5.2 Application to Binary Distillation Column
5.2.4 Composition Sensor Faults Accommodation Using FTIC
The proposed sensor fault tolerant controller is implemented on the distillation column the moment a sensor fault is identified. This is made possible through the use of the relevant redundant controlled variable signal, which is the estimate provided by the DPCR based soft sensor and is used in place of the faulty sensor output in the feedback control loop as presented in Figure 3.3. The controlled variable feedback signals (yp0) used during normal operation is obtained using equation 3.18 as:
yp0 = 1 0 0 1 YD XB + 0 0 0 0 ˆ YD DP CR ˆ XB DP CR (5.3)
When a sensor fault is identified, for instance, a top composition sensor fault (faulty YD
value), the diagonal element corresponding to the faulty top composition sensor output changes to zero to isolate it while the corresponding diagonal element in the redundant backup feedback signal (DPCR estimate) is activated accordingly. The resulting con- trolled variable feedback signal during the sensor fault accommodation is then given as:
yp0 = 0 0 0 1 YD XB + 1 0 0 0 ˆ YD DP CR ˆ XB DP CR (5.4)
The sensor FTIC strategy simply replaces the faulty sensor output with the DPCR inferred estimates to maintain the integrity of the control system and that of the plant, which in this case is a binary distillation column. The same procedure is followed if the bottom
ACCOMMODATION ON DISTILLATION COLUMNS
composition sensor fault is identified. Figures 5.8 and 5.9 present the accommodation of sensor fault cases F 1 − F 2 and F 3 − F 4 respectively using FTIC.
Figure 5.8: F1 and F2 sensor faults accommodation
5.2.5
Results and Discussions
All the four sensor faults investigated in the binary distillation column are detected and properly identified. From the analysis of the T2 and SPE monitoring plots presented in Figure 5.5, sensor faults F 1 − F 4 were all detected. F 1, top composition sensor fault, as presented in Table 5.1 was detected at sample 760 on both T2 and SPE monitoring plots, 5 minutes after it was introduced at sample 750. Fault F 2, bottom composition sensor fault was detected and identified at samples 796 and 781, some 23 minutes and 15 minutes 30 seconds after introduction on T2 and SPE monitoring plots respectively. Fault
F 3 has similar characteristics with F 1, and it was identified 5 minutes after introduction, at sample 760 on both T2 and SPE monitoring plots. It took 22 minutes and 7 minutes
at samples 794 and 764 on T2 and SPE monitoring plots respectively, for fault F 4 to be detected, as presented in Figure 5.5.
Upon detection of a fault, further diagnostics are undertaken to identify the actual fault that has just been flagged through the use of T2 and SPE contribution plots, aided by a good understanding of the process under investigation. Observations from the T2
and SPE contribution plots shown in Figure 5.6 show that top composition (variables 1 and 15) is the major contributor to the sensor fault F 1. The fault was easily detected due to a sudden change in the top composition sensor output, causing the system to drift out of acceptable operating conditions. A similar situation was observed in sensor fault F 3. The top composition sensor is highly sensitive, even a 5% drift in its output will result in a declaration of a sensor fault. The contribution of reflux flow rate (variables 3 and 17) to faults F 1 and F 3 is also significant, but much less than those of variables 1 and 15. Bottom composition sensor faults F 2 and F 4, as presented in Figures 5.6 and 5.7 show variables 3, 4, 5, 6, and 7 (reflux flow rate, steam flow rate, temperature of stages 10, 9 and 8) as the major contributors to the fault declared from the T2 contribution plot. The
SPE contribution plot for F 2 is rather more conclusive, indicating variables 2 (bottom composition) and 5 (bottom stage temperature) as perhaps the only contributing variables to the faulty situation. Good knowledge of the process together with contribution plots aided the fault identification.
The proposed sensor FTIC is implemented on the binary distillation column for the fault cases F 1 − F 4 upon identification. Figures 5.8 and 5.9 present the responses of the top and bottom compositions under the sensor fault accommodating strategy. The inferential control strategy used the soft sensor estimates in place of the faulty sensor
ACCOMMODATION ON DISTILLATION COLUMNS
measurements for feedback control, thereby accommodating the sensor fault. The effects of feed flow and feed composition disturbances after the faults were well compensated for by the fault tolerating control approach, as can be observed from Figures 5.8 and 5.9. The sensor FTIC strategy works quite well in preserving the integrity of the control system.
5.3
Application to Crude Distillation Unit
Real-time sensor faults detection, identification and implementation of the proposed sen- sor fault tolerant controller, also referred to as fault tolerant inferential controller (FTIC) on a dynamic crude distillation unit to accommodate sensor faults is presented in this section. As demonstrated here, faulty sensors need to be quickly identified and isolated in order to preserve the integrity of both the control system and the process. A detailed de- scription of the dynamic crude distillation unit on which the sensor FTIC is implemented was given in Section 4.4.1, and Figure 5.10 presents the unit with the four sensor faults investigated in this section. Only the important parts of the CDU system that relate directly to the implementation of the sensor FTIC is described here to avoid repetition. Figure 5.11 gives a summary of the CDU schematic showing all the input and output vari- ables into the system, including the disturbance variables (DV) and the products quality variables (PQV).
The crude distillation unit has a total of 71 variables which include flow rates and temperatures of all the streams and temperature measurements of all the column stages. There are three disturbances in the system, namely the crude composition, temperature and flow rate as shown in Figure 5.11. Crude oil is fed into the atmospheric distillation unit developed in HYSYS at a temperature of around 15oC. The crude is then heated to
185oC through series of heat exchangers by exchange with hot intermediate streams from the crude and vacuum columns, before entering the furnace where its temperature is raised to 360oC, the temperature at which it enters the atmospheric column flash zone. The column has naphtha, kerosene, diesel, atmospheric gas oil (AGO) and the CDU residue as its products. ASTM D1160 cut-points at 0% and 100% for kerosene, ASTM D1160 cut-points at 90% and 95% for diesel, ASTM D93 flash points for kerosene and AGO, and AGO viscosity at 210F are the product quality variables used to determine the quality of the products. The nominal values for the product quality variables are presented in Table 5.3. Details of the sensor faults investigated in the system as shown in Figure 5.10 are given in Section 5.3.2.