Fault Diagnosis System of CSTR Process Based on Stacking Classifier Algorithm
Clynton Roger Guastti de Oliveira1; Daniel Cruz Cavalieri1; Cassius Zanetti Resende1
1 Instituto Federal do Espírito Santo
doi:10.20906/CPS/CILAMCE2017-0664
Resumo
In recent years, fault diagnosis in industrial processes has been widely studied. Various methods based in artificial neural networks and statistical classifiers have been used to extract the root causes of faults. This paper present a classification performance comparison between a Support Vector Machine (SVM) classifier and an Artificial Neural Network (ANN) applied to solve a multiclass fault diagnosis problem. A chemical process simulator containing a Continuous Stirred Tank Reactor (CSTR) provides the analyzed syntactic data with linear and nonlinear faults. Additionally, a stacking algorithm is proposed to improve the classification accuracy by combining the two classifiers. Finally, a comparison between the three classifiers and their performance for classifying ten types of faults is presented, showing the improvement of the classification with the use of the stacking algorithm proposed.
Palavras-chave: Support Vector Machine; Artificial Neural Network; Continuous Stirred Tank Reactor; Stacking Algorithm; Multiclass Classifiers