Sand filters are an essential part of the processes of treatment and purification of water. Just in the Netherlands, about 20000 sand filters are used just for the purification of drinking water and their use for the treatment of waste water is increasing. As part of the process of purification of drinking water, sand filters are used to eliminate particle in suspension and/or iron, ammonium, manganese, bacteria, with or without the help of chemical and biological processes taking place inside the filters.

Although sand filters have been used for many years and have been the subject of many studies regarding dimensioning and operation, there is little known regarding what are the mechanisms taking place inside the filter. The reason is simple: it is not possible to look inside the filters. In particular, a detailed understanding of the mechanisms of cleaning (backwash) of the filter is missing.

The project FilterExpert is a collaboration of parties involved in the water treatment business (water companies, consultancy companies) and research parties, aiming at significantly improving the functioning (quality of product and cost of operation) of sand filters by modeling the mechanisms, allowing knowledge of the condition of a filter and derivation of an automated control model. It consists of 3 parts:

• Development of techniques derived from acoustical remote sensing to obtain a tri-dimensional image of content of the filter (tomography).
  
• Development of a model of the mechanisms taking place inside the filter and soft-sensors allowing insight into the internal state of a sensor and its control.
  
• Development of improved control processes based on the knowledge obtained from the preceding points.

In this project, Thales is involved, via de D-CIS lab, in the development of an advanced data analysis system allowing fusion of historical data and operator knowledge,

The fusion system will coupled with soft-sensors in order to determine the current condition of the filter and predict the efficiency of the operational actions.
To achieve this goal, we apply Bayesian techniques allowing (i) fusion of various types of information from different sources and (ii) integration of historical data and diverse sources of knowledge including heuristics derived from operations and practical experience.

Consequently, filters are operated based on a few external measurements and a lot of not formalized experience and historical knowledge. In addition, improvements in the operations until now have been mostly directed at improving the water quality in the output and trial and errors approach were used to optimize the operation of filters on an individual basis.
The conjunction of these factors results in high costs of operations of sand filters, work-intensive operation in some cases, and dependency towards historical knowledge of each installation making automation of operation impossible.