Detecting pneumatic failures on temporary immersion bioreactors

Abstract

Temporary immersion bioreactors are an effective procedure to increase plant multiplication rates. The pneumatic system is an important part of a bioreactor, which should be controlled to guarantee both the efficiency and efficacy in the system. Therefore, bioreactors have been automated using a pneumatic drive to execute the immersion time. Sometimes, the pneumatic system presents failures which can affect the plant quality; therefore, pneumatic failure detection is an important task. Since failures are a few compared with the normal behavior, it is a class imbalance problem. In this paper, we study the use of contrast patternbased classifiers, designed for class imbalance problems, for creating an understandable and accurate model for detecting pneumatic failures on temporary immersion bioreactors. Our experiments over eight real-world databases show that a decision tree ensemble obtains significantly better AUC results than other tested classifiers.