (inflow velocity, riser-bending, stiffness and tension as a function of water depth) against various output parameters (strain and amplitude). This network served as the predictive mechanism within the digital twin, actually predicting the occurrence and amplitude of VIVs in a given environment.

Taken together, the three-pronged approach demonstrate how machine learning algorithms and deep learning neural networks can infer riser dynamics from disparate sources of data, Dr Triantafyllou said. This enables the creation of a digital twin that can accurately incorporate the complexities of riser motion. 

Next steps

The MIT researchers are currently working with the operator members of the consortium to create demo digital twins of marine risers from current E&P projects, customizing the algorithms to fit each operator’s specific needs.

“Every company involved here has a different niche application that they want the digital twin to address. You’ve got companies that are more interested in the artificial intelligence, and they want to see if the algorithms are suitable for other applications in addition to the marine risers,” Dr Triantafyllou said. “Another company is using instrumented risers, and they want to use the digital twin to update the predictive models that they’re using for those sensors. The demonstration will give every stakeholder an opportunity to apply the digital twin the way they want.” 

ABS, which has published several guides and recommended practices on marine riser systems, will provide classification and technical services for the MIT researchers as they customize the design of each riser in the digital twin. The goal is for each operator to begin using the algorithms in their workflows either by the end of this year or early next year.

“The design and functionality of this digital twin has changed by the year, so we expect to see some more changes as we talk to the various companies and incorporate their designs into our system,” Dr Triantafyllou said, explaining that ABS will provide guidance to both MIT and operators on how to certify certain methodologies that will be used to create the customized digital twins. “By the time this is ready, we should have something really powerful.”  DC 

Vortices, or closed circular flows of air or water, are the primary cause of motion for marine drilling risers. Understanding when these vortices may occur, as well as the potential force of vortex-induced vibrations (VIVs), is important in predicting potential fatigue damage to the riser. However, VIVs are still difficult to anticipate due to the geometric complexity of the riser configuration, as well as the sheared and unsteady nature of ocean currents.

Over the past five years, researchers from the Massachusetts Institute of Technology (MIT) and Brown University have been working with a consortium of operators, drilling contractors and service companies to create a digital twin applicable to marine risers, DigiMaR. The twin was trained using a variety of data like field sensor data, computational flow dynamic simulations and existing databases, and has semi-empirical codes that incorporate various assumptions about the data to predict VIVs. The result is an accurate reconstruction of riser motion and prediction of fatigue damage, which can help to optimize the placement of sensors on the risers and extend their field life. 

“Artificial intelligence has become an essential tool for the design and operation of complex systems, and digital twins provide an unprecedented capability to explore the parametric space,” said Michael Triantafyllou, Professor of Mechanical and Ocean Engineering at MIT. The digital twin utilizes a parametric representation of the data characterizing VIV behavior in a series of databases, which Dr Triantafyllou said would provide a comprehensive view of how VIVs affect riser motion. 

The twin takes streaming data from physical marine risers and their environment and feeds it into AI predictive models and neural networks. Once the data is assessed and interpreted, the twin outputs information characterizing the magnitude of parameters like the riser’s displacement, velocity and acceleration. Those outputs can then be used to predict the potential fatigue damage of the structure. Additional machine learning algorithms within the digital twin also predict optimal times for servicing of the riser. 

During the 2023 Offshore Technology Conference (OTC) in Houston on 2 May, the American Bureau of Shipping (ABS) announced it had joined the project to provide third-party classification services for the design of the digital twin, which is now being incorporated into the workflows of major operators. “As a safety-focused organization, we’re able to share the insight we have learned with MIT to help support the adoption of new tools and systems that can improve safety and performance,” said Patrick Ryan, ABS Senior VP and Chief Technology Officer. 

The consortium includes operators Petrobras, ExxonMobil and Shell, as well as drilling contractor Saipem. 

Digital twin design process

Algorithms used in the digital twin were developed under a three-pronged approach. First, active learning algorithms were used to enhance the capabilities of the codes typically used to build a model of a marine riser. Digital twins of marine risers usually rely on the data captured from the field that measure the motion of fluids and the forces, such as VIVs, acting on solid bodies immersed in fluid, also known as hydrodynamic data. This data is stored into a database, while a separate database houses data modeling the geometry of the riser. 

Because this method is not comprehensive enough to adequately measure riser motion, Dr Triantafyllou said, other parameters such as the Reynolds number (a ratio that helps predict fluid flow patterns in different situations), surface roughness of the riser and external turbulence were included into the database.

Additionally, long-term issues that can affect riser motion, such as biofouling and the impact of equipment age on structural integrity, are impossible to measure in a hydrodynamic database, making long-term prediction and monitoring even more challenging.

To handle this challenge of incorporating hard-to-measure data in the digital twin, the MIT-led team developed an active learning neural network that uses algorithms that can interact with a user to sort unlabeled data. By interacting with the user, the network learned how to associate certain unlabeled data with certain parameter labels, eventually taking over the task by itself. 

MIT also developed the Intelligent Towing Tank (ITT), a robot that is capable of learning about complex fluid-structure phenomena, to test the validity of the active learning network. The robot designed, performed and analyzed experiments, using field data, to explore and map the complex forces that govern VIVs. An “explore-and-exploit” methodology also helped to reduce the number of experiments needed, improving the efficiency of the active learning network. 

The second prong of the algorithm development involved creating a neural network (LSTM-ModNet) that could use sensor measurements to construct and analyze the motion of a riser in deepwater. The network allowed the researchers to combine different types of sensor measurements, such as strain and acceleration, to reconstruct the motion of the riser, as well as fatigue, over time. 

The third prong involved using another neural network, DeepONet, to map a separate set of input parameters (inflow velocity, riser-bending, stiffness and tension as a function of water depth) against various output parameters (strain and amplitude). This network served as the predictive mechanism within the digital twin, actually predicting the occurrence and amplitude of VIVs in a given environment.

Taken together, the three-pronged approach demonstrate how machine learning algorithms and deep learning neural networks can infer riser dynamics from disparate sources of data, Dr Triantafyllou said. This enables the creation of a digital twin that can accurately incorporate the complexities of riser motion. 

Next steps

The MIT researchers are currently working with the operator members of the consortium to create demo digital twins of marine risers from current E&P projects, customizing the algorithms to fit each operator’s specific needs.

“Every company involved here has a different niche application that they want the digital twin to address. You’ve got companies that are more interested in the artificial intelligence, and they want to see if the algorithms are suitable for other applications in addition to the marine risers,” Dr Triantafyllou said. “Another company is using instrumented risers, and they want to use the digital twin to update the predictive models that they’re using for those sensors. The demonstration will give every stakeholder an opportunity to apply the digital twin the way they want.” 

ABS, which has published several guides and recommended practices on marine riser systems, will provide classification and technical services for the MIT researchers as they customize the design of each riser in the digital twin. The goal is for each operator to begin using the algorithms in their workflows either by the end of this year or early next year.

“The design and functionality of this digital twin has changed by the year, so we expect to see some more changes as we talk to the various companies and incorporate their designs into our system,” Dr Triantafyllou said, explaining that ABS will provide guidance to both MIT and operators on how to certify certain methodologies that will be used to create the customized digital twins. “By the time this is ready, we should have something really powerful.”  DC