Two AI methods for classification of water pipes damage

Abstract

The technical condition of municipal infrastructure is continually in the centre of attention. It is necessary to maintain water management in stable conditions. Many failures of water pipes could lead to destabilize the whole water management in the cities. Not only the number of damage, but also their kinds should be analysed using technical, economical and reliability analysis. The comparison of classification results of kinds of water pipes damage using two artificial intelligence methods (classification trees and neural networks) was presented in the paper. The aim of the work was to check if classification accuracy using multilayer perceptron is higher than in other original investigations carried out several years ago when classification trees were analysed. Exploitation data from water supply system were used for modelling purposes. Several configurations with different size of input vector were investigated. Obtained results are not satisfactory. Only the most numerous classified in approx. 80%. For other kinds of failures the classification accuracy was minimal. It is necessary to change, in the next work stages, the investigation approach of quality variables prediction.