In just its initial year of functionality, the groundbreaking Intelligent Towing Tank (ITT) successfully executed approximately 100,000 experimental trials, effectively accomplishing what would typically require a doctoral student's five years of intensive research within mere weeks.
This innovative automated research facility, engineered within the MIT Sea Grant Hydrodynamics Laboratory, autonomously and adaptively performs, analyzes, and designs investigations exploring vortex-induced vibrations (VIVs). These phenomena, crucial for engineering offshore ocean structures such as marine drilling risers connecting subsea oil wells to surface platforms, continue to puzzle researchers due to the multitude of parameters involved.
Powered by active learning algorithms, the ITT conducts experimental sequences where the parameters for each subsequent investigation are algorithmically determined by computer systems. Employing an "explore-and-exploit" methodology, this revolutionary approach dramatically minimizes the number of experiments required to explore and map the intricate forces governing VIVs.
The innovation originated from then-PhD candidate Dixia Fan's ambition to streamline the process of conducting approximately one thousand painstaking manual experiments, which ultimately led to the development of this cutting-edge system and a recently published paper in the prestigious journal Science Robotics.
Fan, now serving as a postdoctoral researcher, alongside a team of scientists from the MIT Sea Grant College Program, MIT's Department of Mechanical Engineering, École Normale Supérieure de Rennes, and Brown University, unveils a potential paradigm shift in experimental research, where enhanced human-computer-robot collaboration could significantly accelerate scientific discovery.
The impressive 33-foot research tank operates continuously without interruption or supervision, tackling complex fluid-structure interaction challenges. However, the researchers envision broader applications of this active learning and automation approach across multiple disciplines, potentially generating new insights and models in complex multi-input/multi-output nonlinear systems.
VIVs represent inherently nonlinear motions induced on structures within irregular cross-streams, making them particularly challenging to investigate. The researchers report that the volume of experiments completed by the ITT already rivals the total number of experiments conducted worldwide on VIVs to date.
This remarkable efficiency stems from the numerous independent parameters—from flow velocity to pressure—involved in studying these complex forces. According to Fan, a systematic brute-force approach—blindly conducting 10 measurements per parameter within an eight-dimensional parametric space—would necessitate approximately 100 million experiments.
Through the ITT, Fan and his collaborators have expanded exploration into previously impractical parametric spaces. "If we had employed traditional techniques for our research problem," he explains, "the experimentation would require 950 years to complete." Recognizing this impossibility, the team integrated a Gaussian process regression learning algorithm into the ITT, thereby reducing the experimental burden by several orders of magnitude, requiring only a few thousand experiments.
The robotic system automatically conducts an initial sequence of experiments, periodically towing a submerged structure along the tank's length at constant velocity. Subsequently, the ITT assumes partial control over each experiment's parameters by minimizing appropriate acquisition functions of quantified uncertainties while adapting to achieve various objectives, such as drag reduction.
Earlier this year, Fan received the MIT Mechanical Engineering de Florez Award for "Outstanding Ingenuity and Creative Judgment" in developing the ITT. "Dixia's design of the Intelligent Towing Tank exemplifies how novel approaches can revitalize established research fields," remarks Michael Triantafyllou, Henry L. and Grace Doherty Professor in Ocean Science and Engineering, who served as Fan's doctoral advisor.
Triantafyllou, a co-author on this paper and director of the MIT Sea Grant College Program, notes, "MIT Sea Grant has invested resources and funded projects employing deep-learning methods for ocean-related challenges for several years, and these investments are already yielding significant returns." Funded by the National Oceanic and Atmospheric Administration and administered by the National Sea Grant Program, MIT Sea Grant represents a federal-Institute partnership that leverages MIT's research and engineering expertise to address ocean-related challenges.
Fan's research aligns with numerous other initiatives utilizing automation and artificial intelligence in scientific exploration: At Caltech, a robot scientist named "Adam" generates and tests hypotheses; at the Defense Advanced Research Projects Agency, the Big Mechanism program analyzes tens of thousands of research papers to generate new models.
Similarly, the ITT leverages human-computer-robot collaboration to accelerate experimental efforts. The system demonstrates a potential paradigm shift in research methodology, where automation and uncertainty quantification can significantly accelerate scientific discovery. The researchers assert that the machine learning methodology detailed in this paper can be adapted and applied beyond fluid mechanics to numerous other experimental fields.
Additional contributors to the paper include George Karniadakis from Brown University (also affiliated with MIT Sea Grant); Gurvan Jodin from ENS Rennes; MIT mechanical engineering PhD candidate Yu Ma; and Thomas Consi, Luca Bonfiglio, and Lily Keyes from MIT Sea Grant.
This research received support from DARPA, Fariba Fahroo, and Jan Vandenbrande through an EQUiPS (Enabling Quantification of Uncertainty in Physical Systems) grant, as well as funding from Shell, Subsea 7, and the MIT Sea Grant College Program.
