Challenge
Our goal for the 21st century is to achieve food production with optimal use of resources, minimal environmental impact, sufficient production capacity, and maximally efficient human labour input. There is continuous development in precise and efficient management, and automated control through precision technology and machine intelligence. However, food production systems are particularly challenging due their extreme complexity, fragility, variability, and limitations in observability and controllability.

Approach
The research in my group focuses on observation, prediction, and control methods to understand how to optimize performance by improving machine intelligence and precision technology. The key is to combine knowledge from biophysical models with information available from sensing technology, and capability of actuation technology.
Expertise
Our methodology includes biophysical modelling, feedback control, model predictive control, state and parameter identification, optimization, and uncertainty analysis.

Our key domain expertise is on protected horticulture (greenhouse, vertical farm), and animal systems.

Figure: Decision support for operational management. Two feedback loops are shown: a low-level management loop (with a low-level controller) with input consisting of the set-points provided by users, and a high-level management loop, in which users make decisions on settings and set-points based on sensor information, forecasts and decision support (observation, prediction or control advice).

I am involved in the following research projects:

LED it be 50%  (led by Leo Marcelis, Horticulture and Product Physiology)
The aim is a 50% energy reduction in greenhouses via smart LED light management. We investigate possible ways to control LED light and other climate factors in a cost effective manner. http://www.stw.nl/nl/content/p13-20-save-led-it-be-50

Energy saving in greenhouse crop production by flexible management
The goal of this project is to develop greenhouse climate management support that substantially saves on gas, by employing data streams on weather and on the energy grid. This type of management will at the same time reduce the peak loads on the electricity grid, thereby helping the transition to cleaner energy. https://www.nwo.nl/onderzoek-en-resultaten/onderzoeksprojecten/i/50/30550.html

Sensing and modelling morphological traits and social interactions to identify vulnerable cows in dairy herds Together with Lely Industries, we focus on the following objectives: 1) to sense and model morphological traits and social interactions in dairy cows, 2) to identify vulnerable cows based on the dynamics of model outputs to improve dairy herd management.

Veerkracht 2 Together with Wageningen Livestock Research we focus on improving animal welfare by predicting the resilience of cows during vulnerable periods. We do this by integrating experiment and statistical testing, in order to investigate which signal, or combination of signals, are key for predicting vulnerability.

Low-cost low-risk irrigation under uncertain agricultural circumstances Various types of unknown or unpredictable variation, limit the manageability of farming processes, making them inefficient and costly. This causes risk avoiding behaviour of farmers, e.g., in the form of excessive over application of water, fertilizer, pesticide, and antibiotics, with drastic consequences for our environment. As a case study, we investigate how to optimize irrigation under uncertainty with respect to water stress in lettuce. We consider various way to optimize; from irrigation scheduling to hardware innovation.

Modelling uncertainty in controlled environmental agricultural systems Inaccurate climate sensing in greenhouses are associated with considerable energy loss due to mismanagement. Increasing the number of sensors maybe costly and impractical. We investigated the added value of diagnosing the climate state via filtering sensor data using a climate model and a noise model.

The following courses are aimed at students that conduct research of biosystems under variable and uncertain circumstances.   

1. Data analysis (FTE-26306) for BSc students, aimed at finding and analysing data driven evidence. The course is inspired by the Advanced Statistics courses, and tailored to biosystems engineering applications. The underlying model framework is a general linear model, built solely upon data.
2. Machine learning (FTE-35306) for MSc students. This course is an introduction to automated learning in order to make diagnosis and predictions. The underlying modeling framework is consist of various nonlinear structures, but all are black-box (not based on underlying principles).
3. Precision Farming (FTE-35806) for MSc students. In this course we build upon classic control engineering theory, and design ways to create methods for precise dosage, timing, and allocation of inputs. In particular, the focus is on prediction, diagnosis, and control. The model framework consists of nonlinear differential equations.
4. Statistical uncertainty analysis of dynamic models: PhD tutorial. Introduction to uncertainty propagation in dynamic input-state-output models. The model framework is general, to allow nonlinearity, and first principle models. The latter is important for investigating the underlying mechanisms of a system, management design, and systems design. For more info, see https://www.pe-rc.nl/postgraduate-courses/uncertainty-analysis-dynamic-models