Detection and localization of damages in the nets of fish farms

Short Introduction:

Norwegian companies like SalMar produce high quality aquaculture products via fish farms where nets are one of the main components. Damages such as narrow tears, twine fractures, irregular holes in the nets happen regularly and lead to the escape of large volume fishes with negative economic and biologic effects. Thus, the early detection of damages in the nets of a fish farm is vital. Additionally, the fish farms operate under varying environmental conditions (ECs) such as weight height (WH) which partially or fully “masks” the effects of damages on the structural dynamics, rendering damage detection highly challenging. The current methods for monitoring the nets’ health are based on the acquirement of thousands of photos through Remote Operating Vehicles (ROVs) whose use is sparse and very expensive. This is why there is a need of further research for a remote, effective and less expensive monitoring method based on sensors continuously collecting signals. In the context of the project Sustainable Offshore Aquaculture (SusOffAqua)”, Norwegian Research Centre – NORCE AS created a bachelor / master project where a scaled fishing net will be placed in a water tank and damage detection and localization in the net will be investigated via statistical methods under varying environmental conditions.

Project Description:

In this Bachelor/Master project, the student initially will prepare the experimental set-up which will involve a scale fishing net placed inside a large plexiglass tank filled with water (Figures 1(a,b)). The net will be excited by waves created though a plate moved via an electromechanical cylinder (Figure 1(c)). Sensors (Figure 1(d)) will be attached on the net and they will collect data and send it to a data acquisition device., Experiments will be conducted based on the experimental equipment under a varying EC which will be the wave height (WH). Different WH values will be represented by different values of the cylinder’s moving velocity.

After setting up the experimental set-up, the student will conduct indicative experiments under the healthy state and various damage states, process the signals through various techniques such as filtering, subsampling, scaling and then the optimum positions of the sensors will be selected. This selection will be achieved through criteria based on the coherence function or the Power Spectral Density (PSD) ((Figure 1(e)). The signals based on the selected sensor positions should include most of the information (natural frequencies) about the net’s dynamics under the healthy state. Additionally, the analysis of these signals should reveal the highest differences between healthy and damage states. In parallel to the selection of the sensor locations, experimental details such sampling frequency and length of received signals will be decided as well. 

Subsequently, statistical models with a non-parametric form such as PSD and with a parametric form such AutoRegrssive (AR) models, AR - AR with eXogenous Input (AR-ARX) models will be identified based on the acquired signals. The student will extract the dynamic characteristics (natural frequencies) from the models and check if these models are able to describe the net’s dynamics under varying WH.

Finally, the student will achieve detection and localization in the net via machine learning methods and under the varying WH. In these methods, features will be obtained from the identified PSD and AR-ARX/AR models and they will be used either directly for damage detection and localization or for the training of classifiers like Support Vector Machine (SVM) or K-Nearest Neighbor (KNN) which will be used for damage detection and localization. Information compression techniques such as the Principal Component Analysis (PCA) will also be used with the employed methods in an effort to optimize the detection and localization. A damage will be simulated by small holes at a single or two locations in the net.  The results will be presented in a conference.

Keywords:

•    Fishing Farms Nets

•    Laboratory experiment

•    Non-parametric and parametric modelling

•    Damage detection / localization

•    Statistical methods

•    Varying environmental condition

Additional Information:

This master project constitutes a part of the project “Sustainable Offshore Aquaculture (SusOffAqua)”. SusOffAqua has been funded by the Research Council of Norway and it is a collaborative research between University in Stavanger, NORCE and Norwegian Veterinary Institute.

Contact information:

Rune Schlanbusch (Deputy Director), rusc@norceresearch.no, 90960133

Christos Sakaris (Researcher), csak@norceresearch.no, 47742610