Filtering of industrial data using the artificial neural networks
Andrzej Stanisławczyk, Jolanta Talar, Piotr Jarosz, Jan Kusiak · Computer Methods in Materials Science. · 2007
The form of registered data (superimposed by the measurement noise) and the lack of some data (difficulties in measurement of some process parameters) cause the problems in modelling and control of many real industrial processes. The filtering process of the data with the imposed noise is a complex problem and it is difficult to find appropriate general filtering method, which gives the reliable results. Sometimes, the filtering procedure eliminates important information, and sometimes leaves the unnecessary noise. Therefore, the registered data are often not suitable for the further analysis and modelling of the process. The main goal of the work is presentation of elaborated filtering procedure, which is able to eliminate these components of the output signals, which can not be predicted on the basis of the registered input signals. Proposed filtering algorithm gathers the advantages of different techniques: adaptive filters, Fourier Transform Method and techniques based on the Artificial Neural Networks. The paper presents the idea of data filtering system and the results of filtering of the industrial data of the copper flash smelting process. The filtered data can be used to work out a control system based on the Artificial Neural Network.