(200a) Lti Model Based Transition Monitoring Applied to Ethylene Plant Dryer Switching

Smooth process transitions during equipment start-ups and shutdowns are very important to maintain process safety and reliability. Equipment changeover transitions are susceptible to faults due to operator variability or equipment failure that, when undetected, can upset the process steady state and decrease the production yield. There are many transition examples in ethylene production where process equipment is taken offline for regeneration or maintenance, and it can be very challenging to isolate transition responses from other miscellaneous disturbances. Operator assisted transitions typically include manual valve operation outside combined with set point changes inside and are accompanied by controller set point and load change responses.

In this work, multivariable transition monitoring is performed by modeling equipment start-up and shutdown transition response steps using linear time invariant (LTI) models. The methodology is tested on a simulated dryer switching procedure in an ethylene plant. Real-time transition monitoring with DCS data could be used to notify operators about transition faults as they occur. We model transition responses with LTI models relating transition variables to valve outputs; however, for manual operations where valve outputs are not available, we use other related readings like pressure and/or flow changes to model the transition responses. Monitoring is performed by calculating the residuals between the LTI model predictions and the actual transition response and using an appropriate fault detection filter.