Methods for root cause diagnosis of plant-wide oscillations



Plant-wide oscillations are common in many industrial processes. They may impact the overall process performance and reduce profitability. It is important to detect and diagnose such oscillations. This paper reviews advances in diagnosis of plant-wide oscillations. The main focus of this study is on identifying possible root causes of oscillations using two techniques, one based on data analysis in the temporal and spectral domains and the other based on process connectivity analysis. The process data-based analysis provides an effective way to capture the difference between the root cause variable and the secondary propagated oscillating variables. It is shown that process topology-based methods are capable of finding oscillation propagation pathways and, thus, help in determining the root cause. This paper discusses and compares five such methods—spectral envelope, adjacency matrix, Granger causality, transfer entropy, and Bayesian network inference methods— by application to an industrial benchmark dataset. © 2014 American Institute of Chemical Engineers AIChE J, 60: 2019–2034, 2014