Aviation Safety Program Integrated Vehicle Health Management

National Research Council Assessment of NASA s Aeronautics Research Program Aviation Safety Program Integrated Vehicle Health Management Principal Investigator: Ashok N. Srivastava, Ph.D. Project Manager: Joseph E. Grady, Ph.D. Associate Project Manager: Susan Johnson 1

Project Goals and Objectives Agenda Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Project Goals and Objectives Aviation Safety Program Goals Develop technologies, tools, and methods to: - Improve aircraft safety for current and future aircraft - Overcome safety technology barriers that would otherwise constrain full realization of the Next Generation Air Transportation System - Concurrently, leverage these technologies to support space exploration activities, such as enabling self-reliant and intelligent systems necessary for the long-duration travel requirements of future space vehicles. IVHM Project Goals Develop technologies to determine system/component degradation and damage early enough to prevent or gracefully recover from in-flight failures in both the near-future and next-generation air transportation systems, and to provide enabling technology for space exploration. http:///pdf/168652main_nasa_fy08_budget_request.pdf (ARMD-12) 3

IVHM Project Goals

Project Goals and Objectives Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Selected IVHM Milestones

Project Goals and Objectives Agenda Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Cross-Project and Program Coordination Project or Program IRAC IIFD AAD Space Operations Mission Directorate Science Mission Directorate Exploration Systems Air Force Activity C-17, F/A-18, Airstar shared testbed Systems Analysis Architecture analysis Data mining for cockpit switching Crack propagation, aging and durability tools, providing SHM data Virtual Sensors and anomaly detection methods for ISS and Shuttle Virtual Sensors for anomaly detection Anomaly detection algorithms for Ares NASA-AF Executive Research Committee

IVHM4Info% National Aeronautics and Space Administration Partnership Status New Space Act Agreements: FLX Micro: to develop a high temperature, wireless pressure sensor Moog: to develop prognostic technologies for electromechanical actuators Boeing: to develop data mining technologies for anomaly detection Developing Space Act and Other Partnership Agreements: Sun Microsystems: Sunspot wireless sensors for prognostics Sikorsky: Testing and integration of Sunspot wireless sensors Lockheed Martin: Data mining anomaly detection methods on an F-16. 10

Moog SAA Prognostics for Electro Mechanical Actuator (EMA) Systems Moog Activities Moog has a set of 40+ EMAs dedicated for this work (power rating in aircraft actuator range) Provide guidance on developing new and/or simplified models for EMAs Assist in the EMA prognostic algorithm verification and validation via tests with seeded faults Data will be shared with NASA NASA Activities Develop physics-based models for actuator damage propagation Ball Screw Failures Screw Nut Circuit Screw Fracture Nut Fracture Ball Return Wall Thickness / Structural Circuit Orientation Ball Fracture Return Failure Thread Shear Wall Thickness / Structural Great Circle Generation Fatigue Damage Core Ductility Load Distribution Classical Brg Race Failures Contact Fatigue Brinelling False Brinelling Lubrication Misalignment Thread Shear Core Ductility Load Distribution Classical Brg Race Failures Contact Fatigue Brinelling False Brinelling Lubrication Misalignment Classical Brg Ball Failures Dimpling Wear Design Out Fail / Operate Jam Wear Damage Misaligment Rupture Contamination Ice Damming External Internal Screw / Nut Seperation 11

IVHM4Info% National Aeronautics and Space Administration Project Goals and Objectives Agenda Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Section D5: Fault-tolerant and Integrated Vehicle Health Management systems Fault-tolerant aircraft systems, coupled with CBM, may improve aircraft safety and reduce aircraft life-cycle maintenance and ownership costs. Critical research tasks include developing: 1. Robust and reliable hardware and software tools for monitoring components, detecting faults, and identifying anomalies; 2. Prognosis analysis tools for predicting the remaining life of key components; 3. Approaches for recovering from detected faults, including reconfiguration of the flight control system for in-flight failures of manned and unmanned aircraft; and 4. Low-cost, lightweight, wireless, self-powered sensors with greater memory and processing capability. From the Decadal Survey of Civil Aeronautics.

Rotordynamic Analysis NASA Decadal Survey: D5 and C1 Development of a physics based diagnostic system Disk imbalance causes synchronous whirl x T y Current model accounts for shifting eccentricity as function of crack characteristics and speed z Notch used to emulate crack for global, vibration response Machined flaw Milestone 1.3.1

Modeling Approach Equations of motion 2 2 && x + 2ξω x& + ω x = eω cosωt n n 2 2 && y + 2ξω y& + ω y = eω sin ωt y n C M e n U θ V Solution for undamaged case (fixed imbalance) { x y X = X cos t = Y sin t θ = tan = Y 1 = ( ω θ ) ( ω θ ) 2ξr 1 r 2 er 2 2 ( 1 r ) + ( 2ξr ) 2 2 r = ω/ω n These terms define the synchronous whirl amplitude. O x

Modeling Approach Modeling the crack dependence Total eccentricity is a function of crack characteristics and speed Sum of initial imbalance vector and imbalance due to crack y x Crack dependent solution 2 2 2 r e0 + ecr + 2e0ecr cos β = cos t 2 2 ( 1 r ) + ( 2ξr ) 2 ( ω ( θ + α )) α e 0 e β e 0 +e cr cos β e cr x e cr sin β 2 2 2 r e0 + ecr + 2e0ecr cos β y = sin t 2 2 ( 1 r ) + ( 2ξr ) 2ξr 2 ( ω ( θ + α )) 1 1 cr ( θ + α ) = tan + tan 2 1 r e0 + ecr cos β e sin β e = e + e = α = tan 2 2 0 + ecr 2e 0 ecr 1 e 0 e cr + e cos β sin β cr cos β Phase angle has new term due to the introduction of a crack Amplitude of e cr is defined by numerical analysis 6/4/2007

Results Phase, degrees 250 200 150 100 50 No notch 1.2 in notch at 0 degrees 1.2 in notch at 90 degrees 1.2 in notch at 180 degrees 1.2 in notch at 270 degrees 0 0 5000 10000 15000 Shaft speed, rpm 0.0025 Synchronous whirl amplitude, in 0.002 0.0015 0.001 0.0005 0 0 5000 10000 15000 Shaft speed, rpm

ADAPT Advanced Diagnostics and Prognostics Testbed Provide development platform with real, representative Electrical Power System (EPS) components for: Diagnostic and prognostic models and algorithms Advanced caution and warning methods Allow performance assessment of these algorithms Standardized testbed Repeatable failure scenarios 18

Preliminary Characterization of Sensor Placement Methodology Iterative Down-Select Process Candidate Sensor Suites i=0 i>0 Systematic Sensor Selection Strategy (S4) Architecture NASA Decadal Survey: D5 and B3 Candidate Selection Complete No Yes Final Selection Collection of Nearly Optimal Sensor Suites System Diagnostic Model Sensor Suite Merit Algorithm Down-Select Algorithm (Genetic Algorithm) Statistical Evaluation Algorithm Large Commercial Turbofan Engine Problem Characterization Health Related Information System Simulation Knowledge Base Optimal Sensor Suite Target Num Fan LPC HPC Combustor HPT Sensor Selection Results: LPT FADEC Control Sensors Optional Sensors Advanced Sensors T2A P2A P17 T17 T25 P25 W17 W25 W3 Target Wt. XN2 XN25 P2A T2A P17 T17 P25 T25 PS3 T3 T49 P5 T5 WF36 Fitness 9 0.000 X X X X X X X X X X X X 27.708 12 9 0.010 X X X X X X X X X X X X 26.901 12 9 0.100 X X X X X X X X X X 23.161 10 9 0.500 X X X X X X X X X 20.857 9 10 0.001 X X X X X X X X X X X X 27.653 12 Num Sensors T3 PS3 WF36 XN2 XN25T49 T41 P41 W41 Blue = typical sensors P48 W49 Green = optional sensors Red = advanced sensors T5 P5 W5 The overall fitness of different sensor suites selected by the S4 algorithm 19

Comparison of Unsupervised Anomaly Detection Methods, JANNAF 07 Outstanding Achievement in Liquid Propulsion Award in the Operational Systems category at the JANNAF 3rd LPS Meeting in May, 2007 NASA/Decadal Survey D5 and B3 Failure Data Failure Type Time of failure (in seconds) as recorded by each method Actual Orca GritBot SVM IMS LDS GMM STS-77 (#2) Anomalous Spike in Sensor Reading 74.42 N/A N/A N/A N/A N/A N/A STS-91 (#1) Sensor Failure 32.76 77.82 N/A N/A N/A N/A N/A STS-93 (#1) Controller Failure 11.38 25.04 15.44 25.04 25.04 N/A 12.64 STS-93 (#3) Fuel Leak and Controller Failure 11.62 23.44 176.24 23.44 23.44 17.44 11.84 A20619 Knife Edge Seal Crack 119 20.66 31.44 183.44 N/A 156.64 N/A A10853 Turbine Blade Failure 130 67.04 295.84 277.84 553.05 Blue: Fastest response times. Red: Highest accuracy. 130.54 N/A Milestone 1.6.1 20

Section D5 cont d: Fault-tolerant and Integrated Vehicle Health Management systems Evaluate component capability in a simulated environment (ground test and hardware in the loop). That is, take a particular subsystem, such as a real landing gear system that has been represented by an appropriate behavioral model as specified and insert simulated faults to test for proper operation of the health monitoring system. From the Decadal Survey of Civil Aeronautics.

727 Landing Gear Test Preparations in ALDF Description: The Aircraft Landing Dynamics Facility (ALDF) is capable of exercising a landing gear, for a brief extent, through the touchdown, rollout, and braking phase of aircraft landings at operational load and speed conditions while injecting subsystem faults. Data collected during the tests can be used to develop and verify diagnosis and prognosis algorithms. Preparations are underway for a series of 727 main landing gear tests in the ALDF. Over ten test runs are planned, including fault injections in the hydraulic system, anti-skid valves, brakes, and wheel speed sensors.

Section C1: Integrated Vehicle Health Management IVHM holds the promise of reducing vehicle cost, weight, and maintenance downtime, as well as speeding the introduction of new material systems and structural concepts. Real-time onboard sensor systems that monitor the actual state of materials and structural components enable more efficient use of material, including novel concepts With a better model of itself, the aircraft can trace back system anomalies through the multitude of discrete state and mode changes to isolate aberrant behavior. Fault-tolerant systems combine simple rule-based reasoning, state charts, model-free monitoring of cross-correlations among state variables, and model-based representations of aircraft subsystems. From the Decadal Survey of Civil Aeronautics.

Classification of Damage Signatures in Composite Plates Using One-Class SVMs, submitted to AIAA Journal 1 a b Classification Rate c 0.4 0.9 0.8 0.7 0.6 0.5 30 40 50 60 70 Torque 80 90 100 Milestone 1.6.1

Project Goals and Objectives Agenda Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Future Work Research and development of diagnostic and prognostic methods across structures, propulsion, and aircraft systems. Release of NASA Research Announcement. Fostering cross agency technology advancements to benefit IVHM. Data Mining in Aeronautics Science and Exploration Systems conference Building an IVHM Data Mining Lab Developing publicly available IVHM data. 26

Project Goals and Objectives Agenda Project Structure and Milestones Program Coordination and Partnership NASA IVHM Relationship to the Decadal Survey Future Work The IVHM Data Mining Lab

Mission of the IVHM Data Mining Lab The lab enables the dissemination of Integrated Vehicle Health Management data, algorithms, and results to the public. It will serve as a national asset for research and development of discovery algorithms for detection, diagnosis, prognosis, and prediction for NASA missions. Serves several High Priority R&T Challenges including C1, D5, E8B, and E8C.

Features of the IVHM Data Mining Lab Datasets Propulsion, structures, simulation and modeling ADAPT Lab Icing Electrical Power Systems Systems Analysis Flight and subscale systems Fleet-wide data Multi-carrier data Open Source Code Papers Generation of an IVHM community Selected Discovery Tools Inductive Monitoring System (IMS) cluster-based anomaly detection Mariana Text classification algorithm Orca Distance-based outlier detection ReADS Recurring anomaly detection system for text sequenceminer anomaly detection for discrete state and mode changes in massive data sets.

Key Research Issues Addressed in the IVHM Data Mining Lab Real-time anomaly detection Model-free prediction methods Hybrid methods that combine discrete and continuous data Distributed and privacy-preserving data mining Analysis of integrated systems

Appendix

DATA IVHM elements create large amounts of data The results of the data mining analysis, simulated data sets and new tools could be returned to the other research areas as well as disseminated to other NASA directorates, academia etc.., creating a mutually beneficial relationship of enhanced data analysis. IVHM Data Mining Lab There is a suite of tools already available for analysis, so researchers will not only be able to help analyze the data now, but the additional data will allow for continued R&D to develop algorithms and tools to answer evolving and complex questions (such as system-level questions).

From the Decadal Survey.Section B9 The growth of new MEA loads is being driven by advances in the capabilities of electric actuators and controls, and it is being enabled by the development of more flexible and reliable aircraft generators. Intelligent power management and distribution (PMAD) using advanced system models and wireless sensors or Sensor-less control technologies for graceful degradation and failsafe operation. From the Decadal Survey of Civil Aeronautics.

Diagnosing Electrical Power Systems Using Bayesian Networks NASA/Decadal Survey: B9 and D5 INPUT: OUTPUT: Description: Command nodes Bayesian networks offer an effective way to distinguish sensor faults or noise from component or system faults. They can model probabilistic, uncertain behavior but also exploit determinism in the system. For example, in circuit analysis, deterministic behavior takes the form of Ohm's law and Kirchhoff's laws. The goal of this effort is to develop a sensor validation approach that is powerful in domains with a combination of probabilistic and deterministic behavior; an example is the ADAPT lab. Sensor nodes BN Inference Health nodes Outcome/Results: Professor Darwiche and students from UCLA visited the ADAPT lab to continue collaborations with NASA researchers on the development of an approach for modeling both probabilistic and deterministic behavior by means of Bayesian networks. The approach involves the generation of a Bayesian network with the following types of nodes: Health nodes System health: represent health of system Sensor health: represent health of sensors Command nodes: commands from a testbed user Sensor nodes: read from ADAPT or from vehicle Push nodes: mediate flow being pushed from battery to load Pull nodes: mediate flow being pulled by load from battery To Load To Battery 35