Turnaround prediction concept: proofing and control options by microscopic process modelling

Transkript

1 Turnaround prediction concept: proofing and control options by microscopic process modelling GMAN proof of concept & possibilities to use microscopic process scenarios as control options Bernd Oreschko, Thomas Kunze, Tobias Gerbothe and Hartmut Fricke Chair of Air Transport Technology and Logistic, Technische Universität Dresden, Germany ICRAT 2014

2 Chair of Air Transport Technology and Logistics Research Actitvities Trajectory Management Uncertainty in 4D-Trajectories Safety & Security Assessment Simulation Based Risk-Analysis Terminal Operations Expert knowledge exchange Airport & Terminal Operations Turnaround prediction and steering Pushback and Deicing Management Other A320 Cockpit Simulator Safety Analysis in TMA Dipl.-Ing. Bernd Oreschko Turnaround Prediction and Controlling Slide 2

3 Motivation & Background Research Review Turnaround Prediction Modell GMAN Process Controlling by Microscopic Modelling Conclusion & Outlook Turnaround Prediction and Controlling Slide 3

4 ONBLOCK OFFBLOCK Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Background - ACDM EUROCONTROL Airport Collaborative Decision Making Improves situational awareness and rises efficiency by Better information connection and sharing for all airport partners Therefore better capacity use when information is used correctly Establishing Milestone concept Inbound Turnaround Outbound CDM Milestones 20 to 40 TOBT Update Impacts for the Turnaround: Prediction of Target Off Block Time (TOBT) & Turnaround Time (TTT) HOW? => knowing process characteristics for all process and interconnections and control options Turnaround Prediction and Controlling Slide 4

5 ONBLOCK OFFBLOCK Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Motivation Turnaround Uncertainty Uncertainty of start and duration cause of several factors, e.g. delays, extended sub-process duration, airport type or staff skills? Turnaround? Deterministic TA-Planning does not work! Best Guessing by Ramp and Ops Agents does not fit in 4D/ACDM Environment t Turnaround Prediction and Controlling Slide 5

6 ONBLOCK OFFBLOCK Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Motivation Modelling for DST AMAN SMAN AMAN SMAN EXIT DMAN Turnaround TTT SMAN EXIT TTT SMAN DMAN EXOT EXOT xldt xibt The Turnaround within adjacent ATM-tools inline with EUROCONTROL perspective: AMAN Arrival Management Tool DMAN Departure Management Tool SMAN Surface Management Tool Turnaround Modell Output useful for SMAN and DMAN xldt xibt TOBT xtot TOBT time xtot time t Überschrift 1 Verdana 10 Turnaround Prediction and Controlling Slide 6

7 ONBLOCK OFFBLOCK Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Motivation new DST for Turnaround - GMAN AMAN AMAN SMAN AMAN SMAN SMAN DMAN DMAN DMAN EXIT TTT SMAN EXITTurnaround TTT TTT SMAN SMAN EXOT EXOT EXOT xldt xibt xldt xldt xibt xibt TOBT TOBT xtot time TOBTxTOT xtot time time t Turnaround Prediction and Controlling Slide 7

8 Motivation new DST for Turnaround - GMAN AMAN SMAN DMAN EXIT TTT SMAN EXOT xldt xibt TOBT xtot The GMAN output may be used in ramp operations control or schedule planning the following ways: Perform Critical path analysis of TA process Analyze expected buffers between processes constituting a TA event Identify non-achievable target times at earliest times Identify excessive process durations. time t Turnaround Prediction and Controlling Slide 8

9 Motivation & Background Research Review Turnaround Prediction Modell GMAN Process Controlling by Microscopic Modelling Conclusion & Outlook Turnaround Prediction and Controlling Slide 9

10 occurance occurance Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Turnaround Research findings by TU-Dresden Field measurements and data analysis on several airports (MUC, FRA, STR, HAM, DRS, LEJ) show a discrepancy between scheduled and actual times: Actual Turnarounds don t fit fixed plans Process durations and buffers influenced by Delays Airport (category) => Staff Skills Unloading - no delay 0,14 0,12 0,1 A319 Turnaround Plan, source: Airbus SAS 0,08 0,06 0,04 0,02 0 0,14 0,12 Supply Basis Hub 0,1 Unloading - delay Reality Delay influences on process durations & buffers unloading rate [seconds/pax] Example Process Variations due to Airport Category & Delay - Unloading 0,08 0,06 0,04 0, unloading rate [seconds/pax] Supply Basis Hub Dipl.-Ing. Bernd Oreschko Turnaround Prediction and Controlling Slide 10

11 TUD Turnaround Researches The sub-processes comprising a TA should be modeled stochastically as they have uncertainty associated with their processes duration. The TA process is dependent on various parameters like airport category and operational factors (e.g. passenger number, airline, aircraft type), and these information can be obtained from different sources. Incoming delay has an important influence on the individual subprocess duration and process interaction times (buffers). See ICRAT Contributions of last years and others: Turnaround Prediction and Controlling Slide 11

12 Motivation & Background Research Review Turnaround Prediction Modell GMAN Process Controlling by Microscopic Modelling Conclusion & Outlook Turnaround Prediction and Controlling Slide 12

13 TTT Prediction and Controlling Two Step Approach 1. Prediction of TTT and Process Duration with stochastics => comparison with other target times (e.g. TSAT) 2. Control Options by microscopic task simulation => possible handling options GMAN TTT Prediction Target Time comparision e.g.: cttt = TSAT? Mircoscopic Process Simulation Control Option 1 Control Option.. Control Option n Turnaround Prediction and Controlling Slide 13

14 GMAN Process Description Prediction is based on single process description and their interaction results Duration Process Described Processes: deboarding, catering, fuelling, cleaning, boarding, unloading, loading (other possible) Processes description: Each of these process duration and Start time is stochastically described Description source: empirical data from aircraft operators, airports and ground handling companies are used Start - influence of the following parameters: Aircraft type Airline Airport inbound and outbound Airport where the TA is processed Flight distance to destination Flight type, i.e. low cost or legacy Incoming delay (on gate) Number of passengers inbound and outbound Type of aircraft stand Turnaround Prediction and Controlling Slide 14

15 GMAN Critical Path Calculation Duration Repeated n-times Deboarding Start Duration Fuelling Start Start Start Duration Catering Duration Cleaning Critical Path Duration Boarding Duration Unloading Duration Loading Start IBT Start Start TTT TOBT GMAN critical path calculation for one run out of n - with stochastic process start times and duration description Turnaround Prediction and Controlling Slide 15

16 GMAN Stochastic Process Description Finally, probability distribution functions can be fitted to the collected/empiric data. The Turnaround Modell is designed to allow using any distribution functions, while fit best. Deterministic (fallback level) and Weibull distribution Further more arrays of single notfitable times can be used Output for GMAN: Qantiles Start Duration Process / TTT Possible Output for DST-Tool GMAN: Qantiles Reliability Accurateness Turnaround Prediction and Controlling Slide 16

17 GMAN-Look-Ahead Time A higher prediction accuracy level requires more reliable information for better stochastic process description Trigger Information Sources Seasonal Flight Plan Operational Flight Plan Fuel Order as the LAT decreases, more accurate trigger information is expected to become available, and therefore a more specific stochastic process description out of the empirical database can be gathered and fitted Ground Operation Manuals Aircraft Characteristics Data Substitution Service Level Agreements Turnaround Prediction and Controlling Slide 17

18 GMAN Prototyp Turnaround Prediction and Controlling Slide 18

19 Proof of Concept (POC) at LEJ Airport LEJ is a medium sized airport with mainly domestic and European flights Modification of the GMAN model due to the lack of necessary date (no Timestamps at the a/c) start and end times of deboarding, fuelling, unloading, loading and boarding were adjusted, s/aibt and s/aobt also available =>adoptions to GMAN model No catering and cleaning process (minimum amount of occurrences of these processes on the critical path in LEJ) Aim of POC: Does the GMAN gives a usable TTT prediction Overview LEJ Airport PAX Facilities Source: airportzentrale.de Turnaround Prediction and Controlling Slide 19

20 POC at LEJ Airport Empiric Data Source Preparation of Empiric Data (> data) from IFL Database 1. all TA with a scheduled TTT (SOBT-SIBT) above 2 hours were skipped => # Data class preparation by trigger information: airline => main, charter and low-cost classes aircraft type by the maximum seats available => eg. ac100- (up to 100 seats), ac156 (101 up to 156 seats) passenger numbers inbound / outbound by cluster of 2: eg. pax25 (0-25 passengers), pax50 (26-50 passengers) 3. creation of classes for start times and durations, regarding the trigger information boarding, deboarding, loading and unloading: 1. aircraft type 2. corresponding passenger number class Fuelling 1. aircraft type 2. the destination airport Turnaround Prediction and Controlling Slide 20

21 POC at LEJ Airport GMAN connected to the local airport information network Only final trigger information => no intermediate steps => no LAT analysis Data analysis for 600 turnarounds in 09/2013 Output of the GMAN: stochastic values of TTT for a single TTT Mean, μ, δ Manual match of predicted TTT and ATTT by operational staff => Questions: Does the prediction cover with the reality? Indications to what should be the target values for the GMAN output? Turnaround Prediction and Controlling Slide 21

22 POC at LEJ Airport -Output Deviation of ATTT to GMAN TTT prediction, all flights No clustering by trigger information No good prediction => Clustering necessairy Turnaround Prediction and Controlling Slide 22

23 POC at LEJ Airport Output Deviation of ATTT to GMAN TTT prediction, A319 to CGN and STR Lowcost Turnaround (25 min) => Simple/stable TA, acceptable prediction Turnaround Prediction and Controlling Slide 23

24 Motivation & Background Research Review Turnaround Prediction Modell GMAN Process Controlling by Microscopic Modelling Conclusion & Outlook Turnaround Prediction and Controlling Slide 24

25 xibt Start: Groundpower Start: Unloading Start: Deboarding Deboarding Unloading End: Deboarding End: Unloading Start: Catering Start: Cleaning Start: Fuelling Start: Water Start: Waste Water Start: Air Conditioning Catering Cleaning Fuelling Water Waste Water Air Conditioning End: Catering End: Cleaning End: Fuelling End: Water End: Waste Water End: Air Conditioning Start: Security Security Turnaround Time End: Security Ground Power Start: Boarding Start: Loading Boarding End: Boarding Baggage Loading Start: Headcounting Headcounting End: Headcounting End: Loadding Start: Air Starter Air Starter End: Air Starter xobt Start: Pushback Start: Deicing End: Pushback Pushback End: Deicing Deicing Fakultät Verkehrswissenschaften Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs Turnaround Process Charts Developing process chart to cover all elementary steps of turnaround processes Dividing process into tasks, Turnaround Time Stamps Passenger Services Baggage Mail Cargo Services? Aircraft Services Overall TA Relevant steps of aircraft catering Dipl.-Ing. Bernd Oreschko Turnaround Prediction and Controlling Slide 25

26 Example: Microscopic Cleaning Model Identification of significant cleaning steps: remove, clean, restock, arrange for seats, lavatories, galleys, vacuum Define sequence of steps using different scenarios (sequence, staff usage), as different control options Dipl.-Ing. Bernd Oreschko Turnaround Prediction and Controlling Slide 26

27 Cleaning Progress Progress of each cleaning step using expected value of duration Remarkable inter-process dependencies, not easy to cover with analytical description Dipl.-Ing. Bernd Oreschko Turnaround Prediction and Controlling Slide 27

28 TTT Simulation with Microscopic Processes Simulation shows anticipated behavior Next step is to prove the usability in live environment Turnaround Prediction and Controlling Slide 28

29 Motivation & Background Research Review Turnaround Prediction Modell GMAN Process Controlling by Microscopic Modelling Conclusion & Outlook Turnaround Prediction and Controlling Slide 29

30 Conclusion and Outlook Proof of concept of TTT Prediction is confirmed The different levels do not generally imply that the TA process time prediction will get smaller by default (or even show a smaller variation) But the reliability of the results increases GMAN principle proofed: Identify non-achievable target times (at earliest times) Identify excessive process durations Output with quality information Next Step: Test and Validation of control options (in LEJ) Turnaround Prediction and Controlling Slide 30

31 Bernd Oreschko Chair of Air Transport Technology and Logistic Technische Universität Dresden, Germany Turnaround Prediction and Controlling Slide 31