Authors:
(1) Ruochu Yang;
(2) Chad Lembke;
(3) Fumin Zhang;
(4) Catherine Edwards. Table of Links Abstract and Intro Anomaly Detection Algorithm Experimental Evaluation Conclusion and References Abstract— Underwater gliders have been widely used in oceanography for a range of applications. However, unpredictable events like shark strikes or remora attachments can lead to abnormal glider behavior or even loss of the instrument. This paper employs an anomaly detection algorithm to assess operational conditions of underwater gliders in the real-world ocean environment. Prompt alerts are provided to glider pilots upon detecting any anomaly, so that they can take control of the glider to prevent further harm. The detection algorithm is applied to multiple datasets collected in real glider deployments led by the University of Georgia’s Skidaway Institute of Oceanography (SkIO) and the University of South Florida (USF). In order to demonstrate the algorithm generality, the experimental evaluation is applied to four glider deployment datasets, each highlighting various anomalies happening in different scenes. Specifically, we utilize high resolution datasets only available post-recovery to perform detailed analysis of the anomaly and compare it with pilot logs. Additionally, we simulate the online detection based on the real-time subsets of data transmitted from the glider at the surfacing events. While the real-time data may not contain as much rich information as the post-recovery one, the online detection is of great importance as it allows glider pilots to monitor potential abnormal conditions in real time. I. INTRODUCTION Underwater gliders are extensively used in ocean research for activities such as ocean sampling, surveillance, and other purposes [1]–[5]. However, given the complexity of the ocean environment and the long-duration of glider missions, unexpected events such as shark attacks, wing loss, or attachment of marine species can cause gliders to operate abnormally or even totally fail [6]–[8]. In such cases, the gliders may drift to unexpected areas, making localization and rescue operations challenging. Furthermore, it can be difficult to detect the abnormal behavior of gliders, particularly when external disturbances arise, due to the lack of monitoring devices [9]–[12]. The deployment of monitoring devices for gliders or the addition of self-monitoring of performance would increase mission costs and pilot complexity. Typically, glider pilots can only rely on heavily subsetted data transmitted by the glider in real time to form hypotheses about potential anomalies. Sometimes, they just resort to climb and dive ballast data to assess if the glider is surfacing or diving as expected. However, this empirical detection can never be conclusively confirmed as the mission is going on. To address this challenge, we develop an anomaly detection algorithm that systematically utilizes simple glider data such as glider speed, heading, and trajectory. This algorithm is feasible for theoretical validation on numerous real-world glider datasets, and runs autonomously in real-time, as opposed to manual detection by human pilots. By monitoring gliders in realtime, the algorithm allows glider pilots to take appropriate actions promptly to ensure the safety and success of missions. Different strategies have been in the field of underwater robotics to identify abnormal behavior of underwater gliders. Some anomaly detection algorithms focus on changes in robot motion, such as roll angle or pitch angle, to detect possible motion deviation or a foreign object attached to the glider [13], [14]. Some algorithms monitor the power consumption or motor performance of the glider, as variations in these parameters can indicate degeneration of individual components, such as propellers and rotors [15]–[17]. Other algorithms utilize machine learning techniques to identify anomalous behavior by analyzing sensor data collected by the glider over time, such as changes in the speed, roll, pitch, or depth [18]–[20]. However, most of the existing research relies on shore-based manual implementations and does not resolve issues like inability to perform online detection on the gliders or lack of real-time experimental verification. In addition, it is essential to determine whether the detected anomaly is false positive [21], [22]. When the ocean current speed is significantly greater than the maximum speed of the marine robot, it can lead to a considerable performance degradation. Under such circumstances, false alarms should be avoided since the anomaly caused by an unexpected ocean current is unrelated to the glider itself. In practice, it is challenging to separate flow speed and glider speed due to hardware limitations, but leveraging the Controlled Lagrangian Particle Tracking (CLPT) framework [23], the anomaly detection algorithm in [24] generates real-time estimates of the glider speed and flow speed from the trajectory and heading angles. The estimated glider speed is compared with the normal speed range to detect anomalies, while the algorithm-estimated flow speed is compared with the gliderestimated flow speed to avoid false alarms. We initially validate the anomaly detection algorithm by using two real-life deployment datasets [25]. Building upon this previous work, we aim to extend the algorithm to largescale datasets, thus effectively handling various anomalies in diverse missions. We also plan to simulate online implementation of the detection to enable real-time interaction with glider pilots. This objectives constitute primary motivation of this paper, and our main contribution are summarized as follows. • We demonstrate generality of the anomaly detection algorithm based on four glider datasets collected in real deployments featuring diverse anomalies. • We simulate online mode implementation of the algorithm to a real glider deployment with limited data streams in real time for the first time. The SouthEast Coastal Ocean Observing Regional Association (SECOORA) glider Franklin, operated by Skidaway Institute of Oceanography (SkIO), and the University of South Florida (USF) gliders USF-Sam, USF-Gansett, and USF-Stella provide numerous examples of valuable experimental data in which anomalies may be associated with marine bio-hazards. Promising anomaly detection results of these datasets are shown to well match glider pilots’ hindcast analysis. Building off its efficacy, the real-time anomaly detection algorithm is incorporated into the autonomous glider navigation software GENIoS Python [26] to better assist human pilots as an add-on warning functionality. This paper is organized as follows. Section II illustrates the framework of the anomaly detection algorithm. Section III describes the experimental setup of glider deployments, verifies the algorithm by detecting anomalies in large-scale real experiments, and simulates the online implementation on subsetted glider datasets. Section IV provides conclusions and future work. This paper is available on arxiv under CC 4.0 license. Authors: (1) Ruochu Yang; (2) Chad Lembke; (3) Fumin Zhang; (4) Catherine Edwards. Authors: Authors: (1) Ruochu Yang; (2) Chad Lembke; (3) Fumin Zhang; (4) Catherine Edwards. Table of Links Abstract and Intro Abstract and Intro Anomaly Detection Algorithm Anomaly Detection Algorithm Experimental Evaluation Experimental Evaluation Conclusion and References Conclusion and References Abstract— Underwater gliders have been widely used in oceanography for a range of applications. However, unpredictable events like shark strikes or remora attachments can lead to abnormal glider behavior or even loss of the instrument. This paper employs an anomaly detection algorithm to assess operational conditions of underwater gliders in the real-world ocean environment. Prompt alerts are provided to glider pilots upon detecting any anomaly, so that they can take control of the glider to prevent further harm. The detection algorithm is applied to multiple datasets collected in real glider deployments led by the University of Georgia’s Skidaway Institute of Oceanography (SkIO) and the University of South Florida (USF). In order to demonstrate the algorithm generality, the experimental evaluation is applied to four glider deployment datasets, each highlighting various anomalies happening in different scenes. Specifically, we utilize high resolution datasets only available post-recovery to perform detailed analysis of the anomaly and compare it with pilot logs. Additionally, we simulate the online detection based on the real-time subsets of data transmitted from the glider at the surfacing events. While the real-time data may not contain as much rich information as the post-recovery one, the online detection is of great importance as it allows glider pilots to monitor potential abnormal conditions in real time. Abstract — Underwater gliders have been widely used in oceanography for a range of applications. However, unpredictable events like shark strikes or remora attachments can lead to abnormal glider behavior or even loss of the instrument. This paper employs an anomaly detection algorithm to assess operational conditions of underwater gliders in the real-world ocean environment. Prompt alerts are provided to glider pilots upon detecting any anomaly, so that they can take control of the glider to prevent further harm. The detection algorithm is applied to multiple datasets collected in real glider deployments led by the University of Georgia’s Skidaway Institute of Oceanography (SkIO) and the University of South Florida (USF). In order to demonstrate the algorithm generality, the experimental evaluation is applied to four glider deployment datasets, each highlighting various anomalies happening in different scenes. Specifically, we utilize high resolution datasets only available post-recovery to perform detailed analysis of the anomaly and compare it with pilot logs. Additionally, we simulate the online detection based on the real-time subsets of data transmitted from the glider at the surfacing events. While the real-time data may not contain as much rich information as the post-recovery one, the online detection is of great importance as it allows glider pilots to monitor potential abnormal conditions in real time. Abstract I. INTRODUCTION Underwater gliders are extensively used in ocean research for activities such as ocean sampling, surveillance, and other purposes [1]–[5]. However, given the complexity of the ocean environment and the long-duration of glider missions, unexpected events such as shark attacks, wing loss, or attachment of marine species can cause gliders to operate abnormally or even totally fail [6]–[8]. In such cases, the gliders may drift to unexpected areas, making localization and rescue operations challenging. Furthermore, it can be difficult to detect the abnormal behavior of gliders, particularly when external disturbances arise, due to the lack of monitoring devices [9]–[12]. The deployment of monitoring devices for gliders or the addition of self-monitoring of performance would increase mission costs and pilot complexity. Typically, glider pilots can only rely on heavily subsetted data transmitted by the glider in real time to form hypotheses about potential anomalies. Sometimes, they just resort to climb and dive ballast data to assess if the glider is surfacing or diving as expected. However, this empirical detection can never be conclusively confirmed as the mission is going on. To address this challenge, we develop an anomaly detection algorithm that systematically utilizes simple glider data such as glider speed, heading, and trajectory. This algorithm is feasible for theoretical validation on numerous real-world glider datasets, and runs autonomously in real-time, as opposed to manual detection by human pilots. By monitoring gliders in realtime, the algorithm allows glider pilots to take appropriate actions promptly to ensure the safety and success of missions. Different strategies have been in the field of underwater robotics to identify abnormal behavior of underwater gliders. Some anomaly detection algorithms focus on changes in robot motion, such as roll angle or pitch angle, to detect possible motion deviation or a foreign object attached to the glider [13], [14]. Some algorithms monitor the power consumption or motor performance of the glider, as variations in these parameters can indicate degeneration of individual components, such as propellers and rotors [15]–[17]. Other algorithms utilize machine learning techniques to identify anomalous behavior by analyzing sensor data collected by the glider over time, such as changes in the speed, roll, pitch, or depth [18]–[20]. However, most of the existing research relies on shore-based manual implementations and does not resolve issues like inability to perform online detection on the gliders or lack of real-time experimental verification. In addition, it is essential to determine whether the detected anomaly is false positive [21], [22]. When the ocean current speed is significantly greater than the maximum speed of the marine robot, it can lead to a considerable performance degradation. Under such circumstances, false alarms should be avoided since the anomaly caused by an unexpected ocean current is unrelated to the glider itself. In practice, it is challenging to separate flow speed and glider speed due to hardware limitations, but leveraging the Controlled Lagrangian Particle Tracking (CLPT) framework [23], the anomaly detection algorithm in [24] generates real-time estimates of the glider speed and flow speed from the trajectory and heading angles. The estimated glider speed is compared with the normal speed range to detect anomalies, while the algorithm-estimated flow speed is compared with the gliderestimated flow speed to avoid false alarms. We initially validate the anomaly detection algorithm by using two real-life deployment datasets [25]. Building upon this previous work, we aim to extend the algorithm to largescale datasets, thus effectively handling various anomalies in diverse missions. We also plan to simulate online implementation of the detection to enable real-time interaction with glider pilots. This objectives constitute primary motivation of this paper, and our main contribution are summarized as follows. • We demonstrate generality of the anomaly detection algorithm based on four glider datasets collected in real deployments featuring diverse anomalies. • We simulate online mode implementation of the algorithm to a real glider deployment with limited data streams in real time for the first time. The SouthEast Coastal Ocean Observing Regional Association (SECOORA) glider Franklin, operated by Skidaway Institute of Oceanography (SkIO), and the University of South Florida (USF) gliders USF-Sam, USF-Gansett, and USF-Stella provide numerous examples of valuable experimental data in which anomalies may be associated with marine bio-hazards. Promising anomaly detection results of these datasets are shown to well match glider pilots’ hindcast analysis. Building off its efficacy, the real-time anomaly detection algorithm is incorporated into the autonomous glider navigation software GENIoS Python [26] to better assist human pilots as an add-on warning functionality. This paper is organized as follows. Section II illustrates the framework of the anomaly detection algorithm. Section III describes the experimental setup of glider deployments, verifies the algorithm by detecting anomalies in large-scale real experiments, and simulates the online implementation on subsetted glider datasets. Section IV provides conclusions and future work. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Real-Time Anomaly Detection in Underwater Gliders: Anomaly Detection Algorithm

Oceanography.Tech

Too Long; Didn't Read

Make resilience your competitive advantage

Real-Time Anomaly Detection in Underwater Gliders: Abstract and Intro

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

A Deep Discussion On Our Study Looking at Near-Inertial Waves and Mesoscale Eddies: Diving In

The Noonification: Anomaly Detection with Privileged Information—Part 1 (12/13/2023)

The Noonification: Craig Wright is Satoshi Nakamoto (12/24/2023)

3 Types of Anomalies in Anomaly Detection

A Zero-Knowledge Anomaly Detection Approach for Robust Federated Learning

A Deep Discussion On Our Study Looking at Near-Inertial Waves and Mesoscale Eddies: Diving In

The Noonification: Anomaly Detection with Privileged Information—Part 1 (12/13/2023)

The Noonification: Craig Wright is Satoshi Nakamoto (12/24/2023)

3 Types of Anomalies in Anomaly Detection

A Zero-Knowledge Anomaly Detection Approach for Robust Federated Learning

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps