Fault Diagnosis by LAMDA methodology Applied to Drinking Water Plant

Water industry is facing increased pressure to produce higher quality treated water at a lower cost. The coagulation-flocculation unit is a major step in the production of potable water, allowing the removal of colloidal particles and contamination sources. Currently, the coagulant dosing is manual and it is determined by the way of jar test (off-line dosing tests). This test cannot be applied to supervision and diagnostic real-time. In order to obtain a simple model to describe the water treatment plant, a behavior model sets out, from the analysis of raw water characteristics to the entrance of the plant: (1) prediction on-line the amount of optimal coagulant dosage (software sensor) and (2) the determination of the functional states in real time. The innovative aspect of this paper resides in the design of intelligent control techniques of information water treatment plant (fuzzy logic for the classifier and neuronal networks for the software sensor).