PhD Position Physics-informed Learning and Layered Control Architectures fo

at  TU Delft

JOB DESCRIPTION

High-tech greenhouses play a crucial role in ensuring sustainable, affordable, and reliable local food production. The construction and operation of high-tech greenhouses is therefore expected to grow significantly over the next decade, but this increase is not matched by a commensurate increase in the number of growers capable of effectively managing them. Achieving high resource-use efficiency of the greenhouse, e.g., maximizing the kg of food produced per unit of energy, requires growers to consider a complex relationship between the short and long term consequences of each operational decision. These factors have lead to a significant interest in autonomous systems for greenhouse management.
Although autonomous greenhouse systems have achieved remarkable improvements in performance over the last couple of years, the current technology cannot be scaled to the entire greenhouse industry as current methods require either enormous amounts of operational data or high fidelity simulators. In addition, current autonomous greenhouse control systems focus almost entirely on the highest level of decision making, while ignoring the lower control layers that actually implement these high-level decisions. This approach leads to a control architecture that is poorly integrated and lacks flexibility in responding to short term perturbations, thereby producing suboptimal performance in terms of profits, resource efficiency, and carbon footprint.
This PhD position is a part of the LEAP-AI project that will address these challenges in autonomous greenhouse control. The project team includes PhD students and researchers at TU Delft, Wageningen University and Research, and Erasmus University Rotterdam, as well as industrial partners that specialize in machine learning, climate control, and computer vision for high-tech greenhouses. The goal of the LEAP-AI project is to collaborate with this team to design the next generation of autonomous greenhouse control systems and will culminate with several large-scale experiments including both research trials as well as validation/demonstration of the new autonomous greenhouse control system in an actual high-tech greenhouse.
In this PhD project, you will explore how physics-informed graph neural networks (GNNs) in combination with nonconventional sensors (such as computer vision) can be used to enable generalizable, explainable, and data efficient machine learning for crop-greenhouse systems. A particular focus will be placed on using these GNNs in predictive and optimization-based control methods with opportunities to explore tailored control formulations and optimization algorithms for optimal control with GNN models. The ultimate goal is to use these insights to redesign the control layers and communication strategy of current autonomous greenhouse systems to improve both the flexibility and explainability of the system.

JOB REQUIREMENTS

Applicants should have:

• Completed a relevant MSc degree in systems and control, engineering, applied mathematics, or a related field.
• A strong background or interest in systems and control, machine learning, and biological systems.
• Some experience conducting, designing, and/or managing experiments for physical/biological systems is preferred, but not required.

Please refer the Job description for details