PhD Position Learning and Control for Complex Large-Scale Systems with Applications in Greenhouses

Delft University of Technology; Delft; On-site; EUR 30,000 - 42,000; Full time

Job summary

Delft University of Technology invites applications for a PhD position focused on learning and control in complex large‑scale systems, with applications to autonomous greenhouse control within the GreenControl project.

You will develop reduced‑order, data‑driven and model‑based control methods for climate regulation in controlled environments, study Koopman operators, and validate results on greenhouse demonstrators.

Qualifications

• Completed MSc degree in systems and control, applied mathematics, engineering, or a related field.
• Strong background or interest in systems and control, applied mathematics, machine learning, and affinity with biological systems applications.
• Some experience conducting, designing, and/or managing experiments for physical or biological systems is preferred.

Responsibilities

• Methods for reduced-order hybrid model learning, namely control-oriented and transferable models of airflow dynamics.
• Optimal sensor and actuator placement for cost/benefit trade-offs and improved ventilation.
• Hybrid data-driven and model-based control algorithms for online decision making.
• Data-driven predictive control design of PDEs based on Koopman operators and/or SINDy for ventilation strategies.
• Research trials and experiments to validate and iteratively improve the control design.

Skills

Systems and control

Applied mathematics

Machine learning

Biological systems applications

Experiment design/management

Education

MSc in systems and control

MSc in applied mathematics

MSc in engineering

Job description

This PhD position focuses on developing methods for learning and control in complex large‑scale systems, with applications in autonomous greenhouse control within the GreenControl project.

Job description

Recent trends in Controlled Environment Agriculture (CEA) development, design, and operations related to energy saving and greenhouse gas emission reduction are mainly focused on the ventilation process, which typically considers the use of window and mechanical air treatment based on average climate measurements. Airflow affects crop transpiration, growth, development, yield and quality, but despite its importance, related control strategies in practice are often very crude and rule-based without incorporating complex plant/microclimate interactions or economic considerations. State‑of‑the‑art approaches in optimal climate control of greenhouses are based on implementing economic objective functions exploiting a time scale decomposition between short‑term climate control and energy use, and long‑term crop management goals. While several algorithms have shown promising results in energy savings and crop yield, most of these methods have only been tested in simulation and make use of average climate measurements, which are then used to control the overall climate setpoints. Awareness of micro‑climate insight and fine‑grained, model‑based control of locally applied ventilation is lacking in these approaches.

This PhD position aims at developing methods for learning and control in complex large‑scale systems. This will be carried out as part of the GreenControl project, whose primary objective is to address these shortcomings in autonomous greenhouse control. The project team includes PhD students and researchers at TU Delft, Wageningen University, University of Twente, and TU Eindhoven, as well as industrial partners that specialize in greenhouse design and installation, plant breeding, climate control, sensing and monitoring with microdevices, software development, and technology provision for high‑tech greenhouses.

The goal of the GreenControl project is to support the transition to climate neutrality in CEA by using a new operating philosophy that puts plants at the center of the control strategy to achieve 25% energy savings and 35% reduction in energy costs. This is accomplished by moving from average climate control to direct crop‑centric control. This paradigm shift relies on breakthroughs in microclimate sensing, interpreting crop performance by integrating sensor data at different temporal and spatial scales into a crop modelling framework, and predicting daily targets for photosynthesis and transpiration rates. Based on these targets and fluctuating electricity prices, the main objective will be to develop a control‑oriented model and algorithm to alter lighting, CO2 dosing, and air circulation to satisfy crop needs while minimizing resource use and costs. The developed control scenarios will investigate targets with increasing complexity, including daily respiration target, daily photosynthesis target, and cost and energy use optimization, and will be improved iteratively culminating in validation trials and experiments.

In the GreenControl project, the primary aim is to demonstrate the added value of capturing microclimate effects via new sensing and modelling approaches and adjusting control strategies to improve photosynthesis efficiency while reducing overall energy use. Research by project partners will show how crops respond to microclimate setpoints based on detailed sensing data and microclimate and crop models, obtained through mechanistic and CFD‑based approaches. These provide the basis for learning reduced‑complexity control‑oriented models that will be exploited by two control strategies:

1. Airflow control by forced convection to improve photosynthesis efficiency by CO2 delivery to the leaf surface and to couple crop and climate regulation more tightly.
2. Energy‑saving strategies where the application of lighting, active ventilation and heating are optimized based on plant performance and energy price fluctuations.

The learned models and resulting data‑driven control approaches will be designed to be more easily transferable between different greenhouses than constructing detailed CFD models for each location.

Expected innovations include a hybrid data‑driven and model‑based predictive control approach that uses Koopman operators, plant imaging, and 3D microclimate sensors to enable crop control instead of only indirect climate control. The Koopman operator formalism offers applicability in data‑driven settings for the analysis and control of large classes of nonlinear and high‑dimensional systems, such as airflow dynamics in complex greenhouse environments.

In this PhD project, you will explore and conduct research at the intersection of learning theory, PDEs, and systems and control, likely using RKHSs or similar function spaces, Koopman operators, and neural networks to study classes of controlled PDEs, develop suitable learning schemes, and design control policies accordingly. You will use large‑scale optimization for implementation of the obtained results, first using high‑fidelity numerical simulations, and then implementing and verifying them on a greenhouse demonstrator.

Main research and development tasks

• Methods for reduced‑order hybrid model learning, namely control‑oriented and transferable models of airflow dynamics, including CO2 level, temperature, and humidity, using microclimate sensor and CFD simulation data from other researchers.
• Optimal sensor and actuator placement, aiming at cost and benefit trade‑offs of sensor and actuator layouts, for example for improved ventilation.
• Hybrid data‑driven and model‑based control algorithms for improved online decision‑making, for example for ventilation performance, possibly using the learned reduced‑order models and the spatially distributed sensors.
• Data‑driven predictive control design of PDEs based on Koopman operators and/or relying on sparse identification of nonlinear dynamics (SINDy) for model predictive control, for example for adapting ventilation, including on/off fan operating schedules, heating, artificial lighting strategy and screens, while accounting for energy price fluctuations.
• Research trials and experiments to validate and iteratively improve the control design, supported by consortium partners.

Job requirements

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

TU Delft (Delft University of Technology)

Working at TU Delft means contributing to solutions that really make a difference. At TU Delft, you will have the opportunity to take initiative, work with others, and grow as a professional in an international community of professionals and students.

Faculty Mechanical Engineering

The Faculty of Mechanical Engineering is a dynamic and innovative faculty with high‑tech lab facilities and international reach. Its broad disciplinary scope enables unique connections across fields, reflected in its education and research environment.

Conditions of employment

Doctoral candidates will be offered a 4‑year period of employment in principle, in the form of 2 employment contracts. An initial 1.5‑year contract includes an official go/no-go progress assessment within 15 months, followed by an additional contract for the remaining 2.5 years assuming performance requirements are met.

As a PhD candidate you will be enrolled in the TU Delft Graduate School. The Graduate School provides an inspiring research environment with an excellent team of supervisors, academic staff and a mentor. The Doctoral Education Programme is aimed at developing your transferable, discipline‑related and research skills.

TU Delft also offers a customizable compensation package, discounts on health insurance, a monthly work costs contribution, and flexible work schedules.