Root cause isolation of propagated oscillations in process plants
Persistent whole-plant disturbances can have an especially large impact on product quality and running costs. There is thus a motivation for the automated detection of a plant-wide disturbance and for the isolation of its sources. Oscillations increase variability and can prevent a plant from operating close to optimal constraints. They can also camouflage other behaviour that may need attention such as upsets due to external disturbances. A large petrochemical plant may have a 1000 or more control loops and indicators, so a key requirement of an industrial control engineer is for an automated means to detect and isolate the root cause of these oscillations so that maintenance effort can be directed efficiently. The propagation model that is proposed is represented by a log-ratio plot, which is shown to be ‘bell’ shaped in most industrial situations. Theoretical and practical issues are addressed to derive guidelines for determining the cut-off frequencies of the ‘bell’ from data sets requiring little knowledge of the plant schematic and controller settings. The alternative method for isolation is based on the bispectrum and makes explicit use of this model representation. A comparison is then made with other techniques. These techniques include nonlinear time series analysis tools like Correlation dimension and maximal Lyapunov Exponent and a new interpretation of the Spectral ICA method, which is proposed to accommodate our revised understanding of harmonic propagation. Both simulated and real plant data are used to test the proposed approaches. Results demonstrate and compare their ability to detect and isolate the root cause of whole plant oscillations. Being based on higher order statistics (HOS), the bispectrum also provides a means to detect nonlinearity when oscillatory measurement records exist in process systems. Its comparison with previous HOS based nonlinearity detection method is made and the bispectrum-based is preferred.