Anomaly detection and diagnosis in smart greenhouses

Maintaining suitable indoor conditions in a greenhouse requires the continuous operation of heating, ventilation, dehumidification and lighting — energy-intensive systems whose anomalies lead to energy losses, higher costs and reduced product quality. Anomaly detection is challenging: measurements are affected by natural fluctuations, noise and complex physical interactions, causing conventional methods to raise false alarms.

This doctoral project develops an integrated anomaly detection and diagnosis framework combining physics-informed probabilistic modeling, uncertainty quantification, causal reasoning and computationally efficient artificial intelligence — able to identify an anomaly's root cause: actuator fault, sensor fault, control problem or abnormal environmental condition.

Main objective

Develop and validate an intelligent framework for anomaly detection and diagnosis in greenhouses, combining physical-system knowledge, uncertainty quantification, causal reasoning and artificial-intelligence methods.

Methodology

Four stages: literature review; modeling of normal behavior via physical relationships and probabilistic learning accounting for aleatoric and epistemic uncertainty; causal representations for root-cause analysis; efficient AI methods validated on a physics-based simulator and a real test bench (precision, recall, false-alarm rate).