Active Interrogation of Helicopter Rotor Faults Using Trailing

Active Interrogation of Helicopter Rotor Faults Using Trailing Edge Flap Actuation Patricia Stevens Doctoral Candidate Mechanical Engineering Penn State University Doctoral Dissertation Defense April 2, 2001 1

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Conclusions 2

Documented Rotor System Problems Civil • 1990 -1996: 35 civil rotorcraft accidents were caused by rotor system failures AH-64 Apache - Early Blade Problems • Original aluminum blades pitted by sand disabled by hail • Composite blades suffered from delamination CH-46 Sea Knight - Prior to Upgrade • Inspections as often as every 8 hours of flight time for some rotor components 3

What makes helicopter rotor damage detection so difficult? Aerodynamic Loads Gyroscopic System Centrifugal Stiffening 4 Complex Components Inaccessible Locations Noisy Environment

Previous work: Localized fault detection • Acoustic Emission Schoess et al. (1997) – Passive Approach – Acoustic Emission sensor “listens” for crack propagation Ultrasonic sensor crack • Wave Mechanics Lakshamanan & Pines (1997) & Purekar et al. (1998) – Active approach – Scattering of structural waves due to impedance changes • Limitation: – Requires a priori knowledge of fault location 5 stress waves Acoustic Emission PZT actuator / sensor flaw scattered waves Wave Mechanics

Previous work: Rotor Diagnostics using Fuselage Measurements Azzam & Andrew (1992, 1995) Ganguli, Chopra & Haas (1995 -98) • Passive generation of fixed frame loads • Measurements • relative blade position • fuselage vibration • Measurements in hover and forward flight • Limitations: • Limited detectability of small faults • Neural net required to classify faults • Forward flight condition measurements required Dissimilar blade model Seed fault Simulate response Measure tip displacement hub loads (vibs) Next flight condition Next fault 6 Fault profile at each flight condition Train Neural Net

Previous work: Application of Structural Damage Detection Kiddy & Pines (1997 - 1999) • Applied Modal Based SDD Technique to rotor blade environment • Modified Eigenstructure Assignment Technique to accommodate – Centrifugal Stiffening – Aerodynamic Damping • Limitations – Sensitive to noise – Limited fault coverage – Measurability & actuation not assessed Will an active interrogation structural damage detection approach yield improved results? 7

Next Generation Rotorcraft… Active Trailing Edge Flaps • Installed for vibration and noise control • Potential actuator for damage interrogation MD 900 blade with trailing edge flap Flap Actuator Tab Actuator Composite Blade Assembly Active Control Flap, Noise and Vibration Trim Tab, In-Flight Tracking HH 10 Airfoil Section Flap Actuator Bearingless Tab Actuator Hub BLADE CROSS-SECTION 8

Goal: Design and Evaluate the Active Interrogation Concept Interrogation signal sensors trailing-edge flap Blade Response Measured 9 Damage Evaluation Algorithms

Objectives v Determine if active interrogation of rotor faults using trailing edge flap actuators is a viable concept. v Develop active interrogation techniques appropriate for the rotor blade environment. v Demonstrate effective damage evaluation in hover. v Demonstrate damage evaluation in the presence of noise and modeling errors v Evaluate limitations of the approach. 10

Outline Background & Motivation Objectives of Work Modeling Approach Rotor Trailing Edge Flap Damage Identification Conclusions 11

Rotor Model - Bearingless Main Rotor • Finite Element Approach – Flap, torsion – 10 beam elements • Hingeless rotor - cantilever boundary condition • Dissimilar blades Flexbeam Pitch Link Stiffness f • Aeroelastic rotor in hover • Response via time integration 12 • Response measured at each node W Cantilever boundary condition Nodal Degrees of Freedom Nel = 10

Trailing-Edge Flap Model d Physical Description • Size 10% of rotor radius • Location 80 -90% rotor radius • Frequency 0 - 50 Hz. • Amplitude up to +/- 5 deg (using +/- 2. 5 deg) • Lift 120 lb/deg at 0 Hz 70 lb/deg at 50 Hz • Moment 25 ft-lb/deg 13 Aerodynamic Environment • Mach No. 0. 45 - 0. 6 in hover • Reduced Frequency up to 0. 5 (k=wc/2 V) • Requires subsonic compressible flow unsteady aerodynamic model (Leishman, et al)

Damage Models • Flexbeam Degradation – Bending Stiffness – Torsional Stiffness • Control System Stiffness • Flexbeam Crack • Outboard Stiffness Defect – Bending – Torsional • Outboard Crack • Ballistic Damage • Trim Mass 14

Flexbeam Degradation • Distributed stiffness fault • Change in EI or GJ over flexbeam element ® 15 5% reduction in EI or GJ for 0. 0 -0. 1 R (flexbeam element)

Control System Stiffness • Crack in pitch rod or fatigue failure in connecting hardware • 5% reduction in axial stiffness of pitch rod ® 16 5% effective reduction in torsional spring at end of flexbeam

Outboard Stiffness Defect • Adopted from Ganguli, Chopra and Haas (1995 -98) • Manufacturing Defect • Delamination ® 17 5% reduction in EI or GJ for 0. 6 -0. 7 R

Ballistic Damage • Experimental study of effects of ballistic damage (Robinson & Leishman, 97 -98) • Ballistic damage affects: – Cla, Clmax, Cd – aerodynamic center location – mass • “In some cases significant damage produced surprisingly mild effect on the aerodynamics” • “Mild decreases in lift, but major increases in drag” ® 18 Ballistic Damage = 5% decrease in mass from 0. 6 -0. 7 R

Loss of Trim Mass Discrete change in mass of 0. 6 lb at 95% radius mass nominal mass ® x/Lel feat her axis Trim Mass 19 Le l

Crack Model - a new finite element Krawczuk et al. (2000) Boundary Conditions A CRACK I II a LB H A L w 1(x) q 1 q 2 I x=0 20 f 1 (x) w 2 (x) cb=1/kb x=LB II x=LB q 3 f 2 (x) x=L q 4 From moment equilibrium

Crack Model - a new finite element • Converges to standard beam element as K 0 LB=L/2 Krawczuk et al. (2000) 21 • Only bending slope terms are affected

Elastic Crack Model Relating Crack Depth to Crack Constant • Correction function, F(a/H), takes into account crack and body geometry (from stress intensity factor): Effect of depth on crack constant K/H • Correction function governs flexibility (elastic crack) • Flexibility determines constant, K 22 a/H

Crack Model - Validation 23 [reproduced from Krawczuk et al. (2000)]

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Theory & Results Effect of Modeling Errors Noise & Noise Mitigation Alternate Extent Quantification Approach Measurability Conclusions 24

Structural Damage Detection Background (Modified from Rytter, 1993) Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life Can I safely complete my mission? 25

Structural Damage Detection Background Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life There’s a problem! 26

Structural Damage Detection Background Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life . . . in the pitch link! 27

Structural Damage Detection Background Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life It’s a crack! 28

Structural Damage Detection Background Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life It’s a small crack. 29

Structural Damage Detection Background Four Levels of Damage Identification – Level 1: Detection – Level 2 a: Level 1 + Location – Level 2 b: Level 1 + Characterization – Level 3: Level 2 + Quantification of Severity – Level 4: Level 3 + Prediction of Remaining Life Safe to complete the mission! 30

Damage Detection, Location & Characterization The "DAMAGE VECTOR" EOM of damaged system: Damage is perturbation matrix: Rearranging results in two equivalent vector expressions -d(jw) = the Residual Force or “Damage Vector” (1) d(jw) has non-zero elements only at DOFs associated with damage (2) d(jw) can be calculated from known parameters 31

Interpretation of the Damage Vector Physical interpretation: Ojalvo & Pilon (1988) The harmonic amplitude of nodal forces required to force the healthy system model to respond as if damage were present degrees of freedom: 1, 2 3, 4 5, 6 7, 8 9, 10 healthy d 3 measurements: 1, 2 d 4 3, 4 d 5 fint d 6 5, 6 7, 8 9, 10 damaged fint 32

Results for. . . • Flexbeam Degradation – Torsional Stiffness • Control System Stiffness • Outboard Stiffness Defect – Bending Stiffness • Outboard Crack • Ballistic Damage Need to • detect & locate • differentiate between similar faults Does interrogation frequency affect the results? 33

Damage Vector for Flexbeam Torsional Stiffness Damage is 5% decrease in GJ of element 1 displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M Torsional stiffness damage manifests as damage vector f DOFs connected to damaged element end-node twist f. A 34 measurement location

Damage Vector for Pitch Link Stiffness Damage is 5% decrease in torsional spring representing pitch link displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M end-node twist f. A 35 measurement location Pitch link stiffness damage manifests as damage vector f DOF connected to damaged element -- a single DOF

Damage Vector for Outboard Bending Stiffness Damage is 5% decrease in EI of element 7 displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M Outboard bending stiffness damage manifests as damage vector w & w’ DOFs connected to damaged element end-node twist f. A 36 measurement location

Damage Vector for Outboard Bending Crack Damage is crack of depth a/H=0. 05 at midpoint of element #7 displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M end-node twist f. A 37 measurement location Crack damage manifests as damage vector w' DOFs connected to damaged element

Damage Vector for Ballistic Damage is 5% decrease in mass of element 7 displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M Ballisitic damage manifests as damage vector w, w’, and f DOFs connected to damaged element end-node twist f. A 38 measurement location

Damage Vector for Ballistic Damage is 5% decrease in mass of element 7 displacement w bending slope w' 50 Hz mid-node twist 10 Hz f. M Why is damage vector contaminated? end-node twist Centrifugal Stiffening f. A 39 measurement location

Damage Vector for Compound Damage is displacement w –Root bending stiffness –Pitch link stiffness –Ballistic damage bending slope w' 50 Hz Results show mid-node twist 10 Hz f. M end-node twist f. A 40 measurement location –Each damage type is identified –Combined damage vector is equal to sum of individual damage vectors

Damage Detection, Location & Characterization Summary • Residual force vector (a. k. a. damage vector) requires – refined model of healthy system – measured response of damaged system – model or measurement of external force • All fault types studied were detected and located via residual force vector • Similar faults are distinguishable • Compound fault damage vector = sum of individual damage vectors • No clear frequency recommendation • Requires a single interrogation frequency 41

Why are rotor system damage extent calculations difficult? • Aerodynamic Loads – Non-symmetric aerodynamic matrices – Damping • Centrifugal Stiffening – large CF stiffening – mass / stiffness coupling • Coriolis Forces – Skew symmetric matrices 42

Damage Extent for Gyroscopic Systems • Yap and Zimmerman (1999) solved the gyroscopic problem via the “Asymmetric Minimum Rank Perturbation Theory” – Modal based model update – Find the perturbation matrix of minimum rank subject to constraint of null symmetry • This modal analysis based approach was extended to a FRF based approach as part of the current work 43

Damage Extent (step 2) FRF -"Asymmetric Minimum Rank Perturbation Theory” Stiffness damage: Damping damage: Mass damage: Where [ B ] =matrix collection of damage vectors (step 1) = [ d 1, d 2, …, dp ] [ j. Wint ] = diagonal matrix of interrogation frequencies [ X ] = matrix collection of damaged system response = [ {X(jw 1)}, {X(jw 2)}, …{X(jwp)} ] The number of independent columns of [ B ] and [ X ] is equal to the rank of the perturbation matrix (e. g. flap only: mass=4, stiffness=2) BUT! Must know nature (mass, damping, stiffness) a priori. 44

Calculation of Parameter Change Exact DK • AMRPT results in perturbation matrix of full dimension • For damage located in a single element, change in physical parameter is calculated using structure of elemental matrix • e. g. 45 1 5 10 DOF • Non-zero terms describe change in elemental matrix x 104 0 15 20 -1 25 30 -2 35 40 5 10 15 20 25 30 35 40 DOF

Mass Damage in Rotating Structure Observations: • Off diagonal terms in mass and CF stiffness matrices depend on c. g. offset - typically small • CF affects inboard elements in flapwise motion only • Neglecting off-diagonal terms, problem is now (3 x 3) in twist • Solve problem using twist DOFs only - still coupled in mass & stiffness Solution: • Iterate on coupled twist problem 46

Damage Extent Summary AMRPT Damage Extent Quantification Error AMRPT results show improvement using higher interrogation frequencies Errors stem from small errors in damage vector where x is damaged parameter (EI, GJ, r. A) 47

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Theory & Results Effect of Modeling Errors Noise & Noise Mitigation Alternate Extent Quantification Approach Measurability Conclusions 48

Effect of Modeling Errors Model Error: 10% stiffness error in baseline model Damage: 5% outboard bending stiffness 10% modeling error no error Damage Detection Destroyed!! 49

Correction of Modeling Errors Model Error: 10% stiffness error in baseline model Damage: 5% outboard bending stiffness Interrogation: +/- 2. 5 deg. , w = 40 Hz no correction corrected d no error Use damage vector correction: d=dd-dh 50

Effect of Modeling Errors on Damage Extent Calculations Damage: 5% outboard bending stiffness Interrogation: +/- 2. 5 deg. , w = 32 & 40 Hz AMRPT Extent Quantification Error Case % Error 10% Increase in Baseline Stiffness 45. 4 10% Decrease in Baseline Mass -17. 22 • Extent quantification error for perfect model = 0. 04% • Damage vector correction d=dd-dh is utilized Small errors in damage vector result in large errors in frequency domain AMRPT 51

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Theory & Results Effect of Modeling Errors Noise & Noise Mitigation Alternate Extent Quantification Approach Measurability Conclusions 52

Noise Uniform Random Noise in Harmonic Signal • How does measurement noise affect the results? 10% • How can noise effects be reduced? 5% 2% 53

Effect of Noise on Damage Vector Noise: uniform random noise Damage: 5% outboard bending stiffness Interrogation: +/- 2. 5 deg. , w = 40 Hz Damage Vector Magnitude Damage Vector Displacement DOFs Damage Vector Rotation DOFs 54 5% noise 2% noise no error

Noise Mitigation Procedure Cycle Averaging of Harmonic Signal with Noise % RMS Noise 10% uniform random noise 5% uniform random noise 2% uniform random noise 1% uniform random noise Number of Cycles in Average 55

Benefits of Cycle Averaging Noise: 5% uniform random noise Damage: 5% outboard bending stiffness Interrogation: +/- 2. 5 deg. , w = 40 Hz 5% noise 40 cycle average no error Damage Vector Magnitude Damage Vector Displacement DOFs Threshold = 1. 8 Damage Vector Rotation DOFs 56 Threshold = 2. 5

Effect of Noise on Damage Extent Calculations Damage: 5% outboard bending stiffness Interrogation: +/- 2. 5 deg. , w = 32 & 40 Hz AMRPT Extent Quantification Error Case % Error 2% noise 20 cycles 114% 5% noise 40 cycles 130% • Extent quantification error for perfect model = 0. 04% Small errors in damage vector result in large errors in frequency domain AMRPT 57

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Theory & Results Effect of Modeling Errors Noise & Noise Mitigation Alternate Extent Quantification Approach Measurability Conclusions 58

Damage Vector Sensitivity Approach • Motivation – Damage vector is clean & reliable – AMRPT very susceptible small errors in damage vector – AMRPT only applicable for null symmetric systems – Aerodynamic damage is non-symmetric • Does magnitude of damage vector indicate damage severity? 59

Damage Vector Sensitivity Approach Damage Vector Magnitude vs. Damage Extent Outboard Bending Stiffness Fault Damage Vector Magnitude • Simple relationship relates damage severity to damage vector magnitude • Nearly linear for small damage Damage Vector Magnitude 60

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Theory & Results Effect of Modeling Errors Noise & Noise Mitigation Alternate Extent Quantification Approach Measurability Conclusions 61

Measurability • Sensitivity – How large is the response of the healthy system at sensor locations? • Resolution – How does damage change the magnitude of response? 62

Measurability: Sensitivity Direct Measurements – Displacement < 0. 25" – Rotation < 0. 25 deg. – Twist < 1 deg. Strain Measurements – Bending Strain < 250 m-strain – Shear Strain < 60 m-strain • Frequency averaged 10 -50 Hz, 2 Hz step. • Peak-peak harmonic response amplitudes 63

Measurability: Change in Direct Measurement Response root bending stiffness root torsional stiffness root crack a/H=0. 05 root crack a/H=0. 2 pitch link tboard bending stiffness board torsional stiffness outboard crack a/H=0. 05 outboard crack a/H=0. 2 ballistic damage trim mass * Results averaged over frequency and blade length 64

Outline Background & Motivation Objectives of Work Modeling Approach Damage Identification Conclusions 65

Recommendations & Conclusions Summary – A unique active-interrogation damage evaluation approach for helicopter rotor systems using trailing-edge flap actuation has been designed and implemented in a numerical rotor code. – Residual force vector and AMRPT adapted for active interrogation approach – Residual force vector sensitivity approach formulated as alternative extent quantification approach – Detection & Extent demonstrated in hover using trailing edge flap actuation within the bounds set by vibration control requirements (< 50 Hz, +/- 2. 5 deg. deflection) – Effects of noise & modeling errors assessed and mitigated – Preliminary measurability study 66

Recommendations & Conclusions Successes – Damage detection • very clean for mass and stiffness faults • not sensitive to interrogation frequency • faults detected & characterized in the presence of 5% noise with cycle averaging • faults detected & characterized with 10% baseline model errors using damage vector correction – Damage extent measurement via AMRPT • stiffness faults within 5% error (without noise or modeling errors) 67

Recommendations & Conclusions Limitations – Damage extent very sensitive to noise & errors – AMRPT damage extent algorithm modified to account for CF stiffening effects BUT sensitivity to errors in damage vector is severe – AMRPT damage extent algorithm inappropriate for aerodynamic faults – Measurability • Typical change in response = 1% • Change in response << 1% for cracks, flexbeam torsional stiffness fault 68

Recommendations & Conclusions Remarks – Damage detection in helicopter main rotor using active interrogation with trailing edge flap is promising – Damage extent using frequency domain AMRPT is difficult due to sensitivity to errors in damage vectors Recommendations – Extent calculations using damage vector sensitivity – Optimize sensor placement – Optimize interrogation frequency – Implementation of strain-based approach – Investigate alternate detection and extent algorithms • non-linear time series feature extraction (Todd et al, 2001) 69

Acknowledgments This work was funded in part by – The ONR MURI in Integrated and Predictive Diagnostics through the Penn State Applied Research Lab – The ARO MURI in Active Noise and Vibration Control Technologies for Jet Smooth, Quiet Rotorcraft 70