Reliability and Risk in Theory and Practice

Transcription

1 Reliability and Risk in Theory and Practice Panel 1: Frontiers of Reliability Engineering University of Maryland April 2, 2014 Elias Anagnostou Engineering Fellow, Research and Technology

2 Need For Risk-Based Fleet Management Issues Uncertainties in legacy approach force conservatisms Austere budgets now drive the need to extract all remaining capabilities while minimizing risks Pervasive objectives to increase readiness and lower life cycle costs necessitate a change in the current paradigm New vehicle requirements for reduced weight and longer life dictate a need for highfidelity methods to manage risk Approach Advanced modeling and simulation tools that link materials-design-manufacturingsustainment (Digital Thread) Virtual representation of a system as an integrated system of data, models, and analysis tools applied over the entire life cycle on a tail-number unique basis (Digital Twin) Concurrent Uncertainty Management across the material system life cycle Enables improved reliability, affordability and maintainability with an overall goal to reduce total ownership costs Sustainment - Approximately 40% more life can be extracted without structural modifications (DARPA/Navy Structural Integrity Prognosis System (SIPS) demonstrated results) 2

3 DARPA/Navy Structural Integrity Prognosis System (SIPS) Prognosis system to manage uncertainty and provide actionable information for risk-informed fleet management Increase asset availability and reduce cost w/o increasing risk Approach Develop the underlying critical technologies that enable prognosis and the demonstration of these in an integrated PROGNOSIS system: Physics-based modeling that captures interactions between structural damage drivers and material failure mechanisms Sensors that measure critical vehicle and materials parameters Reasoning and predictive modules that accept, compare, interpret and correlate the data from the sensors and models to provide structural reliability predictions Defect Size Probabilistic Predictions Updated By Imperfect Sensor Evidence Anticipated Usage Update With Sensor Data Reasoning & Prediction Software System Physicsbased and future state Systems probabilities OUTPUT: Sensor Current Models Actual Usage 3

4 SIPS Program Organization Prognosis Program Program Manager - Madsen Principal Scientist - Papazian Materials & Modeling Anagnostou Sensor Systems Silberstein Reasoning & Predictions Engel System Architecture Teng Demonstrations Anagnostou Engel An integrated team of 75 engineers, scientists, professors and graduate students Computer Science Sensor Science Chemistry Info Management Experimentalists Structures Material Science Manufacturing Mathematics 4

5 SIPS Uncertainty Management Current State As-manufactured state Maintenance induced damage Repair effectiveness Material properties Missing & corrupted data Environment Usage uncertainty Assessment uncertainty False and missed indications Physics- Based Models Model uncertainty 5 Combines all the available information while accounting for their respective uncertainties

6 SIPS Research Progression to Flight Demonstration Fixed-Wing Structures Application 6

7 Multi-scale Environment Hole #14 Row of fastener holes 1 mm 5.7 mm 100 µm Microstructurally small cracks 7

8 Microstructural Origins of Fatigue (7075) 1 mm Microstructurally small cracks 8

9 Failure Progression From Initial State to Failure Multiple cracked particles cause multiple micro-structurally small cracks. Some arrest, some grow then link together and form the dominant crack that leads to failure. * Typical images from multiple samples 9

10 Physics-Based Models for Fatigue Life Prognosis Loading Direction 7075-T651 N total = N incubation + N nucleation + N small crack + N large crack K eff dc dn Ni Ci Keff K 2 max 1 Kc S S c F max x x FASTRAN/UniGrow FASTRAN (& Crack Coalescence) (Models for small crack growth & link-up) (a) (b) (c) c b 10 mm a 250 mm Cycle: Multi-Stage Fatigue/ VPS-MICRO (Models for nucleation & small crack growth) Geometric Approach (Models for incubation & nucleation stages by coupling experimental observations with micromechanical crystal plasticity simulations) 2 P max C inc N inc da G CTD CTDTH dn MSC / PSC n GS U ˆ GS CTD f C 0 II a CI GS Sut GS 0 HCF loading dominated LC 10

11 Multiscale Fatigue Modeling Environment 11

12 Multi-Scale Modeling: 3D Digital Materials Statistical Characterization of Material Digital Replication of microstructure Two program materials: 7075-T651 & 7050-T7451, and seven 7075-T651 legacy wing panel materials 12

13 Investigation of Damage Mechanics Experimental methods to characterize damage evolution Calibrate fatigue models at various length scales/damage mechanisms 13 Three specimens tested, ~1000 particles monitored per test

14 Framework Validation Experimental characterization of damage evolution Validation of probabilistic framework 14 Total of 35 Specimens tested, 5668 cracks measured

15 Model Integration Captures critical microstructurally-sensitive damage mechanisms Captures probability of occurrence of life-limiting fatigue mechanisms Produces naturally-occurring initial crack sizes for the start of small crack growth analysis to failure Tailored to the as-built manufactured state per aircraft tail number Initial Crack Size Distribution Bulk Material mstructure Statistics Grain Orientation Particle Aspect Ratio Particle Size Physics-Based Models for Crack Nucleation Response Surface to Response Surface to Select Cracked Particles Select Particles Most Most Likely to Spawn a Likely to Crack Crack into the Matrix (Incubation Filter) (Nucleation Filter) Physics-Based Initial Crack Size Distribution P(a) Material Cyclic Response at the Notch Particle Size, a Multiaxial Methods, Neuber & Glinka-ESEB Geometry, Material & Fatigue Loading Spectrum Load UniGrow/FASTRAN Predictions Small & Large Crack Growth to failure Defect Size a 1 Probabilistic Output t 0 p(t) t 1 Flight Hours Probabilistic Predictions vs. Experiment Incubation Nucleated at Cycle 30 80% of 1 st Cycle 1000 Cycles 15

16 EA-6B Outer Wing Panel Fatigue Test Prognosis Validation Original life predictions Prior flight history of panel Panel swap history Pre-test NDI Distribution of constituent particle sizes in 7075-T651 Predictions modified by: Null sensor readings, detection at sensor threshold, crack size estimates (all accounting for sensor accuracy and uncertainty characteristics) Bayesian reasoning system to make a probabilistic prediction based on uncertain input data As the test progressed: Significant decrease in life uncertainty Significant increase in predicted usable life Observed crack sizes validated predictions Predictions for largest crack Predictions converged to truth as test progressed 16

17 SIPS P-3 Flight Demonstration Workable Executable Prototype demonstration of a combination of systems consistent with Navy fleet management practice Onboard Sensors NDI (Omni-Scan) Sensor Reasoning & Prediction SIPS Physics Based Models Crack size HAZARD CATEGORIZATION FREQUENT (A) = or > 100 / 100K Flight Hours Flight Hours SEVERITY CATASTROPHIC (1) CRITICAL (2) MARGINAL (3) NEGLIGIBLE (4) FREQUENCY PROBABLE (B) / 100K Flight Hours OCCASIONAL (C) / 100K Flight Hours REMOTE (D) / 100K Flight Hours IMPROBABLE (E) = or < 0.1 / 100K Flight Hours NAVAIR Chose Vehicle & Sensor System

18 Results at a Critical Location Model predicted ~50% probability of a significant crack in a critical location Sensor had no indications Their combination reduced the probability to ~ 1% What Caused the Wide Disagreement Between Model and Sensor? During teardown inspections we discovered that the hole had been drilled out We presume this was done to remove an existing crack Based on the amount of material removed, model predicted repair would have been performed about 2001 No repair records were available, however Phased Depot Maintenance was performed 6/00 to 1/01 The Combination of Model Predictions and Sensor Evidence Agreed With All Teardown Findings 18

19