Developing Data-Driven Models for Real-Time Dredging Production Optimization

Boskalis is currently looking for a student to start their graduation project with us. Dredging vessels rely on trained crews to configure dredging process settings that directly affect production output, a task that involves numerous interacting physical processes and a wide range of operating conditions. We currently use a machine learning model to recommend operation settings to crews in real time, but there is clear room for improvement. Physics-informed machine learning, time-series modelling, and adaptive control methods are promising directions worth exploring, as they can capture relationships and patterns that the current approach does not yet take advantage of.

The research project will require the student to conduct a literature review across these fields and determine which methods best fit our data and problem. Key design choices need to be made, such as how to combine physical knowledge with data-driven models, and how to validate improvements using historical data. With a variety of techniques available, an educated choice must be made to effectively improve our existing system.

In this research you will

• Work with other data scientists and meet with multi-disciplinary engineers to understand the challenge.
• Conduct literature research on physics-informed machine learning, time-series modelling, and adaptive control, and how they can be applied to this problem.
• Formulate the problem in a way that combines physical knowledge with data-driven models.
• Develop methods to improve the model’s recommendations and demonstrate how they generalize across different vessels and conditions.
• Report your final findings, demonstrating your design choices and reproducibility of results.

• You are doing a university master’s degree in computer science, AI, mathematics, physics, or a related field.
• Affinity with Python and version control.
• Able to develop, train, and evaluate ML models.
• Theoretical understanding of physics-informed ML, time-series modelling, or adaptive control; practical knowledge is not required but preferred.
• Proficiency in writing reports and findings, and presenting results.
• It is preferred to work at our office in Papendrecht minimally 3 days a week.