1. Analysis of FRMS Forum data using BAM 2 September 2011
FRMS Forum, September 2011, Montreal
DRAFT, not for distribution!
Analysis of FRMS Forum data using BAM 2 September 2011
Tomas Klemets, Head of Scheduling Safety, Jeppesen

2. Background and Purpose About BAM
Content
Background and Purpose
About BAM
Features / Capabilities / Limitations
Analysis of data provided
Upcoming functionality

3. Background and Purpose

4. Jeppesen and Jeppesen Crew Solutions
3,000 employees
Denver, Frankfurt, Gothenburg, Montreal, Singapore, New York, Brisbane...
Navigation, Flight planning, and:
Crew Solutions: 500 people focused entirely on crew management.
Affecting some 250,000 crew daily.
Mostly crew planning, but also day-of-ops solutions

5. Why a need for models and BAM?
Regulatory rule sets, as well as union/pilot agreements are really binary fatigue models
Perfectly safe / Perfectly un-safe
Alignment with current science is so-so...
What you can’t measure...
Mathematical prediction models, even if not perfect, provides a continous metric...
...to be used for influencing, to push, an overall collection of crew schedules away from unneccessary fatigue
Crew scheduling with a metric for human physiology!

6. About BAM

7. The science behind BAM
Based on the Three Process Model of Alertness by Åkerstedt / Folkard
Predicts sleepiness
Sleep prediction enhanced to better reflect flight operation and take sleep oppurtunity into account
No published validation studies to date on airline crew but straightforward for an airline to check applicability
Returns continous predictions on a scale 0-10,000
High resolution a need for optimization
KSS , easy to close the loop
The Karolinska Sleepiness Scale - KSS
The Common Alertness Scale - CAS

8. Science (2) - Most recent and relevant references

9. BAM features
BAM is built to support the complex crew management processes for airlines of all sizes.
Integration with industry strength optimizers generating up to 6000 rosters per second over many hours
Initial state assumptions for pairing construction
Augmentation, acclimatisation...
Customizability
Habitual sleep length, Diurnal type, Transfer times
Sleep/wake overrides
Prediction point

10. BAM features (2) CAPI compliant – easy to connect and exchange
Usable in a more comprehensive risk layer taking mission context into account: weather, airport properties, crew experience, light conditions, etc...
Limitations
Predicts the average of a population
Does not take actual light conditions into account, approximates via time zone
Sleep inertia is not implemented
BAM (as any model?) should primarily be used to rank relative fatigue between flights

11. BAM evolution
BAM is built to self-tune in a closed-loop system to collected data
Airline collections
Crowd sourcing (FDC 2011)

12. Analysis of supplied data sets

13. A B C D The task: Pairings Rosters Short haul Long haul
1094 / 90
[flights / chains]
3693 / 56
Short haul C D
188 / 47
1006 / 64
Long haul
1. Which are the worst flights and why? Example also of good ones.
2. Which are the worst pairings/rosters and why? Examples also of good ones.
(Ignore mission context – all flights are equally ”difficult”).

14. Tools used for the analysis
CFAS
CrewAlert
Scenarios

15. Analysis data set A
– short haul pairings

16. Overall solution statistics

17. Some of the worst flights *)A
*) From a fatigue risk perspective ignoring mission difficulty
Time of day. Slight sleep deprivation previous night, no/little chance for afternoon sleep when departing home 15:46. Many consecutive days and sectors adds a bit to the problem.
615

18. Some of the worst flights (2)A
Time of day. Slight sleep deprivation previous night, no/little chance for afternoon sleep when departing home 15:46. Many consecutive days and sectors adds a bit to the problem.
618

19. Some of the worst flights (3)A
Time of day. Slight sleep deprivation previous night, no/little chance for afternoon sleep when departing home 15:46. Many consecutive days and sectors adds a bit to the problem.
769

20. Some of the worst flights (4)A
Time of day. Sleep deprivation from falling asleep at 3AM. Many consecutive days and sectors adds a bit to the problem.
1154

21. Some of the worst flights (5)A
Time of day. Two-pilot operation through the WOCL. Sleep deprivation. 2h acclimatisation west – but small effect.
1169

22. Data set A – low risk flights…

23. A
Day time. Sensitive to early sleep in the evenings due to early starts.
>5000

24. A
Day time. Long duties, but well placed.
>5000

25. Data set A – worst and best pairings…

26. Risk...
The operational risk we try to adress is for a flight to suffer an ”adverse event” with crew fatigue as a contributor or a direct cause
A flight!
The total operational risk (of this type) for the airline is the sum over all flights.
Most likely a weighted sum...
A pairing or a roster can rarely be modified in isolation!
All flights in a crew scheduling problem need to be assessed at once...

27. Risk for a flight vs. risk for a pairing or a roster...
Lowest point during a flight, top of descent, average, or time below treshold.
Doesn’t really matter when reshuffling a sequence of flights!
What is worst? (recall low is bad...)
Pairing 1; active flights on 600, 3000, 3300, 2700
Pairing 2; active flights on 700, 700, 700, 700

28. The operational risk for an airline...
CP hour on type
CP hours total
CP airport recency
CP predicted alertness
FO hour on type
FO ...
Airport elevation
Runway length
Light conditions
Wind direction/force
Rain/hail/snow
Visibility
Airport equipment
Airport terrain
Airport traffic
Aircraft MEL items
...
(More on this in the proceedings from IASS 2009)

29. BAM and operational risk
There is no sharp threshold on predicted alertness where risk suddenly goes from non-existant to non-acceptable
Risk grows exponentially when approaching 0
BAM is built to adress also the total risk
All flights in the lower tail of the alertness distribution makes sense to improve R A

30. BAM and operational risk (2)
When constructing pairings and rosters – one alertness value per flight is sufficient
BAM is configurable and supports using either:
Lowest point during flight
Lowest point during a customizable part of the flight
Prediction at a certain point in the flight – like TOD
(True fatigue risk management takes mission difficulty into account when prioritising crew assignments.)

31. Analysis data set B
– short haul rosters

32. Overall solution statisticsB

33. Some of the worst flights (1)B
392

34. Some of the worst flights (2)B
405

35. Some of the worst flights (3)B
425

36. Some of the worst flights (4)B
578

37. Data set B – low risk flights…

38. B
>5000

39. B
>3000

40. Example that it does not have to be that bad
>xxx

41. Data set B – worst and best rosters…

42. ...
<added later>

43. Analysis data set C
– long haul pairings

44. Overall solution statistics

45. Some of the worst flights (1)C
842

46. Some of the worst flights (2)C
1377

47. Some of the worst flights (3)C
1402

48. Some of the worst flights (3)C
1402

49. Data set C – low risk flights…

50. C
>5000

51. Data set C – worst and best pairings…

52. ...
<added later>

53. Analysis data set D
– long haul rosters

54. Overall solution statisticsD

55. Some of the worst flights (1)D
487

56. Some of the worst flights (3)D
926

57. Some of the worst flights (3)D
956

58. Some of the worst flights (4)D
971

59. Data set D – low risk flights…

60. D
>5000
Best use of extra crew?

61. D
>5000

62. Data set D – worst and best rosters…

63. Summary of BAM capabilities

64. BAM – science Open Science Open sleep assumptions
Openly available for scrutiny
Open Science
Open sleep assumptions
Data driven improvement

65. BAM - individualisation
Commuting / transfer times
Floating sleep predictions
Augmentation
Acclimatisation
Customizable sleep
Diurnal type
Habitual sleep length
Openly available for scrutiny
Open Science
Open sleep assumptions
Data driven improvement

66. BAM - applicability >150 k p/s Risk layer Paralell computing
Customizable prediction point
Rostering optimization
Integration
Jeppesen Support and development
Pairing optimization
Commuting / transfer times
Floating sleep predictions
Augmentation
Acclimatisation
Customizable sleep
Diurnal type
Habitual sleep length
Openly available for scrutiny
Open Science
Open sleep assumptions
Data driven improvement

67. Questions?

68. Backup slides...

69. BAM evolution

70. BAM and continous improvement
Data driven improvement strategy - MUSIC
Manage
Use
Save
Improve
Compare

74. The crew management process

75. Where should fatigue be managed?
Crew management processes
Follow-up
Day of
operation
Maintain
planning
Planning
Mid term manpower
Long term manpower
FEB
MAR *
MAR
APR
...JUL
…MAR‘10
Today
*Today
Correct data
Salary events
Passenger focus
Legality / feasibility
Secure revenue
Use reserves
Trip trades
Maintain productivity Maintain sby levels Crew quality
Productivity
Real costs
Robustness
Quality of life
Enough instructors?
Leave
Leave of absence?
Move crew btw bases?
Adjust the schedule?
Recruit?
Transition training?
Base size?
Qualification structure in cabin?
Crew negotiations?
Leave
Promote instructors?

76. Answer: Where it’s introduced.
Crew management processes
Follow-up
Day of
operation
Maintain
planning
Planning
Mid term manpower
Long term manpower
FEB
MAR *
MAR
Station, Departure time, Equipment, Augmentation, Choice of hotel, Deadheading, ...
APR
...JUL
…MAR‘10
Today
*Today
Time table planning
Manpower Planning
Applications:
Crew Pairing
Crew Pairing
Maintain what has been planned...
Crew Rostering
Crew Rostering
Crew Tracking
The flight ”context”: Surrounding activities/flights on the roster, Individual history and circumstances

77. Sequence of flights + attr. Sequence of flights + attr.
The overall concept...
Take fatigue / alertness into account while recombining the sequence of flights
Flight Schedule
Risk
Crew Pairing
Rules
Objectives
Crew Rosters
Crew Rostering
CAPI
Fatigue Model
Predicted Alertness
Sequence of flights + attr.
Fatigue Model
Predicted Alertness
Sequence of flights + attr.
The Common Alertness Prediction Interface
Fatigue Model
Fatigue Model 77

78. The Jeppesen FRM portfolio

79. Context sensitive mitigation advice
Products
Context sensitive mitigation advice
(Jan/Feb 2012)
CrewAlert
Get started using a fatigue model
Get aquainted with a model. Investigate individual patterns and see how science “plays out“ on a roster. Collect fatigue data easily from your operation! 1

80. 2 1 Products CFAS Extensive Fatigue Assessment with any solution
CFAS
Extensive Fatigue Assessment with any solution
Fatigue Assessment on thousands of pairings or rosters in seconds through a web service. Show control and progress internally and to a regulator. Learn and improve from bad patterns. Use with any system! 2
CrewAlert
Get started using a fatigue model 1

81. 3 2 1 Products BAM Your current system CAPI CAPI
Integrate in your environment
Your current system
CAPI 3
CFAS
Extensive Fatigue Assessment with any solution
The CAPI1 interface is available for licensing enabling a direct integration with BAM and other compliant fatigue models. Allows for adding direct decision support and visualization for planning / re-planning. Requires a system change. 1) The Common Alertness Prediction Interface 2
CrewAlert
Get started using a fatigue model 1

82. 4 3 2 1 Products Jeppesen Crew Solutions Optimize using a model CAPI
Robustness
Real Costs
Productivity
Quality
Jeppesen Crew Solutions
Optimize using a model 4
Alertness
CAPI
Integrate in your environment 3
CFAS
Extensive Fatigue Assessment with any solution 2
CrewAlert
Get started using a fatigue model
Boost alertness while constructing your crew schedules with Jeppesen optimizers. Implement a minimum alertness level and/or introduce incentives to boost alertness in full control of the balance with other factors. Identify FTL/LBA loopholes and wise alleviations. 1

83. 4 3 2 1 Products BAM, Boeing FAID, Interdynamics SAFE, Qinetiq
Jeppesen Crew Solutions
Optimize using a model
Jeppesen strives to provide all the leading fatigue models throughout the portfolio. The CAPI specification has been shared and confirmed to fulfill the data provisioning needs of several leading models. 4
CAPI
Integrate in your environment
Status Aug’11: Only BAM fully compliant to CAPI 2.0 3
Fatigue Models
BAM, Boeing
FAID, Interdynamics
SAFE, Qinetiq
SAFTE, IBR
CFAS
Extensive Fatigue Assessment with any solution 2
CrewAlert
Get started using a fatigue model 1

84. Consultancy Impact assessments, finding relaxations, …
Services
Consultancy Impact assessments, finding relaxations, …
Fatigue Models
Assess your current planned or actualized rosters. Investigate options. Re-run with world class optimization using also FRM capabilities. Sensitivity analysis…

85. Consultancy Impact assessments, finding relaxations, …
Services
Consultancy Impact assessments, finding relaxations, …
Training Three-day training course for planners, safety pilots and managers
Fatigue Models

86. Consultancy Impact assessments, finding relaxations, …
Services
Consultancy Impact assessments, finding relaxations, …
Training Three-day training course for planners, safety pilots and managers
Fatigue Models
Data collection surveys Using CrewAlert but also “full surveys” offered via Boeing
Data collected with crewAlert is uploaded to Jeppesen and directly structured for further processing/analysis. Avoiding interpetation and formatting of paper work often taking weeks or months to complete…

87. Services Consultancy Impact assessments, finding relaxations, …
Training Three-day training course for planners, safety pilots and managers
Fatigue Models
Data collection surveys Using CrewAlert but also “full surveys” offered via Boeing
Service Bureau Remote Planning using FRM
Often a continuation of a consultancy study allowing for quickly using the improved planning results – with Jeppesen staff doing the scheduling. Pairings and rosters.

88. 4 3 2 1 Products Services www.jeppesen.com/frm
Jeppesen Crew Solutions
Optimize using a model
Consultancy Impact assessments, finding relaxations, … 4
CAPI
Integrate in your environment
Training Three-day training course for planners, safety pilots and managers 3
Fatigue Models
CFAS
Extensive Fatigue Assessment with any solution
Data collection surveys Using CrewAlert but also “full surveys” offered via Boeing 2
CrewAlert
Get started using a fatigue model
Service Bureau Remote Planning using FRM 1

89. Crew Alert

90. Sleep journals with actual sleep/wake Predicted alertness average
CrewAlert
Flight duties
An evolving application to make sleep science more available.
Currently meant primarily for crew schedulers and safety pilots to learn about sleep science (as represented by BAM)
Also built for collecting operational fatigue data
In work…
A schedule communication tool
Fatigue mitigation advice for crew
Predicted sleep
Sleep journals with actual sleep/wake
Predicted alertness average
Self assessments
Predicted alertness 90%

91. CrewAlert – collecting data
Data collection has in the past been cumbersome, fragmented and quite expensive…
Collected data is by design now:
quality assured at entry
well structured
securely delivered back to the airline safeguarding personal integrity.
Not a “full scientific study”, but a very cost effective alternative