MICANTS

1. MICANTS Jon Doyle Robert Laddaga Vera Ketelboeter (MIT) Gabor Karsai Benoit Dawant Chris vanBuskirk Gabor Szokoli Jonathan Sprinkle Karlkim Suwanmongkol (Vanderbilt/ISIS) Russ Currer (Idea Services) Lt Martin (USMC MAG-13 VMA-513)

2. MICANTS Research Goals • How to use • Model-Integrated Computing, and • Agent/Negotiation technology to solve complex resource management problems in (Autonomic) Logistics • To demonstrate the feasibility of the technology through real-life example(s) • Roles • Vanderbilt/ISIS: MIC, implementation, and demonstration • MIT: Concepts, algorithms • Boeing: Modeling, domain knowledge • Idea Services: Domain expertise and scenarios, customer interface • http://www.isis.vanderbilt.edu/Projects/micants/micants.htm • Demo: http://www.isis.vanderbilt.edu/Projects/micants/maplant/index.html

3. MMCO MMCO (sister squadron) options options options approve approve approve Application SummaryVision: Agent-supported Maintenance Process MAPLANT MAintenance PLanning AgeNTs Goal:Assistance through offering negotiated options maintains Commander’s Intent Maintenance Schedule negotiate report W/C OIC discrepancy report Assign mechanic Autonomic response negotiate Agents: • “Helpers” for the users • Implement CO’s intent, business rules, and user guidance • Negotiate solutions autonomically • Offer options for approval negotiate Flight Schedule Current focus: Negotiation between Flight and maintenance schedule Shop Maintenance Schedule CAUTION: Simplified picture

4. MMCO Resource Allocation ArchitectureScheduling and negotiation as CSP Explicit management of constraints during negotiation/scheduling Negotiating agent Other agent Coordination Engine Messaging “High-performance” encoding techniques Data structures representing domain constraints Domain-specific API to the scheduler Schedule Domain-independent SAT techniques Constraint SAT mapper (encoding) Standard SAT Interface (CNF, etc.) • Complexity management: • Encoding strategy • SAT Standard SAT Problem Solver (Tableau,WSAT,ISAMP)

5. ApproachEncoding a scheduling problem as binary SAT • Task constraints • From: Maintenance Plan and Manual • Precedence, Starts after, Ends before, Coherence • Resource constraints: • Capacity (mechanics and tools) • Flight requirements • Guidance: • Preferences for scheduling certain tasks for certain times SCALING: Polynomial in #Tasks, #Resources, #Slots

6. Resource allocation & scheduling problem • Negotiated, joint scheduling of flight operations and maintenance tasks with resource allocation • Long-term version: - IAM-1 • Time span: 5 weeks • A/C allocation w.r.t. usage guidance • Calendar- and usage-based inspections • Resource constraints • Short-term version: - IAM-2 • Time span: next day • Based on current status (snapshot) and tomorrow’s flight schedule

7. IAM-1 Problem: A/C assignment and long-term scheduling • Interactive/Automatic A/C assignment • Flight hour projections • Usage-based & phase calculations • Risk analysis • Monthly maintenance planning

8. MAPLANT/IAM-1 Data Warehouse 1.Inputs Guidance Guidance Knobs A/C Status UpcomingInspections 2. Assignments & Projections 3. Risk analysis CFSA Project Flt Hours Projected Maint CFSA View PFSA Flt Hours Analysis Maint Manuals Overlap Margins 5.Results 4. Maintenance Scheduler Maintenance Schedule Schedule View Resource Margins Scheduler Aircraft Availability MntPlan View SNAP Tools/SE Roster

9. MAPLANT/IAM-1Operational scenario 1. Maintenance Control Tunes Parameters • Aircraft Status Tweaks • Intentionally remove jets from the pool • Mark downed jets with expected up times • Partial Flight Schedule Assignment (PFSA) • Define/Modify Guidance 2. Workload Projection • Complete Flight Schedule Assignment (CFSA) • Day-By-Day Projection of Accumulated Flight Hours per A/C • Project Scheduled Maintenance Workload(dues windows)

10. MAPLANT/IAM-1Operational scenario (cont.) 3. Risk Analysis & Approval of Flight Schedule • CFSA Analysis (possibly override and re-iterate) • Resource Margins • Overlap Margins • Phase-Phase • 56-Phase (w/i and across a/c) • Engine-Phase • Engine-56 • Aircraft Utilization Rates 4. Schedule Computation • Produces • Maintenance Schedule • Aircraft Availability Projection • Considering • PFSA Constraints • Resource Availability • Resource Margins • Phase-Phase Overlaps • Engine-Phase Overlaps

11. Gives AC types (day, night, radar) Guidance Fleet Info Aircraft Assignment 1 Each AC has: 1. Min, Max, Pref for month 2. Can/Cannot fly for each day of mo. 3. Optional deadline with min, max, and pref (takes precedence over month until it is satisfied) Assign the AC’s to flights, by assigning to the most desperate AC’s first (based up Guidance info), using the FS for supporting information, and the Guidance for driving information. G AC Run until all AC’s have reached their minimum, or until no further sorties can be assigned. FS 2 Assign the AC’s to flights, but this time utilize the preference of the airplane, instead of a hard (min) constraint. Similarly, use the Guidance as driving info, and FS to collaborate. Run until all AC’s have reached their preference, or until no further Sorties can be assigned. FS with detailed sortie info 3 Now, use the FS as the driver, to make sure that all sorties are assigned, but utilize Guidance to make sure that max is not exceeded. Run until all sorties have been assigned, or until no further Sorties can be assigned. Flight Schedule

12. MAPLANT/IAM-1Screens – PRELIMINARY A/C Status Flight Schedule/Assignments Projected Flight Hours Risks/Overlaps 56 Day Phase Engine

13. IAM-2 Problem: (In the works)Shift change and short-term scheduling • Squadron status on demand • Consider flight schedule mission requirements for next shift • Finalize A/C to mission assignment • Assign work to Work Centers • Check impact of decisions

14. MAPLANT/IAM-2Operational scenario 1. Maintenance Status Check • A/C status • Open MAF-s (down and up gripes) • Upcoming Daily Special and Usage-based inspections, phases • Events • Start time, duration, A/C requirements • Operational deviations, pits & turns 2. Finalize A/C to mission assignment • Checks legality constraints • Evaluates assignment with respect to guidance

15. MAPLANT/IAM-2Operational scenario (cont.) 3. Work assignment to work centers • View open MAF-s and Other Maintenance Tasks (OMT) • Assign priorities to MAF-s and OMT-s • Assign MAF-s/OMT-s to Work Centers 4. Check impact • MAPLANT generates maintenance schedule for the next shift • Risk factors calculated from schedule and shown • MMCO checks and approves schedule • If needed, MMCO changes priorities, and repeats

16. 1. Scheduling/planning of short-term, tactical, corrective maintenance actions in light of the flight schedule SNAP supplies daily flight schedule to MAPLANT which in turn generates a daily maintenance plan 2. Scheduling/planning of long-term, strategic, scheduled maintenance actions w.r.t. long-term flight schedule SNAP supplies n-week flight schedule to MAPLANT which uses that to generate a long term maintenance plan 3. Scheduling flight operations based on plane availability MAPLANT provides a "best effort" estimates for the number of available aircrafts over time. SNAP creates a schedule based on these estimated generation rates. 4. Scheduling flight operations based on plane availability and capabilities MAPLANT provides a "best effort“ estimates for the number aircrafts, their capabilities, other attributes, and negotiable and non-negotiable constraints associated with them. SNAP creates a schedule based on this. 5. Negotiation between MAPLANT and SNAP The two systems interact using a well-defined messaging protocol to facilitate negotiation between the flight schedule and maintenance schedule. The objective is to explore trade-offs between the two aspects to achieve global optimization w.r.t. some metric (e.g., generation rate, CRP, etc.) CAMERA / MICANTS Integration Plan

17. Demo Scenario (Nov ’01) Maintenance OPS Guidance MAPLANT SNAP First Cut Plan Demo Demo Negotiation Approx. Maintenance Plan (A/C availability) Refined Ops Plan Refined Maintenance Plan

18. Progress to Date: IAM-1 • Flight schedule driven maintenance scheduling • Manual/automatic A/C to mission assignment under guidance goals • Sophisticated guidance input • Usage usage projection algorithms • Scheduling of both calendar- and usage-based inspections under resource constraints • Notify user if constraints fail – to request guidance • 5 weeks under 3 minutes • Generation of A/C availability for second iteration with flight scheduler (SNAP) • Web-based integration framework to support joint negotiation between the two system

19. MIT MICANTS EFFORTS • Assisting in coordination with CACE • Further research on negotiation methods • Further research on prioritized constraint relaxation • Research on metrics • Research on models for preferences for Commander’s Intent

20. Commander’s Intent • Commanders generally think in terms of case-based reasoning • Decisions are strongly contextually linked • Preferences are: • Largely implicit • Multidimensional • Interdependent • More networked than hierarchical

21. Approach • We have investigated detailed scenarios (2) • The scenarios are sufficiently detailed to state clear plans for further action • The plans are examined to determine questions about preferences • Underlying preferences are exposed and discussed.

22. Scenario 1 • Basic elements • Squadron develops 12 month plan for 1200 flight hours • Plan runs 110 hours/month to achieve goal with 10 percent leeway • Squadron achieves 330 hours in first 3 months • All planes are grounded the whole fourth month • During the grounding period, the squadron • Catches up somewhat on maintenance, reducing or eliminating backlog of gripes • Doubles up on maintenance training to permit some skipping of Monday half-days when grounding period ends • Gets in all the training lectures for upcoming sorties • Loses some flight qualifications and generally gets rusty in skills • Problem: What should be the schedule for the remaining 8 months? • Fly original schedule with no remaining leeway for further difficulties? • Increase rate of flying? • Ask for more resources? • Ask for relief from mission? • Fail to accomplish mission?

23. Methods for getting more flight hours into the schedule • Lengthen sorties by X • Changing aircraft configuration (adding tanks, etc.) • Changing sortie profile (high optimal cruise burns less fuel) • Increase flight days • Fly on Saturdays • Steal or defer maintenance days • Move maintenance days to Saturdays • Increase flight hours per day • Extend day to more than 10 hours • Fly more planes (can try flying all available planes, including reserves, but risk wasting all preparations when one breaks down)

24. Commander’s Preferences • General preference order: • Lengthening sorties (from 1.2 to 1.4 hours) is best, if this accomplishes core competencies • Otherwise adding sorties is best • Among others, avoid increasing op tempo • Plan A. If we just need hours (e.g., outage occurs near end of year when core competencies have pretty much been met, but still short on hours) • First, lengthen sorties by profile from 1.2 to 1.4 hours • Then, lengthen sorties by configuration • Then, fly more sorties per day • Then, defer maintenance training periods • Plan B. If we need hours plus numbers of specific sorties to achieve competencies, then • First, fly more sorties per day • Then, fly more days

25. Plans • Timeline • Early 2002: • Short-term scheduling • Shift change support • Later 2002: • Hardening and extensions to support other A/C types • Deployment @ Yuma and Iwakuni • Framework refinements • New negotiation techniques: concurrent constraint propagation as negotiation • Constraints with preferences (MAXSAT or other) • Sophisticated constraint management in scheduler • Complexity experiments • Joint scaling properties (with flight scheduler)

26. Screenshots

27. A/C Status

28. Flight schedule

29. Guidance

30. Flight hours for missions

31. A/C utilization

32. Maintenance Plan

33. Overlaps

34. Maintenance schedule

35. Maintenance schedule

36. A/C availability