holger hermanns
play

Holger Hermanns dependable systems and software Saarland University - PowerPoint PPT Presentation

Holger Hermanns dependable systems and software Saarland University Saarbrcken, Germany Safety? Safety by design? Make sure hazardous situations are unreachable! Safety by design? Make sure hazardous situations are unreachable! Safety by


  1. Holger Hermanns dependable systems and software Saarland University Saarbrücken, Germany

  2. Safety?

  3. Safety by design? Make sure hazardous situations are unreachable!

  4. Safety by design? Make sure hazardous situations are unreachable!

  5. Safety by design? Why bother? Enforced by various standards: DO-178C/ED-12C for airborne systems relates to ARP4761 Functional Hazard Assessment (FHA) Preliminary System Safety Assessment (PSSA) System Safety Assessment (SSA) Fault Tree Analysis (FTA) Failure Mode and Effects Analysis (FMEA) Failure Modes and Effects Summary (FMES) Common Cause Analysis (CCA) ISO 26262 for automotive systems ... Higher/highest safety levels recommend formal methods

  6. Prelude

  7. Fault Trees 9x10 -4 5x10 -3 8x10 -3 5x10 -3 8x10 -3

  8. Fault Trees How to obtain the numbers? 1) Time-independent failure Average number of starts before failure: 200  Failure probability 0.005 9x10 -4 5x10 -3 8x10 -3 5x10 -3 8x10 -3

  9. Fault Trees How to obtain the numbers? 1) Time-independent failure Average number of starts before failure: 200  Failure probability 0.005 2) Time-dependent failure: 9x10 -4 On average once 0.00021 per Mission time : 24h Probability to fail in 24 hours: ���⋅�.����� 5x10 -3 8x10 -3 5x10 -3 8x10 -3 ... and from further models

  10. Fault Trees – Analysis Basics Calculate probability of top-level event … or overapproximation thereof 9x10 -4 5x10 -3 8x10 -3 5x10 -3 8x10 -3

  11. Fault Trees licensed at > 55% of • are often very large nuclear power plants worldwide • are very costly to maintain • are very important • are stateless • give imprecise results - too pessimistic due to stateless view + minimal cutset abstraction - too optimistic if dependencies - …

  12. Models for Safety All models are wrong, but some are useful. George E. P. Box

  13. Useful Models finite automata on? x:=0; on? x:=0; dark dark light light x==50 off!

  14. Useful Models finite automata with clocks all running at the same speed on? x:=0; on? x:=0; dark dark light light x==50 off! Timed Automata

  15. Useful Models finite automata with clocks and with costs incurred as time advances on? x:=0; on? x:=0; dark dark light light x==50 off! Priced Timed Automata

  16. Useful Models finite automata with clocks and with costs modular: composition of automata on? x:=0; y>d on? x:=0; on! y:=0; dark dark light light d:=U[5,55]; someone x==50 off! Automata Networks

  17. Useful Models 1 0,9 finite automata 0,8 0,7 0,6 with clocks 0,5 0,4 and with costs 0,3 U[5,55] Pr (“on!” > t) 0,2 Pr (“on!” > t) 0,1 modular: composition of automata 0 0 5 10 15 20 25 30 35 40 45 50 55 60 on? x:=0; y>d on? x:=0; on! y:=0; dark dark light light d:=U[5,55]; someone x==50 off! with probability distributions Stochastic Timed Automata

  18. Useful Models finite automata with clocks and with costs Exp[5] Pr (“on!” > t) modular: composition of automata on? x:=0; y>d on? x:=0; on! y:=0; dark dark light light d:= Exp[5]; someone x==50 off! with probability distributions Stochastic Timed Automata

  19. Useful Models finite automata with clocks memoryless time and with costs Exp[5] Pr (“on!” > t) modular: composition of automata on? x:=0; y>d on? x:=0; on! dark dark light light someone x==50 off! with probability distributions Markov Automata

  20. Useful Models finite automata with clocks and with costs Exp[5] Pr (“on!” > t) modular: composition of automata on? x:=0; y>d on? x:=0; on! y:=0; dark dark light light d:= Exp[5]; someone x==50 off! with probability distributions Stochastic Timed Automata

  21. Useful Models finite automata with clocks and with costs Exp[5] Pr (“on!” > t) modular: composition of automata on? x:=0; y>d on? x:=0; 98% on! 2% y:=0; dark dark light light d:= Exp[5]; someone T>85 && x==50 off! with probability distributions Stochastic Timed Automata

  22. Useful Models finite automata with clocks and with costs Exp[5] Pr (“on!” > t) modular: composition of automata on? x:=0; y>d on? x:=0; 98% on! 2% y:=0; dark dark light light d:= Exp[5]; someone T>85 && x==50 off! with probability distributions Stochastic Hybrid and continuous dynamics Automata

  23. Model based … Analysis System Model Analysis Focus possible behaviour Model Analysis Results

  24. Model based … Analysis ... Maintenance Failure ... Architecture Requirements Objectives Nominal System Model Analysis Focus possible behaviour Model Analysis Results Diagnostics Fault Trees FMEA Characteristics

  25. Model based … Analysis ... Maintenance Failure ... Architecture Requirements Objectives Nominal System Model Analysis Focus possible behaviour Model Analysis Results Diagnostics Fault Trees FMEA Characteristics

  26. Model based … Analysis ... Maintenance Failure ... Architecture Requirements Objectives Nominal System Model Analysis Focus possible behaviour Results Diagnostics Fault Trees FMEA Characteristics

  27. Model based … Analysis ... Maintenance Failure ... Architecture Requirements Objectives Nominal System Model Analysis Focus possible behaviour Model Analysis Abstraction Iteration Results Diagnostics … Fault Trees FMEA Magic Characteristics

  28. Model based … Analysis System Model Analysis Focus possible behaviour Model Analysis modestchecker.org Results A concrete, mission-critical case

  29. Embedded in Space

  30. GOMX-1 • 2U CubeSat (2 liter) • Launched in November 2013 • Payloads: • software defined receiver for aircraft signals • color camera for earth observation • Telemetry transmitted on amateur radio frequency • Massive amounts of data collected • battery voltage, temperature, solar infeed, … Runs our calibration experiments.

  31. Battery Kinetics

  32. Battery Kinetics 100 % 0 %

  33. Battery Kinetics B A Kinetic Battery Model • can represent ‘rate-capacity effect’ • can represent ‘recovery effect’  a faithful abstraction of modern battery chemistry

  34. Battery Kinetics B A Kinetic Battery Model • can represent ‘rate-capacity effect’ • can represent ‘recovery effect’  a faithful abstraction of modern battery chemistry

  35. Battery Kinetics full B A A empty B

  36. Battery Kinetics full B A A empty B

  37. Battery Kinetics full B A A empty B

  38. Battery Kinetics full B A A empty B

  39. Battery Kinetics full B A A empty B

  40. Battery Kinetics full B A A empty B

  41. Battery Kinetics full B A A empty B

  42. Battery Kinetics full B A A empty B

  43. Battery Kinetics full B A A empty B

  44. Battery Kinetics full B A A empty B

  45. Battery Kinetics full B A A empty B

  46. Concretely. Will the battery survive a one-year mission ? -62 with 5000 mAh

  47. Concretely. Will the battery survive a one-year mission ? With half the capacity? 2500 mAh

  48. Concretely. Will the battery survive a one-year mission ? With a quarter of the capacity? 1250 mAh

  49. Concretely. Will the battery survive a one-year mission ? With an eighth of the capacity ? 625 mAh

  50. Concretely. Will the battery survive a one-year mission ? With a sixteenth of the capacity ? 312.5 mAh

  51. GOMX-2 • 2U CubeSat (2 liter) • Shipped in October 2014 with Cygnus CRS-3 towards ISS • Payloads: • Optical communication experiments from NUS • Highspeed UHF and SDR receiver • Shipping failed after liftoff • Satellite was recovered from wreckage and returned to GomSpace

  52. GOMX-3 • 3U CubeSat (3 liter) • Launched from ISS in October 2015 • Payloads: • L-band communication to geostationary satellit • X-band transmitter for CNES • Highspeed UHF and SDR receiver • Can (and must) rotate in 3 dimensions

  53. GOMX-4 • Two 6U CubeSats (6 liter) • Launch expected in 2016 • Initial design in the making • Focus on support for flexible payload model • Needs strong support for dynamic load scheduling • Battery states are critical

  54. GOMX-3 mission details

  55. GOMX-3 mission planning • Very tight power budget • Needs dynamic and battery aware scheduling • What we do: • Priced Timed Automata modelling • Generate optimal schedules for 1 week or day horizon • Evaluate schedules on random KiBaM for robustness • Send to orbit, observe behaviour, update model

  56. GOMX-3 mission planning • Very tight power budget • Needs dynamic and battery aware scheduling • What we do: • Priced TA modelling with linear battery model • Generate optimal schedules for 1 week horizon • Evaluate schedules on random KiBaM for robustness • Prepare for updates of model state based on orbit data

  57. A one-day schedule (for yesterday) and its depletion risk

  58. Meeting Reality, safely

  59. You saw: Model based … Analysis System Model Analysis Focus possible behaviour Model Analysis modestchecker.org Results on a concrete, mission-critical case

Recommend


More recommend