Prediction model condition-determination of cementitious mains – Data mining phase 2

Erwin Vonk MSc, George Mesman BSc

  • Start date
    01 Oct 2014
  • End date
    31 Aug 2015
  • Principal
    Brabant Water
  • collaborating partners
    Brabant Water

To further structure its existing mains replacement policy, Brabant Water wants to gain more insight into factors that play a role in the leaching of cementitious mains. It is possible, using a previously developed prediction model, to make a rough estimate of the extent of leaching in these kinds of mains. This provides a guide in the prioritisation of the replacement policy, while reducing the possibility that pipes that are still in good condition are prematurely replaced.

The objectives of this follow-up project are the following:

  1. Explore whether the existing (phase 1) prediction model for internal leaching of cementitious pipes could be improved through further dataset enrichment.
  2. Implement a method similar to the one used to predict internal leaching for the prediction of external leaching as well.

Detecting undiscovered correlations through data mining

One existing dataset of Brabant Water of phenolphthalein tests on cementitious mains is enriched with supplementary parameters, such as standards for water quality and residence times of the water in the distribution network. The enriched dataset is used to feed a so-called artificial neural network, after which, on the basis of all available (relevant) information about a pipe section, one can determine the degree of (internal/external) leaching in millimetres.

View on the limits of data mining

For both internal and external leaching, only marginal improvements were found in the predictive capacity of the existing model. This means that, on the basis of existing data sources, the limits of data mining seem to have been reached. Uncertainties remain with regard to the materials quality, lime-aggressiveness of the soil and the precision of the phenolphthalein tests themselves. With the limits of data mining clarified, the path is clear to focus on spatial extension, and on transforming the existing model into a practical system that can support asset management within Brabant Water.