Below are brief descriptions of the project topics supervised by professor Lars Imsland, typically together with others. Contact lars.imsland@itk.ntnu.no for further information. Most topics can be continued in a master thesis project if the candidate and supervisor(s) so wishes. Some of the descriptions are rather advanced/involved for a single project, and should be considered descriptions for a combined project/master project.

Virtual flow metering (VFM) is a technology for real-time prediction of flow rates in process systems. A virtual flow meter makes use of mathematical models and ancillary/existing measurements to infer flow rates by computation. It can be a viable and cheap alternative to installing flow rate measurement devices, especially for subsea process systems. Today, VFM technology based on mechanistic models is widely used in the petroleum industry. A drawback with mechanistic VFM is that it can be hard to maintain for changing process conditions. Recent research results show that VFM based on neural networks can predict flow rates with high accuracy. A key feature with neural network-based VFM is that it learns from the data stream and is able to quickly adapt to changing process conditions without human intervention (maintenance). The Norwegian company Solution Seeker – a spin-off from Engineering Cybernetics at NTNU – is now leading the R&D efforts to provide the first neural network-based VFM to the industry. Several oil companies are already using the technology. Solution Seeker is looking for 1-2 candidates to contribute to the advancement of neural network-based VFM. Assignments will be focused on exploring neural network architectures for VFM, including hybrid architectures which combine neural networks and mechanistic models. Solution Seeker will provide the candidates with supervision, datasets and machine learning infrastructure. If circumstances permit, there may also be an opportunity for a stay abroad at Solution Seeker’s offices in Rio de Janeiro, Brazil.

Supervisors: Prof. Lars Imsland, NTNU; Bjarne Grimstad, NTNU & Solution Seeker; Mathilde Hotvedt, NTNU & Solution Seeker

About Solution Seeker: Visit solutionseeker.com.

Traditional optimization of systems and processes is based on models. It is known that in a number of applications, occurring in particular in oil and gas production systems, this traditional approach can be inefficient due to large model inaccuracies and uncertainties. In this case one needs to apply data-driven optimization methods, like Extremum Seeking Control (ESC) approach. For complex production systems consisting of multiple elements (or agents), this approach, however, suffers from a practical limitation – it can generate large disturbances that make the practical application of ESC infeasible (even though it works in theory). In this project we are focusing on developing data-driven optimization methods that do not suffer from this limitation and that can lead to feasible data-driven optimization solutions for complex production systems. The project will consist on developing such an algorithm and implementing it in MATLAB and simulating it for a complex oil production system.

The project is co-supervised by Prof. Alexey Pavlov (IGP) and Prof. Lars Imsland (ITK) and will be a part of the BRU21 program (www.ntnu.edu/bru21).

Haptic feedback is usually applied to controlling robotic systems or mobile units. It allows a human operator to feel what is happening with the controlled robot by means of feeling of touch and/or force. For example, a pilot can feel the airplane through the resistance force that (s)he feels in the control stick or pedals. Haptic feedback is also very important in medical robotic applications. In this project we will focus on transferring the haptic feedback concept to control of processes, in particular in the oil and gas industry. The work will include setting up a force feedback joystick (we have 2 different joysticks available), integrating it with a process simulator, implementing an automatic system for force feedback generation, coupling it with a process control system and tuning it. The objective is to achieve optimal performance of the overall system consisting of the process, human operator and automatic haptic feedback system. The final outcome of the project will consist of an experimental setup demonstrating the concept.

The project is co-supervised by Prof. Alexey Pavlov (IGP) and Prof. Lars Imsland (ITK) and will be a part of the BRU21 program (www.ntnu.edu/bru21).

Drillbotics is an international student competition on developing and implementing a laboratory-scale drilling rig for fully autonomous drilling (www.drillbotics.com). A multidisciplinary team of 3-5 students with expertise in Petroleum Engineering (drilling), Cybernetics and, possibly, Mechanical/Electrical Engineering will work for a year (specialization project + MSc project) on developing, building, programming and testing/tuning the drilling rig. This year the challenge is to make a fully autonomous system to drill a deviated well that hits targets with coordinates specified by the jury. The work includes designing, building/implementing and tuning of rig hardware, sensors, data acquisition system, algorithms for optimal trajectory generation, automatic control system, safety system, etc. It is a large-scale project being a replica of an industrial technology development project. By the end of the year, the team will participate in the competition, which, depending on the corona situation, will be held in Germany or by video in Trondheim. NTNU Drillbotics team took 2nd and 1st places in the finals in 2017 and 2018 and was put as #1 in pre-ranking in 2019 and 2021. According to the students from the previous teams, it was the best learning experience from their whole NTNU studies. The NTNU Drillbotics team is sponsored by BRU21 and Equinor. You can read more about NTNU Drillbotics here: https://pet.geo.ntnu.no/wordpress/igp/nb/category/drillbotics/.

The team is supervised by a team of professors from Departments of Geoscience and Petroleum and Engineering Cybernetics. The supervisor from ITK is Lars Imsland.

Related to a cooperation project we have with two top Brazilian universities (in Florianapolis and Rio de Janeiro), we seek 1-2 adventurous students to do a combined project/master. The specialization project (fall) will be in Norway, while the master thesis work (spring) will be done in Brazil. The topic will be related to machine learning and control for production optimization in offshore oil and gas production, for instance 'physics-informed neural networks' used for data-driven control of gravity separators or electrical submersible pumps. Some financial support for the stay in Brazil is available.

The greenhouse gas emission offshore accounts for a quarter of the total greenhouse gas emissions in Norway. To comply with the Paris Agreement, companies and the government alike have been developing solutions with the purpose of increasing the share of renewable energy in existing power systems. One of these solutions can be found in offshore wind, which is foreseen to have a massive capacity expansion in the coming years. Integration of intermittent renewable such as offshore wind with existing power sources such as gas turbine requires precise control strategies to enable high energy efficiency and stable electric-power supply. To this end, model predictive control (MPC) may be a preferred control solution as it enables the controller to account for future actions, while taking into account past information. A critical challenge by using MPC is to have a good model in order to predict the future. Producing a good model can be hard, as offshore wind is inherently noisy in the form of uncertain wind amongst many other uncertainties. As such, gaussian processes (GP)s have been proposed to capture the uncertainty in a MPC setting. GPs employs data driven methods to learn a model which in a MPC can be used as the model to predict the future, or simply to tighten or loosen the constraint to take advantage of the uncertainty information.

This project will expand on the existing mature literature of GP-MPC and apply it to an offshore hybrid power system, consisting of batteries, offshore wind and gas turbines. The candidate will learn how to implement nonlinear model predictive control and gaussian processes for a power split problem in either Python or MATLAB. Main outcome of this project for the candidate will be a controller which to some extent, will be able to control an offshore hybrid power system with different uncertainties.

The main tasks of the candidate will consist of:

• Literature review on the system (gas turbines, wind turbines and batteries)
• Literature review on the method (nonlinear model predictive control (NMPC), gaussian processes)
• Implement a full-state feedback NMPC assuming perfect uncertainty information (wind, battery parameters)
• Implement a full state GP-NMPC assuming imperfect uncertainty information, but with existing data
• Compare the nominal NMPC with the GP-NMPC given imperfect uncertainty information.

The main idea, is to have the candidate first implement the full-state feedback NMPC for the offshore hybrid power system assuming perfect uncertainty information in the project thesis. The master thesis will then build upon the project thesis by considering uncertainty and gaussian processes.

Co-supervisor: Kiet Tuan Hoang, ITK.

Tools: Septic (Equinor MPC), Modelica (Process simulator), Equinor Septic configuration generator, Visual Studio code, python unit test programming

Task: Start with Septic course (control of oil well simulated in Modelica). Develop methodology to generate and test an MPC application

1. Configuration file: use configuration generator. Demonstrate how to add a second well, add CV’s, MVs, etc based on reading an IO list
2. Unit tests: write unit tests verifying
   1. Communication
   2. Activate/deactivate MVs and CVs
   3. Calc: logic, e.g. turn on/of dead band when near set point
3. Scenario tests: make a simulation scenario and define a metric for performance. Write a test varying 2-3 tuning parameters to find the best tuning.
   1. Startup well
   2. Large disturbance
   3. Noise
4. Discussion:
   1. What can and what cannot be tested with unit tests?
   2. Way forward?

Co-supervisors: John-Morten Godhavn, Pål Kittilsen, Einar Idsø, Equinor