The text below is a brief description of the project topics supervised by professor Lars Imsland. 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.

Compressors operating in parallel is frequently found in the oil and gas industry. This constellation of compressors are typically called a compressor station. The parallel operation is motivated by capacity, availability or both. The capacity motivation is simply due to the compressor station is required to handle a higher flow rate than can be handled by a single compressor. The availability motivation is to avoid the need to stop the entire plant production in case of a malfunction for a single compressor. The work involves:

1. Literature review and documentation.
2. Modeling of a single compressor system consisting of a compressor and recycle line. The recycle line contains a control (anti-surge) valve for manipulating compressor flow and a liquid/gas heat exchanger with a control valve on the liquid side for manipulating the recycle gas temperature.
3. Simulation of the single compressor system by use of SISO (PID) control for compressor flow and recycle temperature. The compressor speed is set based on SISO (PID) control, by use of a pressure flow cascade, for compressor suction pressure.
4. Simulation of parallel compressors by combining single compression systems. The compressor speed is the same control structure as previous item, but there is now one pressure controller for all compressors and one flow controller per compressor. The flow desired by the pressure controller is distributed evenly on the compressors.
5. Develop a control strategy for minimizing the power consumption for the compressor station. In other words, how shall flow be distributed on each compressor in order for the compressor station to have a minimum power consumption.

Co-Supervisor: Dr. Bjørnar Bøhagen, ABB

Model Predictive Control (MPC) is a control strategy that optimizes the future operation of a dynamic system given the prediction of future operating conditions. While this technique is well established in many disciplines, MPC is relatively new in the field of buildings and heating systems. For this application, the set-point indoor temperature for the heating of the building is adapted as a function of the weather data prediction (typically the outdoor temperature for the next days) or the day-ahead hourly price for electricity. It typically enables to minimize the energy costs or the peak power. The specialization project proposes to investigate the potential of MPC for Norwegian residential buildings heated by electric radiator and, if the student has enough time, heat pump systems.

The project will be done in collaboration with the Department of Energy and Process Engineering at NTNU where a paired specialization project will be proposed. The project at this department will focus on the MPC controller and constructing an appropriate model for MPC from an advanced building simulation tool (IDA-ICE), while the student from Energy and Process Engineering will work on building simulations in IDA-ICE and focus more on building physics and HVAC.

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 exact topic will be decided later, but will be related to control and optimization (model predictive control, machine learning, data-driven production optimization, …), of processes in offshore oil and gas production. Some financial support for the stay in Brazil is available.

Se drillbotics.com. NTNU har gjort det svært godt i denne internasjonale konkurransen som går ut på å bygge en skalamodell av en virkelig boreprosess, som skal fungere helt autonomt. Men det er store muligheter for å forbedre regulering og estimering for å få boringen til å gå enda bedre og mer autonomt. Oppgaven er i samarbeid med Institutt for geovitenskap og petroleum, og følges med stor interesse av selskap innenfor petroleumsindustrien.

See this document for information.

Co-supervisors: Giancarlo Marafioti (SINTEF) and Camillo Bosco

Many algorithms for distributed optimization of structured convex QPs arising from distributed MPC problems have arised over the last years. This problem should investigate the use of «dual ascent» for convex equality-constrained problems combined with active set strategies for such problems.

See this text.

Co-advisor: Francesco Goia, Dept. of Architecture

See this text.

Co-advisor: Francesco Goia, Dept. of Architecture

Se dette dokumentet for detaljer. For mere informasjon, kontakt professor Jäschke og/eller meg.

See this document for information.

Co-supervisor: Leif Erik Andersson.

See this document for information.

Co-supervisor: Leif Erik Andersson.

See this document for information.

Co-supervisor: Leif Erik Andersson.

See this document for information.

Co-supervisor: Leif Erik Andersson.

See this document for information.

Co-supervisor: Leif Erik Andersson.

See this document for information.

Co-supervisor: Leif Erik Andersson.

Click this link for more information.

Co-advisor: Eric Bradford

Click this link for more information.

Co-advisor: Eric Bradford

See this document for information.

Co-Supervisor: Dr. Bjarne Grimstad, Solution Seeker

1-3 studenter.

DNV GL Fuel Fighter participate in the international competition Shell Eco-Marathon, where the challenge is to create the most energy-efficient car. This year we have a focus on optimization and want to exploit sensor data and IoT to create an optimized driving strategy. This is a very open problem that can be tackled in many ways and levels of complexity.

Sensors to capture power consumption, track position, speed, gearing, regenerative braking and more can be placed and used to gather data. Information can be sent through telemetry such as FiPy using MicroPython, then optimized with mathematical models or machine learning tools such as SciPy or Tensor Flow, and sent back as directions to the driver dashboard. The task also includes setting up an efficient motor control system to handle the electromotor and a linear actuator for a potential gearing system. Further, the optimization can be done to create an optimal driving strategy before the race, but could also include a real-time optimization during the race itself. In writing this project you are also automatically a member of DNV GL Fuel Fighter, and will work in a multidisciplinary team environment.

Mer informasjon kommer.

Medveileder: Professor Nadi Bar, institutt for kjemisk prosessteknologi

There are several opportunities linked to the joint NTNU-industry Research and Innovation Program in Digital and Automation Solutions for Oil and Gas industry (www.ntnu.no/igp/bru21).

Oil and gas production can be subject to large uncertainties. During production, such parameters as water and gas contents, reservoir pressure as well as other operational parameters change over time. In many cases, models of the complex interaction between the fluids/gas from the reservoir and injected gas/chemicals from artificial lift systems can be quite inaccurate. This makes the traditional model-based approach not very suitable for optimization of production from such systems. Therefore, in this project we consider an approach based on model-free optimization. This approach, also called extremum seeking control, will be applied to optimization of a production system consisting of several wells with artificial lift. (Artificial lift is a technology for enabling economically reasonable flow rates from wells which do not flow naturally. Examples: gas-lift systems, ESP-lift systems; ESP=Electric Submersible Pump). The challenge is to take into account all the practical aspects arising in application of extremum seeking control to oil and gas production (e.g. lack of measurements, noise, process disturbances, etc). The project will involve setting up the automatic optimization system for a production system with artificial lift, coupling it to a simulator (commercial or simple self-made) and making it work under realistic assumptions.

Optimization of oil and gas production has two time scales: slow (reservoir management level) and fast (daily production optimization). Traditionally, reservoir management/optimization is run on a scale of years. It is done through heavy/complex models and thus it is subject to a lot of uncertainties, as not enough measurements/information is usually available to calibrate the models in a good way. In daily production optimization, the time scale is at the level of hours/days, a lot of sensor data (online and historic) is available, and the uncertainty level is these measurements is relatively low. The question addressed in this project is how to unite together the slow model-based reservoir management and optimization with a lot of uncertainties with the fast data-driven production optimization having relatively low uncertainty. The potential solution is to have production optimization for a mid-term (scale of months instead of years or hours) and find a way to handle the uncertainties in a smart way. The project will involve setting up a simulator (possibly a commercial reservoir simulator), finding a method on identifying reservoir/well dynamic performance from the available sensor data (possibly with simple models) for prediction over a mid-term production period (e.g. months) and developing an optimization algorithm based on these predictions.

This project will investigate select different methods for robust linear MPC. Some keywords are (striped, parametrized) tube MPC and robustly reachable sets. These should be tested and compared on relatively simple examples. This project will fit best for «theoretically inclined» students with fondness for optimization and Matlab programming.