Fault diagnostic method for a water heating system based on. continuous model assessment and adaptation

Sylvain Lalot[1]*, Soteris Kalogirou[2], Bernard Desmet1, and Georgios Florides2

1LME, Universite de Valenciennes et du Hainaut Cambresis, Le Mont Houy,

59313 Valenciennes Cedex 9, France

2Cyprus University of Technology, P. O. Box 50329, 3603 Lemesos, Cyprus
* Corresponding author, sylvain. lalot@univ-valenciennes. fr

The objective of this work is the development of an automatic solar water heater (SWH) fault diagnostic system (FDS). The latter consists of a modelling module and a diagnosis module. A data acquisition system measures the temperatures at four locations of the SWH system (outlet of the water tank; inlet of the collector array; outlet of the collector array; inlet of the water tank). In the modelling module a number of artificial neural networks (ANN) are used, trained with the very first values when the system is fault free. Then, the neural networks are able to predict the fault-free temperatures and compare them to actual values. When the differences are low, the corresponding networks are unchanged. On the contrary the networks are retrained. Then the diagnosis module analyses the difference between the current connection weights and the initial weights. When a persistent significant modification occurs, a flag is set to signify that a default is present in the SWH.

The system can predict three types of faults: collector faults and faults in insulation of the pipes connecting the collector with the storage tank (to and from the tank) and these are indicated with suitable labels. It is shown that all faults can be detected well before the end of the drifts, without any false alarm, when the networks and thresholds are well tuned and that the observation window has the right size. It is shown that this does not depend on the draw off profile.

Keywords: fault diagnostic, model adaptation, neural network, water heating system

type. As it has been shown that continuous drifts can be analysed by neural networks in heat exchangers [3-5], ANN has been chosen here to test such tools. In particular, a method based on neural models is presented according to the study detailed in [6] which shows that a continuous assessment of a model and its adaptation is efficient. In a first part, the solar system is presented along with the drifts that are taken into account. The drift detection tool is detailed in the second part, and results are given in the third part.