Master and Specialization projects

(Make an appointment and get more information about the projects. Contact: Jo Arve Alfredsen)

Integration of a machine vision plankton sensor in an unmanned surface vehicle for real-time autonomous ocean monitoring

The abundance, composition and dynamics of the phytoplankton community have great impact on ocean productivity and health. Plankton blooms may under certain circumstances develop excessively into a critical state that causes oxygen depletion and/or toxic conditions for organisms living in the ocean environment (HAB – harmful algal bloom). HABs can be particularly detrimental to fish farming and other aquacultural production.

The goal of this project is to design a cost-efficient and flexible platform that will enable persistent autonomous plankton sampling and analyses. Focus will be on adapting the sensor payload, a machine vision-based plankton sensor (PlanktoScope), to an unmanned surface vehicle platform (USV Otter) and explore the solution through practical experiments on Trondheimsfjorden.

Implementation of machine learning algorithms on an embedded platform (in cooperation with Nortek AS)

Machine learning (ML) is a branch of artificial intelligence (AI) which deals with the use of data and computer algorithms, where algorithms identify and learn patterns from the available data with the overall goal of self-improvement. ML algorithms deal with large amount of data and involve computationally complex mathematical operations, putting a very high demand on memory and compute power. Recently, a trend called Edge ML has been started where ML algorithms are implemented on embedded platforms, having limited memory and compute power. The rationale behind Edge ML is that an algorithm could be trained on a computationally stronger system, whereas the process of inference i.e., making a decision for a new input based on the ML model is executed on the embedded platform. Tools exist which offer quantization of the ML algorithms, which basically scales down ML models e.g., by a factor of 4 to 16 in size, making them implementable on an embedded computer.

Nortek AS designs, develops and produces scientific instruments that apply the Doppler principle to underwater acoustics in order to measure water in motion, such as currents and waves. Currently, Nortek is evaluating a new processing platform and use of machine learning algorithms in future instruments. This project deals with the implementation and evaluation of the TensorFlow APIs on a Digi ConnectCore® 8M Nano Development Kit (CC8MNDVK). CC8MNDVK is based on NXP i.Mx8M Nano SoC which is a multi-core embedded system offering a quad core ARM Cortex A-53 processor with a dedicated GPU. The master project involves tasks such as, implementation of a simple ML application (e.g. image classification) based on TensorFlow python API on a desktop computer/cloud server, auantization and implementation of the model on the CC8MNDVK platform using TensorFlow Lite python API, training and implementation of a ML algorithm e.g. for bottom tracking using a time series from Norteks DVL instrument on a desktop computer or a cloud server in python, and quantizing the bottom tracking model and run the inference on the CC8MNDVK platform using python and C++ API with and without CPU/GPU delegates.

Robotic fish tracking - integration of unmanned surface vehicle and acoustic fish telemetry

Fundamental understanding of the behaviour and distribution of marine living resources stands high on the international research agenda as it relates closely to our ability to work out sustainable management regimes of the oceans and the global marine environment. Scientific progress in this area is treasured and of broad societal value. This project targets development of new enabling technology that will make significant contributions in this direction.

Figure: Robotic fish tracking system.

Autonomous vehicle systems and acoustic fish telemetry are both research areas of strong tradition and merits at the Department of Engineering Cybernetics. The project aims to enable close integration of these areas to create novel platforms for robotic search, localization and tracking of marine life, migrating fish, and other similarly small and evasive underwater assets. The research will move current operational limits of fish/underwater object tracking and contribute significantly in terms of making new innovative technology available to researchers and enable new discoveries within movement ecology and the marine sciences in general.

The USV Otter and AutoNaut vehicle platforms are available at the department and under development with our own controls, sensors and instruments, where acoustic fish telemetry receivers will be an essential part of the vehicles’ payload. Several interesting student assignments on different topics may be defined within the frame of this project:

· Design of embedded hardware and software for:

o vehicle controls and communications

o sensors and payload integration

· Optimal search, mission and path planning

· Underwater target localization and stealthy tracking

· Multi-agent- and formation control, machine learning

Figure: The fish tracking Otter USV.

Figure: Department's wave-driven USV, AutoNaut.

Instrumentation and control of fish crowding system

In industrial-scale fish farming, fish crowding constitutes a particularly critical operation which may have strong negative impact on fish welfare, mortality and quality of product. Fish crowding is at the same time an important and unavoidable task inherent to many farm operations such as sorting, delousing and fish harvesting. Proper tools for monitoring and controlling the crowding process to stay within safe limits are therefore in great demand. The purpose of this project is to investigate different strategies for equipping the crowding gear and fish cage with sensors and instrumentation that can provide relevant information in real-time about the state of the crowding process and the fish’ response. The project will also investigate how this information can be exploited to control the crowding process in a way that minimizes the risk of injuring the fish in terms of mechanical damage and hypoxia. Sensor integration, embedded systems development and data analysis will be central and there will be good opportunities for experimental testing of prospective solutions. The project will be carried out in collaboration with SINTEF Ocean.

Figure: Fish crowding monitoring and control system.

Figure: Conceptual illustration of a smart sweep-net sensor

LPWAN gateway buoy (LoRa, NB-IoT) for underwater acoustic receiver

This project concerns development of a buoy solution for relaying data received by an underwater acoustic receiver to an internet backend. The buoy will typically be deployed for several months in remote fjord and coastal locations and should be capable of transmitting data wirelessly to a gateway on shore while consuming minimal amounts of power. The wireless link will be based on LoRaWAN and/or NB-IoT to meet the low-power long-range requirements of the system. The project will focus on software design and development on the dedicated embedded buoy controller, LPWAN networks, internet technology, and realization of bridge from LPWAN application layer to the DUNE software framework for heterogenous autonomous vehicles.

Figure: Remote hydrophone buoy concept.

Figure: Deployment of buoy prototype in a remote fjord.

Smart autonomous Lagrangian drifters for flow measurement in large-scale sea cages

A prototype buoyancy vehicle (BV) for underwater actuation has been developed. The BV is capable of generating its own vertical motion by controlling the buoyancy force through a piston-based displacement manipulation mechanism and, hence, its own volumetric mass density. Furthermore, stabilization of the BV at a certain reference depth is made possible through a pressure sensor and feedback control of buoyancy. The ability to stay neutral at a reference depth enables use of the BV as a subsurface Lagrangian drifter for ocean current measurements, a target application for the BV. The underwater positioning system described in project below will enable efficient subsurface tracking of the drifter.

The goal of the project is to bring the prototype BV into a fully operational state that enables testing of its performance as a subsurface Lagrangian drifter. Design and realization of an embedded computer (hw and sw) for controlling the vehicle will be central in this task, as well as experimental validation in a full-scale sea cage. The BV should also provide a practical user interface for configuration of the unit as well as retrieval of onboard logging data.

Figure: Vehicle that can control its own buoyancy using a piston mechanism.

Figure: Typical step response of the buoyancy vehicle.

Underwater ultra-low power acoustic positioning system

Acoustic fish telemetry constitutes a powerful scientific tool for investigating the behaviour of fish and other aquatic animals remotely in the underwater environment. The concept is enabled by developments in microelectronics, MEMS sensors, and ultra-low power embedded microcontrollers, and is sometimes be referred to as “fish & chips”. This project concerns the development of a system for estimating the position and tracking the movement of fish as well as other small subsea objects (e.g. an AUV) carrying miniature acoustic transmitters. The approach will be based on time difference of arrival (TDoA) measurements of signals in a spatially dispersed array of acoustic receivers (hydrophones) spanning the area of interest, e.g. an industry-scale fish cage. Receiver synchronization is achieved through GNSS-based disciplining of receiver clocks through a dedicated battery driven hardware module. The module also determines the position of the receiver and provides an ultra-low power radio interface for communicating TDoA measurements in real-time to a central frontend computer for data processing and presentation. The main tasks of the project will be to develop embedded software for the hardware module that exploits its extreme low-power capability (including GNSS and wireless communication interface) and a flexible frontend solution that allows real-time estimation and visualization of transmitter position, as well as relaying position data to other relevant systems (e.g. an underwater vehicle).

Figure: Localization of a fish in a fish cage (3D and horizontal heatmap).

Figure: Underwater localization of "turtle-robot" UCAT on a mission in fish cage.

IoF - Internet of Fish, online fish monitoring system for sea farms

This project is related to the project described above, but is focused on the development of innovative backend and frontend solutions inspired by the IoT paradigm. The embedded module controlling the acoustic receiver features LoRa wireless radio communication, a key IoT technology, that enables extremely efficient relaying of data from the underwater sensors. The project is mainly a software design project, with identification of use cases and user requirements, high level software design, and implementation of a suitable IoT backend and frontend application layer solution based on Internet technologies (e.g. mqtt, cloud computing, web services).

Ultra-low power electronic sensor tags for fish behavioural tracking

Electronic fish tags have benefitted vastly from the technological progress of microelectronics in terms of miniaturization, energy efficiency, MEMS and signal processing capacity. Together with the increasing availability of data from large-scale earth observation systems (satellites), modern electronic fish tags are currently pushing the frontiers of knowledge in fish movement ecology. The proposed project involves design and coding of a miniature embedded computer that integrates a combined ATT (Acoustic Transmitter Tag) and DST (Data Storage Tag) function. This combination that will allow the electronic tag to work as a traditional DST (sensor data logger) when migrating in the open sea beyond receiver coverage, while the ATT function gets switched on only when the fish dwell in coastal waters, fjords and rivers where receiver coverage is more likely. The transmitter will also function as an acoustic beacon that will significantly increase the likelihood of tag recovery, the primary weakness and impediment of traditional DST-based studies.

The tag platform should be designed to accommodate a variety of sensors, e.g. course of swimming (magnetometer), swimming activity (accelerometer), inclination of geomagnetic field (magnetometer), water temperature (thermometer), swimming depth (pressure sensor), and physiology (ECG, pulse oximetry and plethysmography), and being able to store logging data for extensive time periods in non-volatile memory. The project will be based on a prototype tag platform and may be adapted to current needs in associated research projects.

Figure: Electronic fish tags - miniaturized embedded computers.

Lobster Robot - Computer vision high-throughput behavioural phenotyping system for lobster juveniles

Intensive aquaculture production of the European lobster has proven notoriously difficult accomplish. A primary reason for this relates to the agonistic behaviour (aggression, cannibalism) lobster frequently shown against conspecifics. This makes lobster rearing using the common approach of communal tank cultures unfeasible. It has been proposed that careful selection and breeding on individuals exhibiting more docile behavioural traits, if they can be identified efficiently, could provide a way around this difficulty. However, automatic monitoring and assessment of behaviour as well as selection among thousands of individuals (high-throughput behavioural phenotyping) over extended periods of time, remains a daunting task, which can only be solved through extensive use of robotization. This project deals with the development of a prototype of a high-throughput behavioural phenotyping system for lobster juveniles and incorporates elements of a variety of technological disciplines such as robotics, computer vision and real-time control. It will be possible to define several student assignments within the frames of this project depending on student interests/skills and project requirements. A mechanical prototype of an xyz-gantry robot with machine vision cameras and manipulation tools has already been realized and will be available as a test bed for the student project(s).

Figure: Lobster juveniles fighting.

Figure: Gantry robot for automatic phenotyping and selection of lobster juveniles.