Abnormal situation detection and contro

Abnormal situations are defined as those phenomenon that occur which upset a unit operation and typically require human intervention to address. The area of process control encounters many different forms of abnormal situations that affect the ability to automate chemicals processes in order to create ease of operability for complex unit processes and allow for optimization to maximize assets. This work deals with approaches to mitigate column pump failures, aid in the detection of poor control loop performance, and reduction of high frequency noise in control loops. Each of these three areas has a chapter of its own.;Pump failures on distillation columns are discussed in Chapter 1. These failures are one of the common problems in plant operations. A dynamic simulation can be created to run various control schemes to evaluate these scenarios without using a real operating column. The simulation results can be used to develop mitigating control schemes that automatically assist in securing a column during an upset. For feed and bottom flow pump failures, simulation runs show that a column can be brought to a total reflux condition. However, simulations for reflux pump failure show that the desired composition specification cannot be maintained in a tower once reflux is lost, but the column material balance can be preserved to shorten the time to return to normal operation.;Chapter 2 is the interacting controller section. Many times one poorly tuned loop is the reason for another loop's poor performance. In the case of interacting controllers, the changes in one loop's output have an effect on the performance of a second control loop. Problems such as poor tuning with one loop can cause a variation in the process value (PV) of a second loop in which the second loop moves its own output in an attempt to correct the deviation. In a plant with many loops, one cannot group all controllers with cycling process values as being caused by the same poorly tuned loop. The difficulty lies in determining which ones are related and to focus attention on those for troubleshooting. Analysis of a group of controllers using Fast Fourier Transforms can determine the ones that have a common cause.;Digital filters are discussed in Chapter 3. These filters are widely used for pattern recognition, decision-making, model identification, regulatory control, and inferential/advanced control when there are noisy signals. In regulatory or inferential control, noisy measurements must be smoothed by an exponential moving average (EMA) filter before being sent to control loops or inferential models. In the stock market, closing prices are often smoothed by an EMA or simple moving average (MA) filter to reveal useful trends. Digital filters are low-pass filters, which further attenuate high frequency noises at the expense of a lag. The lag delays display, hampers control, and impedes a timely decision. A centered exponential moving average (CEMA) filter is the average of the traditional or past EMA (PEMA), which smoothes the historical data and a future EMA (FEMA). In online applications, CEMA smoothes past EMA values and future model-based future values with no lag.