Neural network-based detection of mechanical, sensor and biological faults in deep-trough hydroponics

In this work, two separate fault detection models are developed: one for the detection of faulty operation of a deep-trough hydroponic system which is caused by mechanical, actuator or sensor faults, and one for the detection of a category of biological faults (i.e. specific stressed situations of the plants), namely the "transpiration fault". The neural network methodology was proved to be successful in the task of fault detection, in both applications. The general fault detection model was capable of detecting a faulty situation in a very short time, in most cases within 20 or 40 min. In the case of the biological fault detection model, the "transpiration fault" was generally detected within 2-3 h. The results show that both neural networks have useful generalization capabilities. The influences of the conditions of the plants to the measured root zone variables are also investigated and they show that biological faults in general cannot be detected in this kind of cultivation systems using the measurable variables used in this work. The most probable explanation is the inertia of overwhelming mass of the nutrient solution (compared to the mass of the plants), in a way that it becomes a limitation factor to the influence of plants condition to their root zone microenvironment.