By Vitalis Okafor and Maxwell Bl Contents Figures & Tables ........................................................................................................................................ 2 Model Construction ...................................................................................................................................... 3 Data Input ..................................................................................................................................................... 6 Exponential distribution ........................................................................................................................ 6 Gamma distribution .............................................................................................................................. 6 Loglogistic distribution .......................................................................................................................... 6 Uniform distribution ............................................................................................................................. 7 Weibull distribution .............................................................................................................................. 7 Model Verification & Validation ................................................................................................................... 8 Results ........................................................................................................................................................... 9 Case 1 ........................................................................................................................................................ 9 Case 2 ...................................................................................................................................................... 10 Conclusions ................................................................................................................................................. 12 Works Cited ................................................................................................................................................. 13 Appendices .................................................................................................................................................. 14 Appendix 1: Relevant Formulas from ExpertFit Manual [4] ................................................................... 14 Appendix 2: Java code used to compute the formulae of the different distributions ........................... 16
Figures & Tables
Figure 1: Conceptual Model of Airport Baggage Handling System ............................................................... 4 Figure 2: Specification Model of Airport Baggage Handling System ............................................................ 5 Figure 3: Processes Average Total Baggage Seized..................................................................................... 11 Figure 4: Processes Scheduled Utilization .................................................................................................. 11
Table 1: Results for Case 1 ............................................................................................................................ 9 Table 2: Results for Case 2 .......................................................................................................................... 10
Model Construction The baggage handling system simulated in this model was constructed on Arena Simulation Software Student Edition mostly using information from the baggage handling systems at Denver International Airport [1] and Rockford Internal Airport [2]. The model begins with passengers arriving at the check-in counter where their baggage is tagged and checked in. In this model, we simulate the baggage check-in process with three check-in counters represented by a Process module that has a Check-In Counter resource with a capacity of three. After the bags are checked in, they are scanned by an automated barcode scanner; which is actually an arrangement of multiple barcode scanners, scanning from all angles of the baggage in order to find the randomly placed tag. This automated barcode scanner can usually scan the barcode tags on about ninety percent (90%) of the bags that go through it [1]. Any baggage that was not scanned by the automated barcode scanner is routed to another station to be manually scanned by personnel. In the simulation model, the manual and automated barcode scanners are represented by Process modules while the ninety percent scan completion rate is simulated by a Decide module with a 2-way by Chance type and a value of 90 for the Percent True. Once the bags are scanned, they are routed by a network of conveyors to their appropriate destination. During this process, the bags are taken through x-ray machines and security devices such as the Explosive Detection System (EDS), where baggage is checked for explosive material. The routing conveyors are represented by a single process with resources EDS and Conveyor Belt to signify the two main parts of the conveyor network. In order to get to their appropriate destinations, the baggage have to be loaded onto the Destination Coded Vehicle (DCV), which takes them to their respective terminals and the off-ramp at the gate. The DCV consists of a plastic tub sitting on a metal cart with wheels that rides on a track using linear induction motors that are mounted to the track. DCVs possess a passive radio-frequency circuit which broadcasts a unique number by which individual cars are identified [1]. In the simulation model, it was assumed that the DCV was a continuation path for the conveyors and as such, could be modeled as a process module using the same Conveyor Belt resource mentioned above. At the off-ramp at the gate, there is usually a sorting station at which baggage handlers sort the bags and load them onto the plane based on whether they belong to transferring (connecting) passengers passengers who will be boarding another flight at the current flights destination or terminating passengers passengers whose journey ends at the current flights destination. This is simulated in the model by a Decide module with a 2-way by Chance type and a value of 80 for the Percent True signifying that eighty percent of the travelers will be terminating passengers; an assumption based on an average of various similar statistics on the subject. When the plane lands, the baggage follows two separate paths depending on whether it is a transferring bag or a terminating bag. If it is a transferring bag, it is taken through another series of routing conveyors where they go through security scans (EDS); and onto another DCV track which takes them to their appropriate destination. For simplicity, this was modeled by adding a Re-routing Conveyor process, which is a replica of the Routing Conveyor process, and a loop that goes back to the previously mentioned DCV process. If it is a terminating bag, it is taken through a conveyor to the baggage carousel where the bags reunite with their owners at Baggage Claim. This was modeled by adding a Conveyor to Carousel process which had a resource of Conveyor Belt and went to the Dispose module Baggage Claim. An algorithm summarizing the simulation process (Conceptual model) is shown below:
The model was run for twenty-four (24) hours and three hundred and sixty-five (365) days to simulate the year-round daily operations of an airport. It was also run for 10 replications to reduce variability in the simulation values. The specification model can be found in the figure below.
Figure 1: Conceptual Model of Airport Baggage Handling System
Figure 2: Specification Model of Airport Baggage Handling System Data Input In order to obtain necessary distributions to represent process delay times in our model, various assumptions had to be made. It was assumed that the data acquired from Khadgi [2] was relevant to our model; and if our model was well constructed, would yield similar (not necessarily the same) results. This data was collected from the Chicago-Rockford International Airport and was used to develop appropriate data distributions. However, these distributions were developed using Input Analyzer and ExpertFit the data analysis tool used in conjunction with FlexSim simulation software and had different formatting and parameter assignments than Arena. Hence, it was necessary to convert them to values that were suitable for Arena based on some calculations and formulas obtained from the ExpertFit Manual [4]. The relevant distributions from ExpertFit include: Passenger Inter-Arrival Times: Gamma (2.907407, 38.700486, 1.303549) Baggage Check-in Times: Expo (44, 86, 1) Baggage automated scan time: Loglogistic (34.698113, 8.120486, 2.348135) Baggage load in time (assumed to be Plane Loading Time): Weibull (3.933333, 3.199017, 0.744261) Baggage load out time (assumed to be Sorting Time): Gamma (0.000000, 0.906343, 13.653779) Manual scan time: Uniform (235, 313) For each distribution type, the following formulas from [4] were then used to generate Arena- acceptable parameters: Exponential distribution ExpertFit: expo (,) Parameters: Location (shift) parameter (-, ), scale parameter > 0 Mean = + Arena: Expo (Mean) = Expo ( + ) Gamma distribution ExpertFit: gamma (, , ) Parameters: Location (shift) parameter (-, ), scale parameter > 0, shape parameter, > 0 Arena: Gamma (BETA, ALPHA) = Gamma (, ) Loglogistic distribution ExpertFit: LL (, , ) Parameters: Location (shift) parameter (-, ), scale parameter > 0, shape parameter, > 0 Mean = +cosecant () for > 1, where = (/) Variance = 2 {2cosecant (2) [cosecant ()] 2 } for > 2, where = (/) Standard deviation = Variance Arena: *LOGN (LogMean, LogStd) *Assuming Lognormal (LOGN) Loglogistic (LL) Uniform distribution ExpertFit: U(a,b) Parameters: a and b are real numbers with a <b; a is a location parameter, b a is a scale parameter Range = (a,b) = (min, max) Arena: UNIF (Min, Max) = UNIF (a, b) Weibull distribution ExpertFit: Weibull (, , ) Parameters: Location (shift) parameter (-, ), scale parameter > 0, shape parameter, > 0 Arena: WEIB (Beta, Alpha) = WEIB (, ) In order to determine the input parameter values for the conveyor and DCV processes, we assumed security screening and bag wait times in the conveyor from Hafizogullari et al [3]. In this case study, wait and process time statistics for baggage screening were developed by TransSolutions LLC for the baggage handling system at Lambert St. Louis International Airport. It was determined that the Bag Wait + Process Time for Screening includes the time the bag waits while the ID is checked; the wait time on the EDS belt; the wait time in front of the primary and secondary Explosive Trace Detection (ETD) agent; the processing time on the EDS and ETD machines, and the secondary ETD processing time. This parameter was found to have a maximum value of 22.7 minutes; an 85 th
percentile value of 3.3 minutes, and a 95 th percentile value of 7.2 minutes.
Model Verification & Validation In order to verify our model, we tried to determine that the model operates as intended by viewing the animation at slow speeds. It was observed that as the time increased in the simulation, the passenger inter-arrival rate caused some queues at the Baggage check-in and Automated barcode scanner processes as expected. Also, the Decide modules were operating correctly based on the intended logic and percentages wanted. Lastly, the feedback loop from the Re-routing conveyor also followed the intended logic, sending transferring bags back to the DCV to go through the system again. In order to validate the model, the simulation was run according to some cases and scenarios from the Rockford Airport Case Study [2]. Due to differences in model construction methods and simulation software, it was not expected that the results will be the same as in Khadgi [2]; however, it was anticipated that both models will possess similar responses and the results will follow comparable trends if the model was correct. For the first case, the simulation was run in three different scenarios to observe the models response to changes in baggage screening methods. This was also done to show the effect of automation to the baggage handling system by comparing the old system of manual baggage scanning/screening; with the current system of 90% automated scan completion and 10% manual scanning; and a 100% automated scanning system. As in Khadgi [2], it was noted that the model showed reduced wait time and increased baggage throughput with the use of more automatic baggage screening. However, it was also noted that there was an error when 100% of the baggages had to go through the manual scanner as the number of entities in queue were too large to run in the Arena Student Edition software. This signifies that the old system of manual scanning was inefficient and would experience bottlenecks with the current resources; and probably required more manpower to meet up with passenger inter-arrival demands. The second case was set up to observe the sensitivity of the model to passenger arrival with varying numbers of available check-in counters (or agents). The first scenario was run as the current system with 3 available check-in counters; while the second and third scenarios were run with 4 and 5 available check-in counters, respectively. As in Khadgi [2], the model also followed the trend of reduced wait times with additional check-in counters.
Results Case 1 As was noted in the Verification and Validation section above, the model experienced greater baggage throughput and improved wait times with the introduction of more automation. It is also noted that there was an improvement in processing as there were less bags in queue with the increase of automation. The table below lists the results for the scenarios in Case 1: Current System Automatic scanner only Manual scanner only Number out 628625 bags 628496 bags N/A Number of bags auto scanned 565738 bags 628496 bags N/A Number of bags manual scanned 62887 bags 0 N/A Average Wait Time 485.59 seconds 470.47 seconds N/A Average Total Time 748.05 seconds 705.46 seconds N/A Max # in baggage queue 103 bags (Average 15 bags) 82 bags (Average 14 bags) N/A Check-in counter utilization 0.865 0.8644 N/A Table 1: Results for Case 1 The manual scanner only scenario was unable to be simulated due to entity constraints in the student edition of Arena. However, it can still be deduced that the wait times would have been significantly higher and the throughput would have been less than the automated scenarios. This is because a bottleneck would be experienced at the manual barcode scanner which will create a backlog in the system and would temporarily bring the system to a halt until the bottleneck is cleared. Such bottlenecks are the cause for flight and baggage delays and the undesirable baggage mishandlings that were a common occurrence in the past.
Case 2 In the second case, the model was analyzed to observe how additional check-in counters would improve the performance outputs. The system showed significant improvements in baggage throughput and reduced total and wait times with increased available check-in counters. Also, the maximum and average numbers of baggage in queue reduced with the addition of more check-in counters as a result of the improved processing capacity. Lastly, the check-in counter utilization reduced to acceptable levels with more check-in counters available. The results for the scenarios of case 2 are shown below: Current System (3 check-in counters) 4 check-in counters 5 check-in counters Number out 628625 bags 628764 bags 628599 bags Number of bags auto scanned 565738 bags 565956 bags 565716 bags Number of bags manual scanned 62887 bags 62809 bags 62885 bags Average Wait Time 485.59 seconds 285.87 seconds 259.11 seconds Average Total Time 748.05 seconds 548.18 seconds 521.5 seconds Max # in baggage queue 103 bags (Average 15 bags) 68 bags (Average 11 bags) 68 bags (Average 10 bags) Check-in counter utilization 0.865 0.6484 0.5187 Table 2: Results for Case 2 Based on the results from the table above, it can be seen that the 4 check-in counter scenario is the most favorable. In this scenario, the baggage throughput is maximized and the maximum number in baggage queue is at its minimum (68 bags). Although the wait and total times are lesser in the 5 check-in counter scenario; they are only minimally so compared to the huge reduction between the current system and the 4 check-in counter scenario (259.11 seconds < 285.87 seconds <<< 485.59seconds). Also, the check-in counter utilization, at approximately 65%, is in a moderate position where the counters are neither being under- or over-utilized.
Figure 3: Processes Average Total Baggage Seized Using the 4 check-in counter scenario as our optimal solution and case study, we can compare the operation of the various processes through their scheduled utilization and total baggage seized.
Based on the average total baggage seized, the Baggage Handler resource and the Conveyor Belt resource get the most baggage throughput; while the manual scanner gets the least. Based on the scheduled utilization, the Automatic scanner gets the most utilization while the EDS and Vehicle (DCV) get the least. This comparison will help determine areas for improvement and those that will require additional resources to meet the passenger arrival loads.
Figure 4: Processes Scheduled Utilization Conclusions Based on the discrete-event simulation of the baggage handling system and the ensuing analysis of results, it was proved that an increase in automation in baggage screening/scanning will lead to better performance outputs while utilizing fewer resources; as seen in the first case. It was also observed that the manual scanner only system was incapable of meeting the passenger arrival loads and will create a bottleneck in the system; which is an undesirable outcome. It would involve a significant amount of additional resources or manpower to meet the current loading requirements with the old system. However, it was ascertained that the current system of 90% automatic scanning and 10% manual scanning is indeed optimal for the current passenger arrival loads. It was also proved that an increase in the number of available check-in counters will lead to better performance outputs: reduced baggage in queue and lower wait times; as was seen in case 2. There was a significant improvement in wait times and total times with the addition of more check-in counters; while there was a corresponding reduction in number of bags in queue. Lastly, it was determined that the 4 check-in counter system was the most favorable as it had the highest baggage throughput and maintained comparable performance outputs to the 5 check-in counter system.
Works Cited [1]Nice, Karim. "How Baggage Handling Works" 13 June 2001. HowStuffWorks.com. <http://science.howstuffworks.com/transport/flight/modern/baggage-handling.htm> 25 October 2013. [2] Khadgi, P., "Simulation Analysis of Passenger Check in and Baggage Screening Area at Chicago- Rockford International Airport," NIU Engineering Review, . [3] Hafizogullari, S., Bender, G., and Tunasar, C., 2003, "Simulation's Role in Baggage Screening at the Airports: A Case Study," Proceedings of the 2003 Winter Simulation Conference, pp. 1833. [4] Law, A., 2006, "ExpertFit Version 7 User's Guide," Averill M. Law & Associates, Inc., Tucson, AZ, pp. 131.
Appendices Appendix 1: Relevant Formulas from ExpertFit Manual [4]
Appendix 2: Java code used to compute the formulae of the different distributions