PSI - Issue 19

Aaron Stenta et al. / Procedia Structural Integrity 19 (2019) 27–40 Stenta and Panzarella / Structural Integrity Procedia 00 (2019) 000–000

34

8

Using these base costs and benefits, this decision network predicts that the optimal Drum Cycle time is 12 hours and that the coke drum should be operated for a Total Lifetime of about 18 months before it is replaced. A cycle time shorter than this is predicted to lead to higher repair costs that outweigh the expected increased revenue. A cycle time larger than this optimal value will lead to less revenue that is not made up for by a significant enough reduction in the repair costs. Thus, this optimal cycle time represents the perfect balance between expected costs and benefits in order to maximize the expected average annual return (the total return divided by the total lifetime of the coke drum). During this amount of time, the total Annualized Utility is expected to be about $379,400,000 USD. This is an extreme case that was chosen to illustrate the typical trade-off decisions that need to be made. By varying the revenue earned per cycle or the cost of each repair, different optimal decisions would be recommended. For example, if the revenue per cycle is increased to $800,000 USD, keeping all other parameters the same, a shorter Drum Cycle time of 10 hours is found to be optimal along with a shortened run time of only 6 months. In other words, if there is more money to be made, then it is acceptable to run the coke drum more aggressively, incurring more damage and shortening the life, because the increased revenue makes up for the increased damage rate. However, if the revenue per cycle is decreased to $200,000 USD, a longer cycle time of 14 hours is recommended over 18 total months of operation.

Fig. 5. Low Cycle Fatigue Bayesian Decision Network for a Coke Drum using the Structural Stress Level 2 Method C in Part 14 of API 579-1/ASME FFS-1.

As was noted above for the network illustrated in Figure 4, i.e. the fatigue model chosen assumed a simple form based on that presented in Panzarella (2016). However, the network is in no means limited to such a simplified model. This choice was made for simplicity in demonstrating the concepts. The Bayesian Decision Network illustrated in Figure 5 is an example of a more thorough network that could be included in the coke drum network illustrated in Figure 4. Here, the fatigue model chosen is Level 2 Method C (i.e. structural stress method) in Part 14 of API 579-1/ASME FFS-1 (2007), Osage (2018), and Stenta (2016). The inputs to the fatigue model are the coke drum operating, mechanical, and design parameters needed to obtain an accurate estimate of the Membrane + Bending Stress Range per Drum Cycle. These include Shell Course (used to get the Wall Thickness and supplemental Weight Load, where the wall thickness and the weight load decrease with height), Cycle Duration (used to define the intensity of the thermal stresses for each Drum Cycle ), Operating Condition (used to define the probability of ideal or non-ideal conditions on the coke drum walls, that is the potential for local hotspots), and Weld Type (used to identify if the cracks are more likely at the longitudinal or circumferential welds of the coke drum shell courses), see Gassama (2016). The outcome of this network is a distribution of the Mean Cycles to Crack Initiation that feeds directly into the coke drum network illustrated in Figure 4.