Machine Learning Application for Oscillation Detection in Control Loops

Abstract

Oscillations in control loops pose a serious challenge to control engineers. Oscillations may even destabilize the control loops which in turn create a serious threat to the overall safety and efficiency of the plant. Further, there could be thousands of control loops and manually observing each one of them is impossible. Therefore, automatic oscillation detection tools are required by the control engineers in order to maintain and operate the plant at its optimal point. In this work, a new approach for automatic detection of oscillations using machine learning has been proposed, wherein an artificial neural network (ANN), in multilayer perceptron configuration, has been used as an oscillation detection classifier. A total of ten linear predictive coefficients were obtained from the industrial plant which formed input to the classifier as the feature vector. Several benchmark simulation examples and real flow process data were used to successfully validate the applicability of the proposed method.