A Marine Growth Detection System for Underwater Gliders

Abstract

Marine growth has been observed to cause a drop in the horizontal and vertical velocities of underwater gliders, thus making them unresponsive and needing immediate recovery. Currently, no strategies exist to correctly identify the onset of marine growth for gliders and only limited data sets of biofouled hulls exist. Here, a field test has been conducted to first investigate the impact of marine growth on the dynamics and power consumption of underwater gliders and then design an anomaly detection system for high levels of biofouling. A Slocum glider was deployed first for eight days with drag stimulators to imitate severe biofouling; then, the vehicle was redeployed with no additions to the hull for further 20 days. The mimicked biofouling caused a speed reduction due to a significant increase in drag. Additionally, the lower speed causes the steady-state flight stage to last longer and the rudder to become less responsive; hence, marine growth results in a shortening of deployment duration through an increase in power consumption. As actual biofouling due to p. pollicipes occurred during the second deployment, it is possible to develop and test a system that successfully detects and identifies high levels of marine growth on the glider, blending model-and data-based solutions using steady-state flight data. The system will greatly help pilots replan missions to safely recover the vehicle if significant biofouling is detected.

Smart anomaly detection for Slocum underwater gliders with a variational autoencoder with long short-term memory networks

“…Condition monitoring solutions with a large focus on machine and deep learning are evaluated for autonomous ships by Ellefsen et al Ellefsen et al (2019). Specific to underwater gliders, Anderlini et al (2021a) introduced an anomaly detection system for biofouling using ensembles of regression trees.…”

“…492, possibly due to the stronger dynamic effects associated with the shallower depth of the deployment negatively impacting the steady-state model (200 m vs 1,000 m). Furthermore, the VAE is also able to detect the growing natural biofouling on the hull during the second deployment, presenting similar performance to the ensemble of regression trees used inAnderlini et al (2021a) whilst being trained on data from other glider units. Therefore, the anomaly detection system based on VAE and LSTM presents better generality to different anomaly types than the model-based solution and better data handling and transferability to different platforms than the random forest scheme.…”

Anomaly Detection and Fault Diagnostics for Underwater Gliders Using Deep Learning

“…In [4], system identification techniques were employed to detect changes in model parameters which further successfully deduced simulated and natural marine growth. Anderlini, et al [5] further conducted a field test to validate a marine growth detection system for UGs using ensembles of regression trees. In [6], the use of a range of deep learning techniques was investigated to achieve over-the-horizon anomaly detection for UGs.…”