LAUSR

Abstract Oscillation in control loops is a frequent problem faced in process industries. It deviates the process variables from their desired condition, affecting negatively plant productivity. To guarantee profitability, oscillation must be detected, diagnosed, and, finally, eliminated. Dozens of automatic detection and diagnosis techniques have been proposed over the last years. However, the application to real industrial data reveals low efficiencies, which indicates that the techniques require improvement. This work presents a new method for oscillation detection and diagnosis. The technique classifies the loops based on the shape of the PV(OP) diagram by a pattern recognition approach, where the model is trained with simulated examples that cover a large range of processes and different conditions found in industrial data, such as noise and disturbance presence. The performance of the proposed technique is compared to well-established oscillation detection and diagnosis methods, returning better results.