FAILURES DETECTION METHODS IN CHEMICAL PROCESS USING ARTIFICIAL INTELLIGENCE

Any atypical change in a procedure can be characterized as a "failure". Consequently, it may result in economic losses and/or a rise of the operational cost, because most of the times the process will need to be interrupted. Therefore, the concern with the quality and security of the processes has stimulating studies of diagnosis and monitoring failures in industrial equipments. In light of this, the present article has as purpose to apply three different methods (Artificial Neural Networks - ANN, Fuzzy Logic - FL and Support Vector Machine - SVM). All of those were applied as detection and classification systems of failure in the processes of a case study in order to diagnose these artificial intelligence techniques so that the efficiency of each method can be compared. All investigation is done by modeling a reactor of Van der Vusse's kinetic causing four types of failures, in the concentration of a reagent (failure 1), in the sensor which measures the concentration of the interested product and temperature (failure 2 and 3), and in the valve locking (failure 4). The data used in this methodology is based in quantitative and qualitative historical information. All methods are able to detect failures, but in different times. ANN is the one which detects faster all the failures. SVM detects some minutes later, however with good precision, even though this method uses less computational effort compared to ANN. Fuzzy, in the most of the cases studied, takes hours to detect any change in the system, which makes this one the less effective among the ones studied.