Prediction of the Filter Life Cycle Based on Artificial Neural Networks

Abstract

This paper aims to show the realization of the system based on artificial neural networks application in the process of monitoring of water filtration procedure, early prediction of filter damage and initial activation of self-cleaning, which is necessary to carry out so that the system would function properly. The control system proposed involves two artificial neural networks which completely define parameters of the process state. Capability of filter self-cleaning gives a possibility of significant autonomous work. The most important characteristic of the filter is the change of differential pressure in the function of water flow, and as this is a nonlinear function, the choice of such supervising system and control process is justified. The learning algorithm used in this nonlinear mapping was back-propagation within the BPnet software. Considering the fact that the system permanently does the acquisition of information about the system state, it is possible, by using data about "rainy days", to define correlation between the characteristic of filter operation and outward atmosphere factor.