Graph neural networks
To join our team in Science Disciplines, we are looking for a student on the topic:
Graph neural networks
ROSEN uses a fleet of tools to perform inline inspection of pipelines. Each tool is specialized to different environmental conditions (e.g. pipeline size), leading to differences in sensor-in-pipeline locations between tools. Reconstructing the location of sensor in pipeline coordinates is an important step in automated data analysis. Furthermore, the resulting (potentially) irregular grid needs to be standardized to allow the application of machine-learning algorithms that assume fixed input sizes.
This project aims to explore whether the coordinate-remapping step can be moved into the machine-learning algorithm for data analysis. The idea is to represent the measurement as a graph of measurement locations in tool coordinates, annotated with measurement values and tool information such as location, velocity and orientation. The task in this project is then to design a graph neural network that can translate the measurement graph into the pipeline graph; entailing that the network successfully solves the sensory inversion task.
The resulting network should be evaluated via appropriate accuracy metrics that measure inversion quality. Additionally, the behavior of the network is to be characterized with respect to perturbations of the tool position, imprecise tool coordinates and dependence on tool geometry.
This thesis is being announced in cooperation with Prof. Dr. Martin Atzmüller, Osnabrück University.
Challenges for the thesis:
- Define in and output graph structures and preprocess existing datasets into usable format.
- Design and train graph neural network
- Evaluate performance of network
RequirementsTo become part of the ROSEN family, you convince with your ability to work in a team and are used to working independently. Moreover:
- Master's student in Computer Science, Cognitive Science or similar fields
- Interest in machine learning and statistics beyond convolutional-neural networks
- Good coding skills in python
- First experience with Deep Learning frameworks (e.g. pytorch or tensorflow)
Our OfferWe offer insights into the work of an international, innovative and long-term oriented group of companies. In an open corporate culture with fast decision-making processes, you can successfully implement your ideas. We will also support you in the following areas as part of your final thesis:
- Know-how development in the mentioned topics
- Individual supervision and independent work
- Exchange with other students
- Invitation to exclusive ROSEN events
ROSENis a leading privately owned company that was established in 1981. Over the last 4 decades, ROSEN has grown rapidly and is today a worldwide technology group that operates in more than 120 countries with almost 4000 employees.
ROSEN offers sophisticated and highly innovative products and services to the oil and gas and other engineering industries. ROSEN is an extended team of people with a passion for technology and innovation. We are always looking for young professionals as well as experienced employees.
Our ongoing organic growth results in career opportunities and gives our employees chances for further development and added experience.
For more information about ROSEN go to www.rosen-group.com.