Exploring New Frontiers with Artificial Intelligence At the Marine Engineering Lab in the Naval Architecture and Marine Engineering Department, exciting research is underway that combines large data sets with cutting-edge technology. Researchers are applying Artificial Intelligence (AI) to predict and prevent problems in ship machinery. This work is especially important for autonomous, or crewless, naval…
Exploring New Frontiers with Artificial Intelligence
At the Marine Engineering Lab in the Naval Architecture and Marine Engineering Department, exciting research is underway that combines large data sets with cutting-edge technology. Researchers are applying Artificial Intelligence (AI) to predict and prevent problems in ship machinery. This work is especially important for autonomous, or crewless, naval vessels, which must be extremely reliable given the absence of crew to maintain and repair machinery plant faults and failures while underway.
How AI is Transforming Naval Engineering
The researchers collect huge amounts of data from various ship systems in a laboratory setting. By analyzing this data, they can detect early signs of equipment faults or failures before they become serious issues. AI helps make sense of this vast information quickly and accurately through custom frameworks and tensor networks. Tensor networks from statistical physics, are mathematical structures that enable the AI to be even more effective at spotting potential problems.
Meet the Pioneers: Recent PhD Graduates
Two recent PhD graduates, Andy Olson and Alex Manohar, have played key roles in this groundbreaking work. This work has been led by Professors Tim McCoy, Matthew Collette and David Singer with funding from the Office of Naval Research and support from undergraduate and graduate students: Ethan Almquist, Henry Zayko, Arianna Kerkmaz, Micah Williamson, August Sturm, Nayah Daniel, Kye Dembinski, Frances Truong, Andreya Ware, Frances Truong, Andreya Ware.
Building the Marine Engineering Laboratory (MEL)
Training AI models requires extensive data, yet data for shipboard machinery systems is not widely available. The team strategized to build a model-scale facility that could generate such data, much like the Marine Hydrodynamics Lab can simulate resistance, propulsion, and seakeeping challenges at model scale. Under the leadership of Professor Tim McCoy, Ph.D. student Andy Olson designed and built the first-of-its-kind multiphysics hardware lab dedicated to marine engineering systems in the United States. The laboratory plant includes simulated diesel engines, cooling and fuel systems, propulsion, and mission loads. This laboratory plant contains hardware and software-based linkages to enable the injection of faults into various systems, which can lead to cooling loss and fuel starvation, which can cascade into level failures. The virtual linkages allow fault injection without physical hardware damage. This unique capability coupled with the embedded LabVIEW control and data acquisition system, enables the collection of data for over 200 machinery plant signals for repetitive run-to-failure profiles. By using the capabilities of the MEL, the University of Michigan generated a unique set of hardware-based failure simulations suitable for training AI models.
Andy Olson’s Dissertation: A Deep Dive
Andy Olson used the data from the MEL to develop an AI framework to predict the time remaining before a critical plant failure occurs, based on a time history of the plant’s operation. Andy studied Long Short-Term Memory (LSTM) networks, which correlate readings over time to predict future events. He tested the framework by modeling common machinery plant failures from literature in software, then injecting these failures into the MEL machinery plant to form a dataset of 100 run-to-failure profiles, where the failures were simulated in 10 sequential steps. By using the initial steps within each profile, the AI method was able to predict both the type of fault present (diagnosis) and the remaining number of steps before the plant’s capacity fell below a critical threshold (prognosis).
In conjunction with this research, Olson conducted hardware simulations that illustrated the impact of common faults at the plant level and showcased potential solutions given knowledge of a future system failure prior to its presence. Fault mitigation strategies included inverter-based energy storage, uneven generator load sharing, and traditional mitigation techniques such as load shedding. Olson’s work has showcased the potential for AI coupled with active fault mitigation strategies to enable resilient machinery plant operation in an unmanned and autonomous environment.
Alex Manohar’s Dissertation: A Deep Dive
Alex Manohar developed the Network of Tensor Networks (NTN) framework. The developed diagnostic framework enables self-adaptive health monitoring systems on board crewless vessels, ensuring they can adapt to any failures that arise. By utilizing techniques from statistical physics, he created a novel framework that leveraged networks of tensor networks to enable lightweight communication among general decentralized systems. Decentralized and centralized agents then utilized the NTN framework to quickly diagnose local and system-wide failures.
To demonstrate the capabilities of the NTN framework, Manohar worked in conjunction with Olson to create a series of failure profiles that captured a wide range of failure states in the Marine Engineering Lab. Manohar then designed autonomous decentralized and centralized agents to diagnose these failures and trace them back to their origin point. These agents assessed the system and made their diagnoses in real time without using highly complex machine learning-based AI algorithms. Agents were able to diagnose failures reliably and quickly with more basic AI. This work sets a great starting point for further work in developing self-adaptive health monitoring systems for crewless vessels, which are critical for the safety and reliability of crewless vessels.
Real-World Experience for Students
The Marine Engineering Lab isn’t just for researchers. It’s also integrated into junior-level courses, such as NA 331 and NA 332. In these classes, students get hands-on experience with machinery plants, including learning how to parallel generators and evaluate electrical harmonics on ships. This practical training is invaluable for students aspiring to work in marine engineering and has been led by Professors Tim McCoy and John Page.
The work being done at the Marine Engineering Lab showcases the potential of AI to revolutionize how we maintain and operate ships, especially autonomous ones. Through the efforts of dedicated researchers and students, the lab is paving the way for a future where naval vessels are safer and more reliable than ever.