UNIVERSITY OF CINCINNATI September 22 03 _____________ , 20 _____ Douglas Morgan Coombs I,______________________________________________, hereby submit this as part of the requirements for the degree of: Master of Science ________________________________________________ in: Mechanical Engineering ________________________________________________ It is entitled: Detection of Structural Nonlinearities Using Combined ________________________________________________ Coherence ________________________________________________ ________________________________________________ ________________________________________________ Approved by: ________________________ Dr. Randall J. Allemang ________________________ Dr. David L. Brown ________________________ Dr. Allyn W. Phillips ________________________ Dr. Jay Kim ________________________ Detection of Structural Nonlinearities Using Combined Coherence A Thesis submitted to the Division of Research and Advanced Studies Of the University of Cincinnati In partial fulfillment of the requirements for the degree of MASTER OF SCIENCE in the Department of Mechanical Engineering of the College of Engineering 2003 by Douglas Morgan Coombs B.S.M.E. University of Portland 1999 Committee Chair: Dr. Randall Allemang Abstract A frequency domain method of detecting and spatially locating structural nonlinearities using measurements made during a typical modal test is further developed and evaluated in this thesis. The combined coherence function contains drops due to digital signal processing (DSP) errors, however it is unaffected by nonlinearities. By comparing combined coherence to ordinary coherence (or multiple combined coherence to multiple coherence) one can determine whether or not a drop in ordinary coherence is due to a nonlinearity or a DSP error. Previous work was done by Tom Roscher and Dr. Randy Allemang with theoretical data generated from a lumped parameter (M, K, C) model [1, 2]. This paper expands on that work in two ways: 1) By applying the formulation to a “real-world” test structure. 2) By expanding the theoretical background to enable the use of post-processed data for multiple input testing with Multiple Combined Coherence (MCCOH) In applying combined coherence to two physical substructures with linear and/or nonlinear connections, the results of several testing scenarios are examined to see how they affect the ability of combined coherence to spatially locate nonlinearities. These include: 1) Cases with and without leakage errors 2) Varying the number of spectral averages 3) Reducing the number of nonlinear paths 4) Varying the input force locations 5) Changing the spatial density of the responses Acknowledgements There are many people I would like to thank for making this work possible. First, I would like to thank Dr. Randy Allemang and Dr. Allyn Phillips for the hours spent working with me on this, supplying needed direction and feedback. Without them this project would not have been possible. I would also like to thank Dr. David Brown and Dr. Jay Kim for the insight they would offer on various problems throughout my education here. I came to UC to learn about experimental vibrations and could not have found a better group of people to do so under. Second, I would like to thank my wife for allowing me to turn down a good job offer and move over 2300 miles away from family and friends in order to pursue further studies at SDRL. For all of her love, encouragement and support, I am most grateful. I would also like to thank Dr. Miroslav Rokos, retired professor of the University of Portland, for seeing my potential when I could not and always pushing me to go to graduate school. Without his impetus I would have remained in the working world and missed out on one of the greatest experiences of my life. Last of all, I would like to thank God for all the gifts he has given me and Mother Mary for keeping me in her prayers. Table of Contents List of Figures ................................................................................................................................ iii List of Tables ................................................................................................................................ vii Nomenclature................................................................................................................................. ix Abbreviations.................................................................................................................................. x Introduction and Background ......................................................................................................... 1 Definition and Application of Combined Coherence and Multiple Combined Coherence ........ 1 Current Methods of Detecting and Identifying Nonlinearities ................................................... 3 Definition and Application of Ordinary and Multiple Coherence.............................................. 5 Combined Coherence Equation Development............................................................................ 7 Theoretical Basis for Equation................................................................................................ 7 Mathematical Development of Combined Coherence ............................................................ 8 Mathematical Development of Multiple Combined Coherence ............................................. 9 Theoretical Example ................................................................................................................. 11 Applying the Theory to a Real Structure ...................................................................................... 16 Test Subject............................................................................................................................... 16 Case 1: 4 Nominally Linear Connection Points Between Structures ....................................... 20 Case 2: Effect of Cyclic Averaging on Nominally Linear Case 1............................................ 25 Cases 3-5: Effect of Varying the Force Input with Four Nominally Nonlinear Connection Points Between Structures ........................................................................................................ 30 Case 6: Effect of Cyclic Averaging on Nominally Nonlinear System ..................................... 41 Case 7: Spatial Resolution of the Sensors ................................................................................ 47 Case 8: Effect of Spatial Variation of the Input Forces............................................................ 49 i Case 9: Effect of Reducing the Number of Connection Points Between the H-Frame and Square Frame to One ................................................................................................................ 53 Cases 10-11: Effect of Spectral Averaging............................................................................... 66 Discussion ..................................................................................................................................... 73 Summary and Conclusions ........................................................................................................... 77 Future Work .................................................................................................................................. 79 Appendix 1: Derivation of Confidence Intervals for Ĥ (ω ) and θˆ(ω ) (i.e., ∠Ĥ (ω ) ) ............... 83 ii List of Figures Figure 1: Three DOF Lumped Mass System .................................................................................. 7 Figure 2: Theoretical Lumped Parameter (M, K, C) System used in example............................. 11 Figure 3: Comparison of FRF and COH for Linear and Nonlinear Theoretical Systems at a DOF directly affected by the nonlinearity ..................................................................................... 13 Figure 4: Comparison of FRF and COH for Linear and Nonlinear Theoretical Systems at a DOF directly affected by the nonlinearity ..................................................................................... 13 Figure 5: Comparison of FRF and OCOH Magnitudes for Linear and Nonlinear Theoretical System at a DOF not directly affected by the nonlinearity................................................... 14 Figure 6: OCOH and CCOH for a Theoretical Hardening Stiffness Case ................................... 15 Figure 7: OCOH vs. CCOH for a Theoretical Hardening Stiffness Example .............................. 15 Figure 8: H-Frame and Square Frame Test Structures ................................................................. 16 Figure 9: Pictures of Shaker Setup showing skew input directions............................................. 17 Figure 10: Accelerometer and force input locations using a spread out spatial distribution....... 18 Figure 11: Accelerometer locations using a more concentrated spatial distribution ................... 19 Figure 12: Layout of connections between the H-frame and the square frame for Cases 1 and 2 21 Figure 13: Picture and schematic of the Bolted Joint for Cases 1 and 2 ...................................... 21 Figure 14: FRF and coherence functions showing the 54.25 Hz resonance peak at which combined coherence improves over multiple coherence ...................................................... 23 Figure 15: FRF and coherence functions showing the 38.25 Hz resonance peak at which combined coherence improves over multiple coherence ...................................................... 24 Figure 16: Overlay of FRF and Coherence Plots Showing the Effect of Cyclic Averaging on the Nominally Linear System ..................................................................................................... 26 iii Figure 17: Overlay of FRF and Coherence Plots Showing the Effect of Cyclic Averaging on the Nominally Linear System ..................................................................................................... 26 Figure 18: Case 2 FRF and coherence functions showing the 177.75 Hz antiresonance valley at which combined coherence improves over multiple coherence ........................................... 28 Figure 19: Case 2 FRF and coherence functions showing the 21.5, 25.25 and 772 Hz antiresonance valleys at which combined coherence improves over multiple coherence .... 29 Figure 20: Layout of connections between the H-frame and the square frame for Cases 3-8...... 30 Figure 21: Schematics of the two connection types between the structures for Cases 3-8 .......... 31 Figure 22: Case 3: MCCOH for all DOFs at 24.25 Hz................................................................. 36 Figure 23: Case 4: MCCOH for all DOFs at 53.75 Hz................................................................. 37 Figure 24: Case 5: MCCOH for all DOFs at 396 Hz.................................................................... 37 Figure 25: Case 4 FRF and coherence functions showing the 24.25 Hz small resonances and 100.75 Hz antiresonance valleys at which combined coherence improves over multiple coherence .............................................................................................................................. 38 Figure 26: Case 4 FRF and coherence functions showing the 53.75 Hz resonant peak at which combined coherence improves over multiple coherence ...................................................... 39 Figure 27: Case 4 FRF and coherence functions showing the 395 Hz antiresonance valley at which combined coherence improves over multiple coherence ........................................... 40 Figure 28: Case 6: MCCOH for all DOFs at 24.25 Hz................................................................. 43 Figure 29: Case 6: MCCOH for all DOFs at 52 Hz...................................................................... 43 Figure 30: Case 6: MCCOH for all DOFs at 53.5 Hz................................................................... 44 Figure 31: Case 6: MCCOH for all DOFs at 397 Hz.................................................................... 44 iv Figure 32: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System................................................. 45 Figure 33: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System................................................. 45 Figure 34: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System................................................. 46 Figure 35: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System................................................. 46 Figure 36: Case 7: MCCOH for all DOFs at 396.75 Hz............................................................... 48 Figure 37: Case 7 FRF and coherence functions showing the 123.25 and 175 Hz antiresonance valleys at which combined coherence improves over multiple coherence ........................... 50 Figure 38: Case 7 FRF and coherence functions showing the 175 Hz antiresonance valley at which combined coherence improves over multiple coherence ........................................... 51 Figure 39: Overlay of H(45x,*x) before and after moving shaker #1 from the H-Frame to the Square Frame ........................................................................................................................ 52 Figure 40: Pictures of the setup for Cases 9-11 showing the removal of 3 connection points and the suspension of the square frame by shock cord................................................................ 53 Figure 41: Layout and schematic of the connection point between the H-frame and the square frame for Cases 9-11 ............................................................................................................. 54 Figure 42: Case 9: MCCOH for all DOFs at 21.5 Hz................................................................... 57 Figure 43: Case 9: MCCOH for all DOFs at 401.5 Hz................................................................. 57 Figure 44: Case 9: MCCOH for all DOFs at 512 Hz.................................................................... 58 Figure 45: Case 9: MCCOH for all DOFs at 632.5 Hz................................................................. 58 v Figure 46: Case 9: MCCOH for all DOFs at 800 Hz.................................................................... 59 Figure 47: Case 9 FRF and coherence functions showing the 21.5 Hz combined coherence improvements over multiple coherence ................................................................................ 60 Figure 48: Case 9 FRF and coherence functions showing the 45 Hz combined coherence improvements over multiple coherence ................................................................................ 61 Figure 49: Case 9 FRF and coherence functions showing the 512 Hz combined coherence improvements over multiple coherence ................................................................................ 62 Figure 50: Case 9 FRF and coherence functions showing the 400 Hz combined coherence improvements over multiple coherence ................................................................................ 63 Figure 51: Case 9 FRF and coherence functions showing the 400 and 635 Hz combined coherence improvements over multiple coherence............................................................... 64 Figure 52: Case 9 FRF and coherence functions showing the 800 Hz combined coherence improvements over multiple coherence ................................................................................ 65 Figure 53: Case 12: MCCOH for all DOFs at 638.25 Hz (25 spectral averages)......................... 68 Figure 54: Case 12: MCCOH for all DOFs at 638.25 Hz (100 spectral averages)....................... 69 Figure 55: Case 12: MCCOH for all DOFs at 800 Hz (25 spectral averages).............................. 70 Figure 56: Case 12: MCCOH for all DOFs at 800 Hz (100 spectral averages)............................ 71 Figure 57: Overlay of FRF and Coherence Plots Showing the Effect of a Spectral Averaging for DOF 43 +X ........................................................................................................................... 72 vi List of Tables Table 1: Various Excitation Types and Their Ability to Mask or Highlight Nonlinear Effects in FRF’s....................................................................................................................................... 2 Table 2: Force, Mass, Stiffness and Damping Values of Theoretical 4 DOF System.................. 12 Table 3: Test Setup for Case 1...................................................................................................... 20 Table 4: Table of Improvements in Combined Coherence for Case 1 ......................................... 22 Table 5: Test Setup for Case 2...................................................................................................... 25 Table 6: Table of improvements of 2% or more in Combined Coherence for Case 2 for adjacent points..................................................................................................................................... 27 Table 7: Test Setup for Cases 3-5 ................................................................................................. 30 Table 8: Summary of Results for Cases 3-5 ................................................................................. 32 Table 9: Case 3 improvements of 20% or more in Combined Coherence for Case 3 for adjacent points in the same direction .................................................................................................. 33 Table 10: Case 4 improvements of 20% or more in Combined Coherence for Case 4 for adjacent points in the same direction .................................................................................................. 34 Table 11: Case 5 improvements of 20% or more in Combined Coherence for Case 5 for adjacent points in the same direction .................................................................................................. 35 Table 12: Test Setup for Case 6.................................................................................................... 41 Table 13: Table of improvements of 10% or more in Combined Coherence for Case 5 for adjacent points in the same direction (consecutive frequencies enumerated but not listed). 42 Table 14: Test Setup for Case 7.................................................................................................... 47 Table 15: Table of improvements of 20% or more in Combined Coherence for Case 7 for adjacent points in the same direction (consecutive frequencies enumerated but not listed). 48 vii Table 16: Test Setup for Case 8.................................................................................................... 49 Table 17: Table of improvements of 20% or more in Combined Coherence for Case 8 for adjacent points in the same direction .................................................................................... 49 Table 18: Test Setup for Case 9.................................................................................................... 53 Table 19: Case 9 improvements of 20% or more in Combined Coherence for Case 9 for adjacent points in the same direction (closely spaced frequencies enumerated but not listed)........... 56 Table 20: Test Setup for Cases 10-12 ........................................................................................... 66 Table 21: 10 Greatest Differences between MCCOH and MCOH for Case 10 (25 Spectral Averages) .............................................................................................................................. 67 Table 22: 10 Greatest Differences between MCCOH and MCOH for Case 12 (100 Spectral Averages) .............................................................................................................................. 67 Table 23: Similarities Highlighted Between the 638 Hz and 120 Hz MCCOH Improvements .. 67 viii Nomenclature []-1 Matrix operation denoting inverse []+1 Matrix operation denoting pseudo inverse []T Matrix operation denoting transpose m Mass [M] or M Mass matrix [K] or K Stiffness matrix [C] or C Viscous damping matrix, first order coefficient [H] or H Frequency response function matrix Hij Entry row I, column j of frequency response matrix ω Rotational velocity, radians/second [I] or I Identity matrix X(ω) Displacement response in the frequency domain F(ω) Input force in the frequency domain GFF(ω) or GFF Autopower spectrum of force in the frequency domain GXX(ω) or GXX Autopower spectrum of displacement in the frequency domain GXF(ω) or GXF Crosspower spectrum of force and displacement in the frequency domain ix Abbreviations SDOF Single Degree of Freedom (system) MDOF Multiple Degree of Freedom (system) DOF(s) Degree(s) of Freedom DSP Digital Signal Processing COHp Ordinary Coherence at output DOF p CCOHp+r Combined Coherence for output DOFs p and r FRF frequency response function MCOHp Multiple Coherence at output DOF p MCCOHp+r Multiple Combined Coherence for output DOFs p and r MIMO Multiple Input Multiple Output SIMO Single Input Multiple Output SISO Single Input Single Output MISO Multiple Input Single Output x Introduction and Background Definition and Application of Combined Coherence and Multiple Combined Coherence The goal of this study is to further evaluate a frequency domain method for detecting nonlinearities using measurements already made during a typical modal test. Drops in ordinary coherence and multiple coherence are caused by one of two factors: nonlinearities and digital signal processing (DSP) errors. If a coherence function could be formulated that was unaffected by nonlinearities, but still was affected by digital signal processing (DSP) errors or vice versa, then one could determine whether a drop in ordinary coherence was due to a nonlinearity or a DSP error by comparing the two coherence functions. The coherence function proposed for accomplishing this is Combined Coherence (CCOH). Previous work was done by Tom Roscher with theoretical data generated from a lumped parameter (M, K, C) model [1, 2]. This paper expands on that work in two ways: 1) By applying the formulation to a “real-world” test structure. 2) By expanding the theoretical background to enable the use of post-processed data for multiple input testing with Multiple Combined Coherence (MCCOH) Combined coherence is a method of detecting the presence of nonlinearities between two given degrees of freedom (DOFs). To further identify the nonlinearity type and parameters values, other analysis methods must be used. The advantage of combined coherence is that it gives a quick and efficient way to spatially locate nonlinearities using data already taken during a typical modal test. No special test setup or data collection are required. 1 The type of excitation has a great effect on whether nonlinear effects are masked or whether they are highlighted. [3-5]. It also affects which type of analysis one should consider when doing linear and nonlinear analysis. For example, nonlinearities tend to show specific types of distortions after the Hilbert transform is applied with stepped-sine force input. However, random excitation tends to mask these effects and even when FRF’s show some distortion due to nonlinearities, the Hilbert transform will not be seriously distorted [3]. Table 1 is given below which characterizes many types of excitation and their tendency to reduce or highlight distortion due to nonlinearities [3-5]. Because the most common modal parameter estimation theories in use are based on the assumption of linearity, random excitation is most commonly used in actual testing situations. Because of this, any practical method for detecting nonlinearities using data already measured must work with random excitation. For this reason, pure random excitation has been used for all analyses involving combined coherence. Table 1: Various Excitation Types and Their Ability to Mask or Highlight Nonlinear Effects in FRF’s Excitation Type Nonlinear Distortions Remain? Swept Sine Yes, Best at Highlighting Distortions Steady State Sine Yes Periodic Chirp Yes Impact Yes Pseudo Random Yes Periodic Random No Burst Random No Pure Random No, Best at Removing Nonlinear Distortions 2 Current Methods of Detecting and Identifying Nonlinearities There are many methods used today to detect and identify nonlinearities. Most methods tend to focus on parameter identification, and in some way assume a model for the nonlinearities. Many methods also rely on data that is not readily available in experimental applications. A sample of the a few methods in use today is given below to give a flavor of the current analysis procedures. Piecewise analysis involves looking at measurement data under various conditions (forcing levels, time periods) and making comparisons to see if the data is consistent. One of the easiest nonlinear detection methods is simply to compare the FRF’s at various force levels throughout the range the structure would experience in actual use checking for differences. Another way would be to break a transient time history up into two or three sections, calculate FRF’s from each time block and compare the results to see if there are frequency or amplitude variations between them [6]. Bendat, Coppolino and Palo developed the Reverse MI/SO technique for identifying nonlinear parameters of SDOF systems based on the conversion of nonlinear SISO models with or without feedback into linear MISO models without feedback [7, 8]. It involves looking at the force as the output and the acceleration response as the input and then connecting them by various linear and nonlinear paths. It tends to be applied to systems with few degrees of freedom where the gross behavior of the structure is nonlinear. Examples involving an automotive shock and the stability of an ocean barge are given [7]. 3 Storer and Tomlinson present a method of detecting the presence of nonlinearities using a sine wave input and higher order FRF’s using a NARMAX model [9]. They comment on the limitations of higher order FRF’s using random excitation and the limited success in their application to real world structures. They present a nonlinear beam experiment in which they are able to determine the frequencies at which the nonlinearities act by measuring higher order FRF’s up to order three. Characterization of these nonlinearities still relies on an a priori model. Chen and Tomlinson proposed a method for identifying nonlinear parameters using a time series model based on acceleration, velocity and displacement (AVD) data [10]. This data need not be independently measured. The acceleration data may simply be integrated to get the needed velocity and displacement data. They demonstrate that using AVD data was more accurate than simply using displacement or acceleration data. The AVD model results of an experiment involving a pendulum with nonlinear supports are compared to NARMAX model results of the same experiment. The argument is then made for the superiority of the AVD model especially for higher (3rd) order FRF’s. Ghanem and Romeo have proposed a method for identifying parameters of an a priori known nonlinear system [11]. When the type of nonlinearity is unknown, a method of selecting the type from various classes of nonlinearities is presented. The method makes use of wavelets and their ability to focus on specific time/frequency bands to identify systems with nonlinearities whose effects would be averaged out using other techniques. Theoretical noise free examples are given. 4 Worden and Tomlinson present a method of detecting and locating nonlinearites that makes use of the Hilbert transform in identifying the part of the FRF that is due to artificial non-causality caused by nonlinearities [3]. They define the non-causal power ratio (NPR) as the ratio of the power of the non-causal part of the impulse response to the power of the total impulse response. Calculating the NPR for each impulse response in the FRF matrix can give an indication of where the nonlinearites are spatially located. A method for detecting nonlinearities using only experimental data was developed by Chong and Imregun using reciprocal modal vectors [12]. This involves going through the modal parameter identification process, selecting the DOFs of interest and comparing the deviation of reciprocal modal vectors from the unit matrix to see which modes are affected by nonlinearities. This method, relying on the results of a mode fit seems to be contaminated by errors resulting from closely spaced modes. Theoretical examples with and without noise are given. Definition and Application of Ordinary and Multiple Coherence Ordinary coherence (COH, Eqn. 1) is a statistical indicator of whether a given input and a given output are correlated (i.e., linearly related) [5, 13-15]. It is bounded by zero and one (Eqn. 2) with one indicating a perfectly linear relationship between the input and output and zero indicating that the input and output are completely uncorrelated. Being a statistical indicator of linearity, the more averages are taken, the more confidence one can have that a high value is a true indication of a linear relationship between the input and output. A discussion of how statistical confidence levels vary with the number of averages is given in Appendix 1. Coherence is not an indicator of causality but of correlation. Because of this, a drop in 5 coherence does not necessarily mean that the response is not being caused by the input (though that may be the case). A few of the more common errors causing drops in coherence are listed below [13-15]. One can see that the errors may be broadly broken down into two categories, DSP errors (1-3) and errors due to nonlinearities (4-8). All errors causing drops in coherence fall into one of these two categories. 1) Leakage 2) Aliasing 3) Measurement noise 4) Friction Forces 5) Saturation, dead-zone 6) Unmeasured inputs 7) Harmonic distortion 8) Hardening or softening stiffness/damping COH p (ω ) = GXF ps (ω ) 2 (1) GXX pp (ω )GFFss (ω ) 0 ≤ COH p (ω ) ≤ 1 (2) Ordinary coherence indicates whether a single input is linearly related to a single output. Often though, multiple inputs are used during testing so that energy can be more evenly distributed throughout a structure and so that repeated roots can be separated. In the case of two uncorrelated inputs both of which have equal output spectra at a given DOF, the ordinary coherence would have a value of 0.5 for both input-output combinations. Multiple coherence (MCOH) was developed to take all inputs into account (Eqn. 3). It is a statistical indicator of whether all inputs taken together are linearly related to a given output [5, 13-15]. Because of 6 this, it can be used in MIMO testing scenarios similarly to the way ordinary coherence is applied and interpreted in SISO or SIMO testing scenarios. MCOH is bounded by zero and one. A value of one indicates that all the inputs together are linearly related to the output, while a value of zero indicates that none of the inputs are correlated with the output (Eqn. 4). Ni Ni MCOH p (ω ) = ∑∑ H ps (ω )GFFst (ω )H *pt (ω ) s =1 t =1 (3) GXX pp (ω ) 0 ≤ MCOH p (ω ) ≤ 1 (4) Combined Coherence Equation Development Theoretical Basis for Equation If one considers the lumped mass system in Figure 1 below, there are both internal and external forces acting on it. The internal forces, due to the springs and dampers may or may not be nonlinear. If there was a nonlinear spring between two specific DOFs and the motion of those DOFs was summed together, the nonlinear internal forces would then cancel out leaving only the remaining external force terms. F1 X1 K01 F2 C01 X2 K12 M1 F3 K23 X3 K30 M3 M2 C12 C23 C30 Figure 1: Three DOF Lumped Mass System The equations of motion for DOFs 1 and 2 are given below. & & & m1 & x& 1 = −(k 01 + k12 + k13 )x1 + k12 x 2 + k13 x3 − (c01 + c12 + c13 )x1 + c12 x 2 + c13 x3 + F1 (5) & & & m2 & x& 2 = −(k12 + k13 )x 2 + k12 x1 + k 23 x3 − (c12 + c13 )x 2 + c12 x1 + c 23 x3 + F2 (6) 7 If one were to sum the motion of the two DOFs together, one arrives at a new equation. 1 [− (k 01 + k12 + k13 )x1 + k12 x2 + k13 x3 ] + 1 [− (k12 + k13 )x2 + k12 x1 + k 23 x3 ] m1 m2 & & & x& 1 + x2 = (7) + 1 [− (c01 + c12 + c13 )x&1 + c12 x&2 + c13 x&3 ] + 1 [− (c12 + c13 )x&2 + c12 x&1 + c23 x&3 ] + F1 + F2 m1 m2 m1 m2 If one makes the assumption that the masses of the two DOFs are equal, then the information about the internal forces drops out. & & & x& 1 + x2 = 1 [− (k01 + k13 )x1 + k13 x3 − (k13 )x2 + k23 x3 ] + 1 [− (c01 + c13 )x&1 + c13 x&3 − (c13 )x&2 + c23 x&3 ] + F1 + F2 m m m (8) Thus, if a coherence function is derived, based on the sum of the motions of two DOFs containing a nonlinearity between them, the drop in coherence due to that nonlinearity will go away. However, drops in coherence due to digital signal processing errors such as leakage and noise will remain [1, 2]. If the masses are not equal, scaling the motion of the two DOFs will be necessary to completely cancel out the effect of internal forces. Mathematical Development of Combined Coherence The standard ordinary coherence equation is given below for input point q and output point p [5, 13-15]. ∑ (X Nave COHpq = GXF pq 2 GFFqq GXX pp = 1 Nave ∑ (F q p Fq* * q F 1 )∑ (X Nave Fq p * p 1 Nave )∑ (X ) * p X 1 (9) ) Since the combined coherence function is based on the sum of the motion between two DOFs, Xp+Xr is substituted for Xp. CCOH ( pr )q ∑ ((X + X )F )∑ ((X + X ) F ) = ∑ (F F )∑ ((X + X )(X + X ) ) r q p * q * * q r p q * r p r p 8 (10) CCOH ( pr )q = ∑ (X ∑ (F q Fq* r Fq* + X p Fq* )∑ (X r )∑ (X Fq + X p Fq r ) * X r* + X r X *p + X p X r* + X p X *p (11) ) Expanding and consolidating terms, one is left with the equation below [1, 2]. CCOH ( pr )q = GXF pq + GXFrq 2 (12) GFFqq (GXX rr + GXX rp + GXX pr + GXX pp ) Mathematical Development of Multiple Combined Coherence The standard equation for Multiple Coherence is given by the equation below [5, 13-15]. Ni Ni MCOH p = ∑∑ H ps GFFst H *pt s =1 t =1 (13) GXX pp Looking at H for a 2 input, 3 output system, GFF ss GFF = GFF ts [GFF ]−1 GFF st GFF tt GFF tt GFF GFF − GFF GFF ss tt st ts = GFF ts GFF ss GFF tt − GFF st GFF ts GXF 1 s GXF = GXF 2 s GXF 3 s GXF 1t GXF 2 t GXF 3 t H 1 = [GXF ][GFF ] H 1t = − (14) −1 H 1s = H 2 s H 3 s GFF st GFF ss GFF tt − GFF st GFF ts GFF tt GFF ss GFF tt − GFF st GFF ts (15) (16) H 1t H 2 t H 3 t (17) GFF st GXF 1 s GFF ss GXF 1t + GFF ss GFF tt − GFF st GFF ts GFF ss GFF tt − GFF st GFF ts 9 (18) H 1s = GFF tt GXF 1 s GFF ts GXF 1t − GFF ss GFF tt − GFF st GFF ts GFF ss GFF tt − GFF st GFF ts (19) Since MCCOH is based on the summation of two degrees of freedom, let X1=Xp+Xr in H1s. H 1s = GFFtt GXF1s GFFts GXF1t − GFFss GFFtt − GFFst GFFts GFFss GFFtt − GFFst GFFts (20a) ∑ (F F )∑ (X F ) ∑ (F F )∑ (X F ) − )∑ (F F ) − ∑ (F F )∑ (F F ) ∑ (F F )∑ (F F ) − ∑ (F F )∑ (F F ) (20b) ∑ (F F )∑ ((X + X )F ) ∑ (F F )∑ ((X + X )F ) = − ∑ (F F )∑ (F F ) − ∑ (F F )∑ (F F ) ∑ (F F )∑ (F F ) − ∑ (F F )∑ (F F ) (20c) * = t ∑ (F F s * s t t * * s 1 * t s t t * t s = t * s p s GFFtt (GXF ps + GXFrs ) GFFss GFFtt − GFFst GFFts s * s t t − t * s s * s * 1 t * t s * s t * t * s * s r * t * s * t t * s * p r * t t t GFFts (GXF pt + GXFrt ) GFFss GFFtt − GFFst GFFts = H ps + H rs t * s t t * s (20d) (20e) Also, let X1=Xp+Xr in GXX11. GXX 11 = ∑ (X X ) 1 * 1 (21a) [ = ∑ (X p + X r )(X p + X r ) * [ ( ] (21b) )] (21c) = ∑ (X p + X r ) X *p + X r* = ∑ X p X *p + ∑ X p X r* + ∑ X r X *p + ∑ X r X r* (21d) = GXX pp + GXX pr + GXX rp + GXX rr (21e) Thus, Ni Ni MCCOH p + r = ∑∑ s =1 t =1 (H (GXX + H rs )GFFst (H pt + H rt ) * ps pp + GXX pr + GXX rp + GXX rr ) 10 (22) Theoretical Example Roscher gave several theoretical examples of the application of combined coherence to a test structure [1]. One is included here to show how combined coherence can be applied to a theoretical lumped mass system. M4 K3 C3 K2 M3 C2 K1 K2 C2 C1 K1 M2 C1 K1 C1 M1 K0 C0 Figure 2: Theoretical Lumped Parameter (M, K, C) System used in example A hardening stiffness is applied between DOFs 1 and 3 of the form k(x+εx3). The stiffness, damping, mass and ε values are given in Table 2 below. Since the nonlinear stiffness is between 11 DOFs 2 and 3, one would expect combined coherence to show improvement when summing those two DOFs, while not improving for other DOF combinations. Table 2: Force, Mass, Stiffness and Damping Values of Theoretical 4 DOF System Ground Mass 1 12 kg Mass 2 7 kg Mass 1: 12 kg K10=22000 N/m C10=6 N-s/m Random Input: F1=30 N rms Mass 2: 7 kg Mass 3: 9 kg Mass 4: 14 kg K12=19000 N/m C12=8 N-s/m No Input Force K13=19000 N/m C13=8 N-s/m K23_lin=24000 N/m ε = 50000 K23=K23_lin(1+ε∆x2) C23=9 N-s/m No Input Force K14=24000 N/m C14=7 N-s/m K24=20000 N/m C24=4 N-s/m Mass 3 9 kg Mass 4 14 kg K34=20000 N/m C34=5 N-s/m No Input Force Comparing the frequency response and ordinary coherence values for the linear and nonlinear systems, one can see that there are several significant drops in coherence for the nonlinear system (Figure 3 through Figure 5). For all DOFs, there is a significant drop due to leakage at the 3.5 Hz resonant peak. For DOFs 2 and 3, which are directly affected by the nonlinearity, there are other drops in coherence at the 13 Hz antiresonance, from 15-19 Hz and at high frequency. There is also a shift in the resonant peak at 17 Hz (Figure 3 and Figure 4). For DOF 4, which is not direcly affected by a nonlinearity, there is a slight drop in coherence at around 17 Hz and some shifting in the resonant peak there as well (Figure 4). 12 H(2,1) = X/F, normally distributed random excitation 50% overlapping blocks, 79 averages & Hanning window applied -2 10 H1 lin(2,1) H1 nonlin(2,1) -4 10 F/ X = H -6 10 -8 10 0 5 10 15 20 Frequency (hz) 25 30 35 40 1 e c n er e h o C f o e d ut i n g a M COH lin(2,1) COH nonlin(2,1) 0.8 0.6 0.4 0.2 0 0 5 10 15 20 Frequency (hz) 25 30 35 40 Figure 3: Comparison of FRF and COH for Linear and Nonlinear Theoretical Systems at a DOF directly affected by the nonlinearity H(3,1) = X/F, normally distributed random excitation 50% overlapping blocks, 79 averages & Hanning window applied -2 10 H1 lin(3,1) H1 nonlin(3,1) -4 10 F/ X = H -6 10 -8 10 0 5 10 15 20 Frequency (hz) 25 30 35 40 1 e c n er e h o C f o e d ut i n g a M 0.8 COH lin(3,1) COH nonlin(3,1) 0.6 0.4 0.2 0 0 5 10 15 20 Frequency (hz) 25 30 35 40 Figure 4: Comparison of FRF and COH for Linear and Nonlinear Theoretical Systems at a DOF directly affected by the nonlinearity 13 H(4,1) = X/F, normally distributed random excitation 50% overlapping blocks, 79 averages & Hanning window applied -2 10 H1 lin(4,1) H1 nonlin(4,1) -4 10 F/ X = H -6 10 -8 10 0 5 10 15 20 Frequency (hz) 25 30 35 40 1 e c n er e h o C f o e d ut i n g a M 0.8 COH lin(4,1) COH nonlin(4,1) 0.6 0.4 0.2 0 0 5 10 15 20 Frequency (hz) 25 30 35 40 Figure 5: Comparison of FRF and OCOH Magnitudes for Linear and Nonlinear Theoretical System at a DOF not directly affected by the nonlinearity As can be seen below in Figure 6, when adding the motions of DOFs 2 and 3, the combined coherence plot shows significant improvement at 13 Hz, 17 Hz and at high frequency where the nonlinear effect was quite noticeable. This indicates that a nonlinearity is acting between DOFs 2 and 3. Looking at Figure 7 below, while combined coherence may improve compared to the ordinary coherence of one of the DOFs, it never improves over the ordinary coherence of both of the DOFs. This leads to the expected conclusion that a nonlinearity is not located between DOFs 3 and 4. 14 COH 2/1, COH 3/1, CCOH(2+3)/1, F=30N 1 0.9 ) 1, 2( H O C 0.8 0.7 0.6 0.5 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 0 5 10 15 20 Frequency (f, Hz) 25 30 35 40 1 0.9 ) 1, 3( H O C 0.8 0.7 0.6 0.5 1 0.9 ) 1, 3 + 2( H O C C 0.8 0.7 0.6 0.5 Figure 6: OCOH and CCOH for a Theoretical Hardening Stiffness Case COH 3/1, COH 4/1, CCOH(3+4)/1, F=30N 1 ) 1, 3( H O C 0.9 0.8 0.7 0.6 0.5 ) 1, 4( H O C 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 1 0.9 0.8 0.7 0.6 0.5 ) 1, 4 + 3( H O C C 1 0.9 0.8 0.7 0.6 0.5 Frequency (f, Hz) Figure 7: OCOH vs. CCOH for a Theoretical Hardening Stiffness Example 15 Applying the Theory to a Real Structure Test Subject The test structure used is a welded metal frame of 2”x6”x.25” steel tubing. Attached to this frame was another welded structure made from 2”x2”x.125” steel tubing. These were connected at 4 discrete points, giving various options for linear/nonlinear connections. The test frames are shown below in Figure 8. Figure 8: H-Frame and Square Frame Test Structures Two shakers were used as inputs in a MIMO type test. Each shaker was connected to the HFrame in a skew direction in order to get energy into the structure in all three principal directions. The shakers were 45o skew to all three principal directions at Points 101 and 102, 16 while shaker #1 was only about 20o skew to the principal x direction at Point 100, due to space/angle constraints. The setup of the shakers is shown below in Figure 9 through Figure 11. Shaker #1 at Point 100 Shaker #1 at Point 100 - Closeup Shaker #1 at Point 102 Shaker #1 at Point 102 - Closeup Shaker #2 at Point 101 Shaker #2 at Point 101 - Closeup Figure 9: Pictures of Shaker Setup showing skew input directions 17 The spatial distribution of the accelerometers varied depending on the test conducted. Some tests had a “spread out” spatial distribution, giving an overall view of the structure and allowing for a determination of structural mode shapes (Figure 10 below). Other tests used a more “concentrated” spatial distribution focusing on the interactions across two specific connection points between the structures (Figure 11 below). Figure 10: Accelerometer and force input locations using a spread out spatial distribution 18 Figure 11: Accelerometer locations using a more concentrated spatial distribution 19 Case 1: 4 Nominally Linear Connection Points Between Structures The first experimental case presented is that of the square frame bolted to the H-Frame at each corner with a wood spacer separating the two. This is meant to simulate a hard “linear” connection as much as possible, without welding the two structures together (see Figure 12 and Figure 13 below). Since the connections are linear, there should not be large improvements in multiple combined coherence (MCCOH) compared to multiple coherence (MCOH) across the connection points. Random excitation was used with 150 spectral averages in order to get a good estimate of the FRF’s. For a complete description of the test setup, see Table 3 below. Table 3: Test Setup for Case 1 Test Setup Description Accelerometer Concentrated, Excitation Type Random Distribution Figure 11 Input Voltage 3V Peak Force 10 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 150 # of Cyclic Averages 0 Attachment Method: Bolted connection with 2x4 wood spacer at each corner of the square frame (Figure 12 and Figure 13). 20 Figure 12: Layout of connections between the H-frame and the square frame for Cases 1 and 2 Figure 13: Picture and schematic of the Bolted Joint for Cases 1 and 2 In comparing combined coherence to multiple coherence, it is useful to compare points located next to each other in the same direction to determine if a nonlinearity is being excited between 21 them (e.g., comparing 44x to 48x). Table 4 below shows the DOF combinations with at least a 10% improvement in MCCOH over MCOH as well as whether the improvement occurred at a resonance or antiresonance. As can be seen from Figure 14 and Figure 15, the largest improvements all came at resonant peaks. This led to the question of whether leakage was contaminating the results at all, contrary to earlier theoretical studies which showed that leakage did not affect combined coherence [1, 2]. Table 4: Table of Improvements in Combined Coherence for Case 1 freq pt1 dir1 pt2 dir2 (Hz) MCCOH MCOH1 MCOH2 13 13 52 51 51 5 13 13 13 54 54 53 53 43 43 +X +Z -Y -X -X +Y +Y +Z +Z +Y +Z -X +Y +Y +Z 15 15 51 46 44 52 54 54 54 53 53 47 45 47 47 -X +Z +Y +X +X -Y +Y +Z +Z +Y +Z +X +Y +Y +Z 30.5 38.5 54.25 54.25 54.5 54.25 54.25 38.25 49.25 11 30.75 11 11 38.25 49.25 0.9823 0.9794 0.954 0.7375 0.9554 0.9925 0.9887 0.9642 0.9983 0.9834 0.9768 0.9552 0.9961 0.898 0.9836 0.8724 0.8663 0.8227 0.6301 0.6864 0.8366 0.838 0.7113 0.7997 0.8388 0.8803 0.8463 0.8736 0.7367 0.8395 22 0.8897 0.8871 0.766 0.5481 0.483 0.8227 0.8837 0.8153 0.835 0.8736 0.8814 0.7797 0.8405 0.7186 0.8713 %difference 10.4063 10.4094 15.9539 17.0329 39.1917 18.6288 11.8851 18.2648 19.559 12.57 10.8262 12.8657 14.0137 21.8927 12.8882 Improvement at (anti)resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance resonance FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 )f bl / n(i e d ut i n g a M F R F 10 10 FRF FRF FRF FRF -1 -2 -3 45 50 55 60 65 Frequency (Hz) 70 75 80 85 Multiple Coherence for Points 5 and 52 directions +Y and -Y. 1 0.8 MCOH 5+Y MCOH 52-Y 0.6 H O C M 5 +Y,100 +X 52 -Y,100 +X 5 +Y,101 +Z 52 -Y,101 +Z 0.4 0.2 0 0 100 200 300 500 600 700 800 Multiple Combined Coherence for Point 5 and 52 directions +Y and -Y. 1 0.8 H O C C M 400 Combined Coherence 5 Frequencies with 5% Improvement above 10 Hz. 0.6 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 14: FRF and coherence functions showing the 54.25 Hz resonance peak at which combined coherence improves over multiple coherence 23 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 10 b n e d u n g a M F R F 10 10 -1 FRF FRF FRF FRF 47 +Y,100 +X 43 +Y,100 +X 47 +Y,101 +Z 43 +Y,101 +Z -2 -3 -4 30 32 34 36 38 40 42 44 46 48 50 Frequency (Hz) Multiple Coherence for Points 43 and 47 directions +Y and +Y. 1 0.8 MCOH 43+Y MCOH 47+Y 0.6 H O C M 0.4 0.2 0 0 100 200 300 500 600 700 800 Multiple Combined Coherence for Point 43 and 47 directions +Y and +Y. 1 0.8 H O C C M 400 Combined Coherence 2 Frequencies with 5% Improvement above 10 Hz. 0.6 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 15: FRF and coherence functions showing the 38.25 Hz resonance peak at which combined coherence improves over multiple coherence 24 Case 2: Effect of Cyclic Averaging on Nominally Linear Case 1 Since the majority of improvements occurred at resonant peaks for the previous case, a test was run to see what would happen if steps were taken to reduce the amount of leakage error. For this purpose, 10 cyclic averages were used (Table 5). Table 5: Test Setup for Case 2 Test Setup Description Accelerometer Concentrated, Excitation Type Random Distribution Figure 11 Input Voltage 3V Peak Force 10 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 150 # of Cyclic Averages 10 Attachment Method: Bolted connection with 2x4 wood spacer at each corner of the square frame (Figure 12 and Figure 13).. As was expected, cyclic averaging vastly improved the multiple coherence function almost universally. Of noted interest is that MCCOH no longer improved over MCOH at the aforementioned resonant peaks, indicating that either both leakage and a nonlinearity had affected the FRF estimate in Case 1, or that an improvement in combined coherence can be an indication of a DSP error and not just a nonlinearity. 25 10 Overlay Plot of FRF's and Multiple Coherence for DOF 5 +Y for Cases with and without Cyclic Averaging 0 No Cyclic Averaging: 5 +Y Cyclic Averaging: 5 +Y e d ut i n g a M F R F 10 -5 0 50 100 150 1 0.8 e d ut i n g a M H O C M No Cyclic Averaging: 5 +Y Cyclic Averaging: 5 +Y 0.6 0.4 0.2 0 0 50 100 150 Frequency (Hz) Figure 16: Overlay of FRF and Coherence Plots Showing the Effect of Cyclic Averaging on the Nominally Linear System 10 Overlay Plot of FRF's and Multiple Coherence for DOF 47 +Y for Cases with and without Cyclic Averaging 0 No Cyclic Averaging: 47 +Y Cyclic Averaging: 47 +Y e d ut i n g a M F R F 10 -5 0 10 20 30 40 50 60 70 80 1 0.8 e d ut i n g a M H O C M No Cyclic Averaging: 47 +Y Cyclic Averaging: 47 +Y 0.6 0.4 0.2 0 0 10 20 30 40 50 60 70 80 Frequency (Hz) Figure 17: Overlay of FRF and Coherence Plots Showing the Effect of Cyclic Averaging on the Nominally Linear System 26 After cyclic averaging to remove leakage errors, MCCOH still improved over MCOH at certain frequencies, though the improvements were much smaller in magnitude. The locations of these improvements in the FRF’s changed in nature from resonances to antiresonances (Figure 18 and Figure 19). Table 6 below lists DOFs next to each other which improved 2% or more. The absence of any improvement in combined coherence at the same frequencies as before for the same setup makes the insensitivity of combined coherence to leakage effects as noted in earlier theoretical studies questionable for distributed mass systems [1, 2]. Table 6: Table of improvements of 2% or more in Combined Coherence for Case 2 for adjacent points pt1 dir1 pt2 dir2 freq (Hz) MCCOH MCOH1 MCOH2 % Difference (anti)resonance 3 -Y 5 +Y 102.5 0.9920 0.6141 0.9683 2.4432 antiresonance 13 +Y 15 -Y 282.25 0.9922 0.9276 0.7788 6.9654 antiresonance 52 -Y 51 +Y 21.5 0.9934 0.8719 0.9714 2.2585 antiresonance 52 -Y 51 +Y 25.25 0.9941 0.6735 0.8594 15.6682 5 +Y 52 -Y 25.25 0.9975 0.9435 0.6735 5.7227 antiresonance 13 +X 54 -X 177.75 0.9896 0.9651 0.9302 2.5389 antiresonance 13 +Y 54 +Y 23.25 0.9712 0.9474 0.9461 2.5158 antiresonance 13 +Y 54 +Y 23.5 0.9762 0.9505 0.9555 2.1700 54 -X 53 -X 177.75 0.9577 0.9302 0.8926 2.9643 antiresonance 15 -X 54 -X 24.75 0.9611 0.9363 0.9282 2.6486 15 -X 54 -X 25 0.9514 0.9121 0.9228 3.1018 antiresonance 15 -X 54 -X 25.5 0.9050 0.7546 0.8788 2.9714 27 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 )f bl / n(i e d ut i n g a M F R F 10 FRF FRF FRF FRF -1 54 -X,100 +X 13 +X,100 +X 54 -X,101 +Z 13 +X,101 +Z -2 165 170 175 180 185 Frequency (Hz) 190 195 Multiple Coherence for Points 13 and 54 directions +X and -X. 1 0.8 MCOH 13+X MCOH 54-X 0.6 H O C M 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Combined Coherence for Point 13 and 54 directions +X and -X. 1 Combined Coherence 1 Frequencies with 1.5% Improvement above 10 Hz. 0.8 0.6 H O C C 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 18: Case 2 FRF and coherence functions showing the 177.75 Hz antiresonance valley at which combined coherence improves over multiple coherence 28 FRF FRF FRF FRF 10 -2 10 )f bl / n(i e d ut i n g a M F R F )f bl / n(i e d ut i n g a M F R F 10 10 51 +Y,100 +X 52 -Y,100 +X 51 +Y,101 +Z 52 -Y,101 +Z -1 -2 -3 10 15 20 25 30 Frequency (Hz) 35 40 -3 720 730 740 750 760 Frequency (Hz) 770 780 790 Multiple Coherence for Points 51 and 52 directions +Y and -Y. 1 MCOH 51+Y MCOH 52-Y 0.8 0.6 H O C M 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Combined Coherence for Point 51 and 52 directions +Y and -Y. 1 Combined Coherence 9 Frequencies with 1.5% Improvement above 10 Hz. 0.8 0.6 H O C C 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 19: Case 2 FRF and coherence functions showing the 21.5, 25.25 and 772 Hz antiresonance valleys at which combined coherence improves over multiple coherence 29 Cases 3-5: Effect of Varying the Force Input with Four Nominally Nonlinear Connection Points Between Structures In order to create nonlinear connections between the Square Frame and H-Frame, some vehicle body mounts sandwiched between a foam with memory were inserted at the four connection points where the wooden blocks had been (Figure 20 and Figure 21). The force was then varied to see at what forcing level the nonlinearities between the structures were most excited. Table 7: Test Setup for Cases 3-5 Test Setup Description Accelerometer Spread Out, Excitation Type Random Distribution Figure 10 Case 3 Input Voltage 1 V Case 3 Peak Force 4 lbf Case 4 Input Voltage 3 V Case 4 Peak Force 10 lbf Case 5 Input Voltage 5 V Case 5 Peak Force 18 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 150 # of Cyclic Averages 0 Attachment Method: Four Vehicle Body Mounts at each corner of the square frame. Two bolts barely hand tightened. Two bolts removed altogether. See Figure 20 and Figure 21. Figure 20: Layout of connections between the H-frame and the square frame for Cases 3-8 30 Figure 21: Schematics of the two connection types between the structures for Cases 3-8 As can be seen in Table 9 through Table 11, when the force level was varied, the number and nature of improvements in combined coherence varied. Figure 25 through Figure 27 below give examples of the FRF and COH plots at the resonances and antiresonances for which MCCOH improved over MCOH. At low forcing levels (Case 3), combined coherence changed more in the 24.25 Hz range at some very small peaks/valleys (Figure 25). At this frequency the DOF combinations that showed the most improvement in MCCOH over MCOH were between DOFs on the lower structure and between DOFs spanning the nonlinear connection (Table 9 and Figure 22). At high forcing levels, there were still changes a couple resonant peaks with MCCOH improvements, but the dominant changes occurred at a very wide frequency range around the 395 Hz antiresonance on the upper frame (Table 11, Figure 24 and Figure 27). This improvement did not appear at the lower force levels. For the 10 and 18 lbf forcing levels, at the 53 Hz resonance peak combined coherence improved throughout both structures (Figure 23 and Figure 26). Similar to the 25 Hz improvements, these DOF combinations were spanning the nonlinear gap, within the square frame and within the H-frame. Table 8 below summarizes these observations. 31 Table 8: Summary of Results for Cases 3-5 Frequency Range Resonant Peak/ Antiresonance Excited More at Low, Medium or High Forcing Levels DOF Combinations with MCCOH Improvements Within HFrame Within Square Frame Across Connection Points 24.25 Hz Resonant Peak Low to Medium Yes Yes Yes 53.5 Hz Resonant Peak Medium to High Yes Yes Yes ~385-400 Hz Antiresonance Medium to High No Yes No One thing to note is that for Case 3 with the 4 lb/shaker input, the largest changes for adjacent points between multiple combined coherence and multiple coherence occurred when combining Points 21 and 12 as well as Points 21 and 13 (Figure 22). At 400%, the changes across these DOFs dwarfed other changes. It is interesting to note that the shortest and most direct path between the two DOF combinations is perfectly linear. Also, oftentimes improvements in MCCOH over MCOH are present not only between various DOFs in the same direction (e.g., 44X and 45X) but also between DOFs in orthogonal directions (e.g., 42Y and 44X). In fact, Figure 24 shows that cross directional improvements are sometimes more common than improvements in the same direction. 32 Table 9: Case 3 improvements of 20% or more in Combined Coherence for Case 3 for adjacent points in the same direction pt1 dir1 pt2 dir2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference (anti)resonance not at 2 +Z 3 +Z 111.75 0.7952 0.6431 0.6614 20.2148 resonance 21 -Y 13 +Y 21.5 0.9729 0.0761 0.798 21.9237 21 -Y 13 +Y 24 0.951 0.2195 0.2365 302.0953 21 -Y 13 +Y 24.25 0.956 0.1791 0.1869 411.3762 21 -Y 13 +Y 24.5 0.9575 0.7518 0.7063 27.3636 small peaks 21 -Y 13 +Y 25 0.9643 0.6023 0.5271 60.1085 21 -Y 13 +Y 25.25 0.9604 0.4645 0.4174 106.7339 21 -Y 13 +Y 26.75 0.9537 0.6867 0.7139 33.5941 21 -Y 13 +Y 27 0.9557 0.6055 0.6553 45.8423 21 -Y 13 +Y 100.75 0.9267 0.6184 0.4771 49.8671 antiresonance 21 -Y 13 +Y 101 0.9326 0.7702 0.586 21.093 21 -X 12 +X 47.5 0.6358 0.4788 0.5266 20.735 antiresonance 21 -X 12 +X 48.5 0.4374 0.2988 0.357 22.508 21 -Y 12 +Y 21.5 0.9763 0.0761 0.7661 27.4387 21 -Y 12 +Y 24 0.9706 0.2195 0.2167 342.1419 21 -Y 12 +Y 24.25 0.9688 0.1791 0.1783 440.9924 21 -Y 12 +Y 24.5 0.9706 0.7518 0.7106 29.1004 small peaks 21 -Y 12 +Y 25 0.972 0.6023 0.5149 61.3782 21 -Y 12 +Y 25.25 0.9653 0.4645 0.4061 107.7982 21 -Y 12 +Y 26.75 0.9763 0.6867 0.7641 27.7582 21 -Y 12 +Y 27 0.9787 0.6055 0.7203 35.872 antiresonance/ 21 -Y 12 +Y 100.75 0.8463 0.6184 0.3673 36.8678 peak for least excited FRF resonance 45 +Z 44 +Z 53.5 0.9477 0.6981 0.6596 35.7501 12 +Y 42 +Y 24 0.3484 0.2167 0.2494 39.7216 small peaks 12 +Y 42 +Y 24.25 0.2937 0.1783 0.1942 51.265 12 +Y 42 +Y 25.25 0.5398 0.4061 0.393 32.924 12 +Z 42 +Z 24 0.9086 0.2901 0.3573 154.3206 12 +Z 42 +Z 24.25 0.9088 0.3965 0.2953 129.2423 small peaks 12 +Z 42 +Z 25 0.95 0.7556 0.5166 25.7258 12 +Z 42 +Z 25.25 0.9123 0.6609 0.5631 38.0488 small peak 13 +Y 42 +Y 25.25 0.5085 0.4174 0.393 21.827 13 +Z 42 +Z 24 0.7111 0.4818 0.3573 47.592 small peak 13 +Z 42 +Z 24.25 0.6659 0.4688 0.2953 42.0393 3 +Z 44 +Z 25 0.7353 0.5973 0.4868 23.0976 small peak 3 +Z 44 +Z 25.25 0.6467 0.4372 0.465 39.067 4 +Z 44 +Z 24 0.6864 0.3511 0.0665 95.4814 small peak 4 +Z 44 +Z 24.25 0.7416 0.4718 0.0394 57.1842 13 +Y 45 +Y 24.25 0.241 0.1869 0.1839 28.9065 small peaks 13 +Y 45 +Y 25.25 0.5577 0.4174 0.3641 33.626 14 +Y 45 +Y 24 0.3211 0.2354 0.2414 33.0263 14 +Y 45 +Y 24.25 0.3296 0.1985 0.1839 66.0124 small peaks 14 +Y 45 +Y 25 0.7249 0.5518 0.5052 31.3702 14 +Y 45 +Y 25.25 0.6444 0.4386 0.3641 46.9463 33 Table 10: Case 4 improvements of 20% or more in Combined Coherence for Case 4 for adjacent points in the same direction (anti)resonance pt1 dir1 pt2 dir2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference 21 21 21 21 21 11 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 12 12 12 13 13 14 -Y -Y -X -Y -Y +Z +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +Z +Y +Z +Z +Z +Y +Y 13 13 12 12 12 12 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 42 42 42 42 45 45 +Y +Y +X +Y +Y +Z +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +Z +Y +Z +Z +Z +Y +Y 24 24.25 53.75 24 24.25 53.75 392.5 392.75 393 393.75 394 394.25 394.75 395 395.25 395.5 395.75 396 396.25 396.75 397 53.5 24.25 24 24.25 24.25 24.25 24.25 0.9939 0.9942 0.8059 0.9964 0.9965 0.9211 0.8623 0.8583 0.8777 0.8699 0.8591 0.8641 0.8808 0.8618 0.8342 0.8538 0.8278 0.7992 0.8131 0.809 0.8304 0.9346 0.775 0.9787 0.9771 0.8925 0.7784 0.8237 0.7672 0.687 0.6533 0.7672 0.687 0.7504 0.6961 0.689 0.7164 0.7112 0.6823 0.6944 0.7186 0.6657 0.5947 0.6063 0.5757 0.5524 0.5685 0.6171 0.6157 0.7615 0.6281 0.7642 0.76 0.7106 0.64 0.6521 34 0.7244 0.64 0.5347 0.7174 0.6281 0.5504 0.6431 0.6078 0.6772 0.6591 0.6337 0.6383 0.6468 0.6228 0.5823 0.5823 0.6167 0.6229 0.6631 0.6738 0.6621 0.6906 0.5719 0.7813 0.6502 0.6502 0.5558 0.5558 29.5463 44.7169 23.3632 29.8729 45.0495 22.7411 23.8802 24.5721 22.5216 22.3078 25.9134 24.4415 22.5627 29.4516 40.2745 40.8342 34.2281 28.2962 22.6324 20.061 25.4282 22.7206 23.4036 25.2672 28.57 25.583 21.6374 26.3235 small peak/valley resonance small peak/valley resonance antiresonance resonance small peak/valley small peak/valley small peak/valley small peak/valley small peak/valley Table 11: Case 5 improvements of 20% or more in Combined Coherence for Case 5 for adjacent points in the same direction (anti)resonance pt1 dir1 pt2 dir2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference small peak 21 -Y 13 +Y 24.25 0.9979 0.8245 0.7946 21.0337 21 -X 12 +X 53.5 0.8404 0.6745 0.6303 24.5933 resonance 21 -X 12 +X 53.75 0.8642 0.7039 0.6831 22.7749 21 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 2 -Y +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X -X 12 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 44 41 +Y +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X +X 24.25 391.5 391.75 392.75 393 393.25 393.5 393.75 394 394.25 394.5 394.75 395 395.25 395.5 395.75 396 396.25 396.5 396.75 397 53.5 0.9987 0.8481 0.8499 0.8367 0.831 0.8641 0.8498 0.8451 0.8338 0.8458 0.8218 0.7964 0.8044 0.8051 0.7866 0.8188 0.8169 0.8177 0.802 0.7823 0.7833 0.7479 0.8245 0.6905 0.7065 0.665 0.6576 0.7029 0.6792 0.6478 0.6365 0.6654 0.6028 0.5458 0.5504 0.5367 0.5354 0.5901 0.5456 0.583 0.5543 0.5127 0.5201 0.6217 35 0.7843 0.6765 0.6822 0.6648 0.6661 0.668 0.6392 0.6163 0.577 0.6054 0.6336 0.6344 0.6344 0.5851 0.5763 0.5552 0.5312 0.5612 0.5855 0.639 0.6209 0.591 21.1232 22.8245 20.2848 25.8344 24.7628 22.9372 25.1206 30.4624 31.0061 27.1037 29.7134 25.5227 26.7836 37.5845 36.4775 38.7574 49.7291 40.2581 36.9719 22.4381 26.1532 20.2894 small peak/valley antiresonance resonance Percent Difference Between MCCOH and the Maximum MCOH at f = 24.25 Hz. Plot Min/Max = -0.05% and 440.9924% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 400 350 300 250 200 150 100 50 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y 43 +Y 45 +Y 0 Percent Difference Between MCCOH and the Maximum MCOH at f = 24.25 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y Figure 22: Case 3: MCCOH for all DOFs at 24.25 Hz 36 43 +Y 45 +Y Percent Difference Between CCOH and the Maximum OCOH at f = 53.75 Hz. Plot Min/Max = -0.05% and 5% 2 # F O D t u pt u O 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y 43 +Y 45 +Y Figure 23: Case 4: MCCOH for all DOFs at 53.75 Hz Percent Difference Between MCCOH and the Maximum MCOH at f = 396 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y Figure 24: Case 5: MCCOH for all DOFs at 396 Hz 37 43 +Y 45 +Y FRF Magnitude vs. Frequency for Various Input/Output Combinations FRF FRF FRF FRF 10 )f bl / n(i e d ut i n g a M F R F 10 10 21 -Y,100 +X 13 +Y,100 +X 21 -Y,101 +Z 13 +Y,101 +Z -2 -3 -4 20 40 60 80 100 120 140 160 Frequency (Hz) Multiple Coherence for Points 13 and 21 directions +Y and -Y. 1 0.8 MCOH 13+Y MCOH 21-Y 0.6 H O C M 0.4 0.2 0 0 100 200 300 500 600 700 800 Multiple Combined Coherence for Point 13 and 21 directions +Y and -Y. 1 0.8 H O C C M 400 Combined Coherence 7 Frequencies with 5% Improvement above 10 Hz. 0.6 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 25: Case 4 FRF and coherence functions showing the 24.25 Hz small resonances and 100.75 Hz antiresonance valleys at which combined coherence improves over multiple coherence 38 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 10 )f bl / n(i e d ut i n g a M F R F 10 10 FRF FRF FRF FRF -1 11 +Z,100 +X 12 +Z,100 +X 11 +Z,101 +Z 12 +Z,101 +Z -2 -3 -4 40 60 80 100 120 140 160 180 200 Frequency (Hz) Multiple Coherence for Points 11 and 12 directions +Z and +Z. 1 MCOH 11+Z MCOH 12+Z 0.8 0.6 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 11 and 12 directions +Z and +Z. 1 0.8 Combined Coherence 2 Frequencies with 5% Improvement above 10 Hz. 0.6 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 26: Case 4 FRF and coherence functions showing the 53.75 Hz resonant peak at which combined coherence improves over multiple coherence 39 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 )f bl / n(i e d ut i n g a M F R F 10 10 10 FRF FRF FRF FRF -2 44 +X,100 +X 45 +X,100 +X 44 +X,101 +Z 45 +X,101 +Z -3 -4 -5 320 340 360 380 400 420 440 460 Frequency (Hz) Multiple Coherence for Points 44 and 45 directions +X and +X. 1 0.8 MCOH 44+X MCOH 45+X 0.6 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 44 and 45 directions +X and +X. 1 0.8 Combined Coherence 44 Frequencies with 5% Improvement above 10 Hz. 0.6 0.4 0.2 0 0 100 200 300 400 500 Frequency (Hz) 600 700 800 Figure 27: Case 4 FRF and coherence functions showing the 395 Hz antiresonance valley at which combined coherence improves over multiple coherence 40 Case 6: Effect of Cyclic Averaging on Nominally Nonlinear System The mounting of the square frame to the H-frame was the same for Case 6 as in Cases 3-5. Since Case 4 seemed to best balance whether the combined coherence improvements were located at resonances or antiresonances, 10 lbf peak input per shaker was used. The only setup modifications from Case 4 were that 75% burst random excitation and cyclic averaging were used. This was done to see if reducing leakage errors drastically modified the frequency and DOF combinations for which MCCOH improved (as in Case 2 compared to Case 1). Table 12: Test Setup for Case 6 Test Setup Description Accelerometer Spread Out, Excitation Type 75% Burst Random Distribution Figure 10 Case 4 Input Voltage 3 V Peak Force 10 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 150 # of Cyclic Averages 10 Attachment Method: Four Vehicle Body Mounts at each corner of the square frame. Two bolts barely hand tightened. Two bolts removed altogether. See Figure 20 and Figure 21. As can be seen from Table 13 below it appears that cyclic averaging not only reduced leakage, but changed the DOFs for which combined coherence detects changes (similar to Case 2). In the case of the small resonance peaks around 25 Hz, which before registered very large changes, virtually no change in MCCOH is seen relative MCOH (Figure 28). This paralleled the expected reduction of leakage at all DOFs at that frequency (see Figure 32 through Figure 35). In the case of the strong resonance at 53.5 Hz, combined coherence again detected no improvement at the resonant frequency (Figure 30). It should be noted, however, that just to either side of the resonant frequency, there was still some improvement when combining DOFs from just the square frame (Figure 29). This again paralleled the expected reduction of leakage on all DOFs for which combined coherence improved (see Figure 32 through Figure 35). In comparing Case 41 6 to Case 4 at the 390-400 Hz antiresonance valley, there was very little change. Virtually the same DOFs registered improvement in MCCOH over MCOH (Figure 31). It should be noted that when comparing MCCOH for the off-peak 55 Hz and for the 396 Hz antiresonance, the DOF combinations which improved, while not exactly the same as in Case 4, are in the same spatial vicinity and are isolated to the square frame. As in Cases 1 and 2, MCCOH for Case 6 no longer improves at frequencies where leakage effects are greatly reduced. Table 13: Table of improvements of 10% or more in Combined Coherence for Case 5 for adjacent points in the same direction (consecutive frequencies enumerated but not listed) Freq Percent Resonance/ # Pt. 1 Dir. 1 Pt. 2 Dir. 2 (Hz) MCCOH MCOH1 MCOH2 Difference Antiresonance 1 2 3 38 39 40 42 +X 42 +X 45 +X 45 +X 44 +Y 44 +Y 45 +X 45 +X 44 +X 44 +X 41 +Y 41 +Y 338.75 339.00 389.00 398.25 204.75 206.00 0.9744 0.9728 0.8807 0.8075 0.9763 0.9763 42 0.8696 0.8741 0.7988 0.4967 0.8808 0.8817 0.8808 0.8837 0.7959 0.6535 0.8813 0.8873 10.6270 antiresonance 10.0823 10.2587 antiresonance 23.5625 10.7755 antiresonance 10.0333 Percent Difference Between MCCOH and the Maximum MCOH at f = 24.25 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y 43 +Y 45 +Y Figure 28: Case 6: MCCOH for all DOFs at 24.25 Hz Percent Difference Between MCCOH and the Maximum MCOH at f = 52 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y Figure 29: Case 6: MCCOH for all DOFs at 52 Hz 43 43 +Y 45 +Y Percent Difference Between MCCOH and the Maximum MCOH at f = 53.5 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y 43 +Y 45 +Y Figure 30: Case 6: MCCOH for all DOFs at 53.5 Hz Percent Difference Between MCCOH and the Maximum MCOH at f = 397 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 42 +X 42 +Z 42 +Y 41 +X 41 +Z 41 +Y 14 +Y 14 +X 14 +Z 13 +Y 13 +X 13 +Z 12 +Y 12 +X 12 +Z 11 +Y 11 +X 11 +Z 21 -X 21 -Y 21 +Z 4 -Y 4 -X 4 +Z 3 -Y 3 -X 3 +Z 2 -Y 2 -X 2 +Z 1 -Y 1 -X 1 +Z 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 2 +Z 4 +Z 11 +Z 13 +Z Output DOF #1 41 +Y Figure 31: Case 6: MCCOH for all DOFs at 397 Hz 44 43 +Y 45 +Y Overlay Plot of FRF's and Multiple Coherence for DOF 45 +X for Cases with and without Burst Random Input and Cyclic Averaging 0 10 Pure Random Input and No Cyclic Averging: 45 +X Burst Random Input and Cyclic Averaging: 45 +X e d ut i n g a M F R F 10 10 10 -2 -4 -6 0 100 200 300 400 500 600 700 800 1 0.8 e d ut i n g a M H O C M 0.6 0.4 0.2 0 Pure Random Input and No Cyclic Averging: 45 +X Burst Random Input and Cyclic Averaging: 45 +X 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 32: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System Overlay Plot of FRF's and Multiple Coherence for DOF 45 +Y for Cases with and without Burst Random Input and Cyclic Averaging 0 10 Pure Random Input and No Cyclic Averaging: 45 +Y Burst Random Input and Cyclic Averaging: 45 +Y e d ut i n g a M F R F 10 10 10 -2 -4 -6 0 100 200 300 400 500 600 700 800 1 0.8 e d ut i n g a M H O C M 0.6 0.4 Pure Random Input and No Cyclic Averaging: 45 +Y Burst Random Input and Cyclic Averaging: 45 +Y 0.2 0 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 33: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System 45 Overlay Plot of FRF's and Multiple Coherence for DOF 12 +Y for Cases with and without Burst Random Input and Cyclic Averaging 0 10 e d ut i n g a M F R F 10 10 -2 -4 Pure Random Input and No Cyclic Averaging: 12 +Y Burst Random Input and Cyclic Averaging: 12 +Y 10 -6 0 100 200 300 400 500 600 700 800 1 0.8 e d ut i n g a M H O C M 0.6 0.4 0.2 0 Pure Random Input and No Cyclic Averaging: 12 +Y Burst Random Input and Cyclic Averaging: 12 +Y 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 34: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System Overlay Plot of FRF's and Multiple Coherence for DOF 12 +Z for Cases with and without Burst Random Input and Cyclic Averaging 0 10 Pure Random Input and No Cyclic Averaging: 12 +Z Burst Random Input and Cyclic Averaging: 12 +Z -1 10 e d ut i n g a M F R F 10 10 10 -2 -3 -4 0 100 200 300 400 500 600 700 800 1 0.8 e d ut i n g a M H O C M 0.6 0.4 0.2 0 Pure Random Input and No Cyclic Averaging: 12 +Z Burst Random Input and Cyclic Averaging: 12 +Z 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 35: Overlay of FRF and Coherence Plots Showing the Effect of a Burst Random Input and Cyclic Averaging on the Nominally Nonlinear System 46 Case 7: Spatial Resolution of the Sensors Case 7 is similar to Case 6, except the accelerometers have been moved to give greater spatial resolution in a particular part of the structure (Figure 11, p. 19). Table 14 below gives details of the entire setup with major changes highlighted. The goal of this test is to determine if it is simply an anomaly that 44X and 45X showed the greatest increase in combined coherence for Case 6, or whether that trend holds as well with a finer spatial resolution on the upper frame. The question arises because points 44X and 45X are DOFs with what appear from visual inspection to be the same mass weighting with no possible nonlinearities in what is by far the shortest, most direct path between them. Table 14: Test Setup for Case 7 Test Setup Description Accelerometer Concentrated, Excitation Type Random Distribution Figure 11 Input Voltage 3V Peak Force 10 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 150 # of Cyclic Averages 8 Attachment Method: Four Vehicle Body Mounts at each corner of the square frame. Two bolts barely hand tightened. Two bolts removed altogether. See Figure 20 and Figure 21. As Table 15 below shows, the results are very similar to Case 6, especially in respect to the improvement of combined coherence in the upper frame. The improvement of combined coherence in the “linear” section from 44 to 48 to 45 is confirmed. In comparing Figure 36 below with Figure 31 (p. 44) one can see that combined coherence improves throughout the upper structure in the 390-400 Hz frequency range and is not isolated to any particular DOF. This leads to questions about whether paths other than the most direct one can play a significant role in the combined coherence calculation. This would make combined coherence much more 47 difficult to use as a tool to spatially locate nonlinearities. The largest improvement in MCCOH over MCOH occurs at 120 Hz. This is very suspect, being a multiple of 60 cycle noise, but is included because the improvement was so much larger than other improvements (e.g., 13Y and 15Y in Table 15). Table 15: Table of improvements of 20% or more in Combined Coherence for Case 7 for adjacent points in the same direction (consecutive frequencies enumerated but not listed) Freq # Pt 1 Dir 1 Pt 2 Dir 2 (Hz) MCCOH MCOH1 MCOH2 % Difference 1 3 -Y 5 +Y 21.5 0.9076 0.3652 0.5746 57.9527 2 3 -Y 53 +Y 21.5 0.9623 0.3652 0.7341 31.0792 3 13 +Y 15 -Y 120 0.9745 0.3109 0.1668 213.4371 4 54 +Y 15 -Y 120 0.9929 0.8178 0.1668 21.4048 5 45 +Z 48 +Z 389.75 0.9389 0.7703 0.708 21.8744 37 45 +Z 48 +Z 398.25 0.7429 0.5414 0.5818 27.6789 38 48 +Z 44 +Z 394 0.6656 0.4648 0.5442 22.3172 48 48 +Z 44 +Z 397.25 0.6052 0.4859 0.4767 24.5447 49 52 +X 45 +X 389.5 0.9489 0.6942 0.7838 21.077 90 52 +X 45 +X 400.25 0.7489 0.4977 0.5566 34.5369 91 52 -Y 45 +Y 28.25 0.9776 0.7797 0.795 22.9631 102 52 -Y 45 +Y 31 0.9721 0.7789 0.4678 24.8047 103 52 -Y 45 +Y 55 0.943 0.6572 0.7116 32.5248 105 52 -Y 45 +Y 55.5 0.9418 0.7789 0.5354 20.903 Percent Difference Between MCCOH and the Maximum MCOH at f = 396.75 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y Figure 36: Case 7: MCCOH for all DOFs at 396.75 Hz 48 45 +Y Case 8: Effect of Spatial Variation of the Input Forces Case 8 is similar to Case 7, except shaker #1 has been moved to the upper frame (see Figure 11, p. 19). Table 16 below gives details of the entire setup with major changes highlighted. Table 16: Test Setup for Case 8 Test Setup Description Accelerometer Concentrated, Excitation Type Random Distribution Figure 11 Input Voltage 3V Peak Force 10 lbf Shaker #2 Location Pt. 101 Shaker #1 Location Pt. 102 # of Spectral Averages 150 # of Cyclic Averages 10 Attachment Method: Four Vehicle Body Mounts at each corner of the square frame. Two bolts barely hand tightened. Two bolts removed altogether. See Figure 20 and Figure 21. The first noticeable result of moving the shaker is that the number of DOFs with improvements in combined coherence has been reduced (Table 17 below). Second, combined coherence no longer detects any nonlinearities in the upper structure. Moving Shaker #1 from point 100 to point 102 on the square frame has eliminated the detection of nonlinearities in that section of the structure. In contrast, Shaker #2 is located at point 101 between DOFs 3 and 5 and a nonlinearity is consistently detected in this part of the structure. These MCCOH improvements are concentrated around the 123 Hz antiresonance valley (Figure 38). Table 17: Table of improvements of 20% or more in Combined Coherence for Case 8 for adjacent points in the same direction pt1 dir1 pt2 dir2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference 3 13 13 13 13 13 -X +Z +Z +Z +Z +Z 5 54 54 54 54 54 +X +Z +Z +Z +Z +Z 123.25 113.50 113.75 114.00 114.25 114.50 0.7135 0.7391 0.7146 0.6890 0.6412 0.7266 49 0.5469 0.5146 0.5695 0.5367 0.4825 0.5963 0.5185 0.6090 0.5382 0.5073 0.4458 0.4763 30.4685 21.3541 25.4825 28.3700 32.8984 21.8531 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 10 )f bl / n(i e d ut i n g a M F R F 10 10 10 FRF FRF FRF FRF -1 3 -X,100 +X 5 +X,100 +X 3 -X,101 +Z 5 +X,101 +Z -2 -3 -4 -5 60 80 100 120 140 160 180 Frequency (Hz) 200 220 240 260 Multiple Coherence for Points 3 and 5 directions -X and +X. 1 MCOH 3-X MCOH 5+X 0.8 0.6 H O C M 0.4 0.2 0 0 50 100 150 200 250 300 Multiple Combined Coherence for Point 3 and 5 directions -X and +X. 1 0.8 H O C C M 0.6 0.4 Combined Coherence 6 Frequencies above 10 Hz with 5% Improvement 0.2 0 0 50 100 150 Frequency (Hz) 200 250 300 Figure 37: Case 7 FRF and coherence functions showing the 123.25 and 175 Hz antiresonance valleys at which combined coherence improves over multiple coherence 50 FRF Magnitude vs. Frequency for Various Input/Output Combinations FRF FRF FRF FRF 10 )f bl / n(i e d ut i n g a M F R F 10 10 10 13 +Y,100 +X 15 -Y,100 +X 13 +Y,101 +Z 15 -Y,101 +Z -2 -3 -4 -5 50 100 150 Frequency (Hz) 200 250 Multiple Coherence for Points 13 and 15 directions +Y and -Y. 1 0.8 0.6 H O C M 0.4 0.2 0 MCOH 13+Y MCOH 15-Y 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 13 and 15 directions +Y and -Y. 1 0.8 H O C C M 0.6 0.4 Combined Coherence 10 Frequencies with 5% Improvement above 10 Hz. 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 38: Case 7 FRF and coherence functions showing the 175 Hz antiresonance valley at which combined coherence improves over multiple coherence 51 Since all of the frequencies with significant multiple coherence improvements are at antiresonances, the question arises as to why combined coherence did not register an improvement at the 390 Hz antiresonance as in Case 6. The only change is the location of one input force. From Figure 39 below it can be seen that MCOH greatly improves for the square frame in this range because dominant FRF for this DOF is now linear. This would lead one to believe that where the inputs are located has a significant effect on the ability of combined coherence to detect nonlinearities. This also leads to more questions about how critical the energy path is in this detection process. For instance, if most of the energy is coming from a linear path, is combined coherence able to detect a nonlinearity from another less dominant path? Also, if all of the energy comes through a nonlinearity some distance away, is combined coherence able to distinguish that fact when combining two DOFs in very close proximity with no nonlinearity between them? Also, does the number of nonlinear paths reduce the ability of combined coherence to spatially locate nonlinearities? 10 e d ut i n g a M F R F 10 10 10 Overlay Plot of FRF's and Multiple Coherence for DOF 45 +X for Various Tests 0 -2 -4 -6 0 100 200 300 400 500 600 Test A: H2(45X, 100 skew) Test A: H2(45X, 101 skew) Test B: H2(45X, 102 skew) Test B: H2(45X, 101 700 800skew) 1 0.8 e d ut i n g a M H O C M Test A: MCOH(45X) Test B: MCOH(45X) 0.6 0.4 0.2 0 0 100 200 300 400 500 600 700 800 Frequency (Hz) Figure 39: Overlay of H(45x,*x) before and after moving shaker #1 from the H-Frame to the Square Frame 52 Case 9: Effect of Reducing the Number of Connection Points Between the H-Frame and Square Frame to One Case 9 is similar to Case 8, except shaker #1 has been moved and mounts were removed. Figure 40 shows the removal of the 3 mounts, while Figure 11 (p. 19) shows the location of the inputs. Table 18 below gives details of the entire setup with major changes highlighted. The goal of this test is to determine whether reducing the number of paths between the H-frame and square frame improves the ability of combined coherence to detect and spatially locate nonlinearities. Table 18: Test Setup for Case 9 Test Setup Description Accelerometer Concentrated, Excitation Type Random Distribution Figure 11 Input Voltage 3V Peak Force 10 lbf Shaker #1 Location Pt. 100 Shaker #2 Location Pt. 101 # of Spectral Averages 200 # of Cyclic Averages 10 Attachment Method: Three vehicle body mounts have been removed and the square frame is supported by shock cord at each of those corners. The one original body mount that remains is not modified in any way. See Figure 40 and Figure 41. Figure 40: Pictures of the setup for Cases 9-11 showing the removal of 3 connection points and the suspension of the square frame by shock cord 53 Figure 41: Layout and schematic of the connection point between the H-frame and the square frame for Cases 9-11 Table 19 below shows that by isolating the square frame and by reducing the number of paths that couple the two structures together, the number of combined coherence improvements greatly increases. Depending on the frequency, combined coherence improvements are spread throughout both structures (Figure 42) or isolated to the square frame (Figure 43 through Figure 46). Of the few improvements in MCCOH from combining a DOF from both structures, none involve adjacent DOFs (e.g., Pt. 52 and Pt. 54), but DOFs with several other DOFs in between the two (e.g. 51X and 15Z at 21.5 Hz in Figure 42). When looking at the frequencies and FRF’s for which combined coherence improves, they are at both antiresonances (Figure 47, Figure 49, Figure 50 and Figure 51) and resonant peaks (Figure 48, Figure 51 and Figure 52). Some of the resonant peaks at which MCCOH improves are atypical in shape compared to other resonant peaks around them. One example would be an extremely small resonant peak in the FRF relative to the surrounding resonant peaks which only shows up in the FRF’s for a single input (e.g. see 54 21.5 Hz FRF in Figure 47). However, combined coherence does not seem to be able to spatially locate these nonlinearities. Another interesting thing to note about combined coherence is that often the largest increases come about from combining two DOFs in orthogonal directions at the same spatial point (e.g., 44Y and 44Z in Figure 46). Off axis sensitivity may also be the cause of some of these improvements in MCCOH over MCOH. The amplititude of some of the improvements, however it unlikely that it is the only cause of cross axis MCCOH improvements. 55 Table 19: Case 9 improvements of 20% or more in Combined Coherence for Case 9 for adjacent points in the same direction (closely spaced frequencies enumerated but not listed) # Pt 1 Dir 1 Pt 2 Dir 2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference 1 2 4 11 12 45 47 48 51 52 55 56 57 59 62 63 65 66 126 128 180 181 246 247 251 278 279 280 281 284 285 291 292 293 333 334 3 13 52 52 52 52 52 51 51 51 51 51 15 44 44 44 46 46 46 43 43 47 47 47 45 45 45 45 48 48 48 48 48 48 48 48 -Y +Y -Y -Y -Y -Y +Z +Z +Z -X -X +Z -Y +X +X +Z +X +Z +Z +X +X +X +X +X +Y +Y +Z +Z +X +X +X +X +Y +Z +Z +Z 5 15 51 51 51 51 51 46 46 44 44 44 54 46 46 46 43 43 43 47 47 45 45 45 48 48 48 48 44 44 44 44 44 44 44 44 +Y -Y +Y +Y +Y +Y +Z +Z +Z +X +X +Z +Y +X +X +Z +X +Z +Z +X +X +X +X +X +Y +Y +Z +Z +X +X +X +X +Y +Z +Z +Z 21.5 21.5 400 402 628.25 637 800 799.25 800 799.25 800 401.25 21.5 799.25 800 512 408.75 624 639.5 626.5 641.25 626.5 643.75 799.25 631.75 640 799.75 800 401 401.75 798.5 800 60 626.5 637 800 0.9074 0.8853 0.916 0.942 0.4678 0.392 0.4277 0.5254 0.3864 0.6923 0.666 0.8723 0.9028 0.761 0.6021 0.8729 0.7683 0.7712 0.8424 0.9378 0.6287 0.9171 0.8435 0.6998 0.856 0.8066 0.3253 0.266 0.8575 0.9045 0.795 0.6192 0.3473 0.6021 0.5697 0.2535 56 0.6074 0.7078 0.7579 0.4908 0.3152 0.2856 0.3035 0.4327 0.2923 0.5536 0.4868 0.5555 0.671 0.3508 0.3313 0.5823 0.625 0.6207 0.6832 0.2972 0.4954 0.7519 0.7023 0.5614 0.5432 0.5618 0.2675 0.2131 0.6553 0.6745 0.6187 0.2388 0.2737 0.4939 0.4552 0.1738 0.6191 0.671 0.6809 0.7799 0.3657 0.172 0.2923 0.3533 0.2579 0.3508 0.3313 0.7103 0.5984 0.609 0.4265 0.6831 0.2912 0.5581 0.5085 0.7519 0.5206 0.3413 0.6937 0.3492 0.7069 0.6634 0.2293 0.1738 0.6813 0.7256 0.576 0.3313 0.2218 0.4511 0.4442 0.1854 46.5721 25.0707 20.8619 20.7799 27.9134 37.25 40.9337 21.403 32.1966 25.0432 36.8184 22.8126 34.5451 24.9547 41.1844 27.7883 22.9373 24.2417 23.3002 24.7151 20.7683 21.9647 20.0991 24.6688 21.0935 21.5931 21.603 24.8584 25.856 24.6483 28.5094 86.8891 26.8904 21.8952 25.1508 36.7231 Percent Difference Between MCCOH and the Maximum MCOH at f = 21.5 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 42: Case 9: MCCOH for all DOFs at 21.5 Hz Percent Difference Between MCCOH and the Maximum MCOH at f = 401.5 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y Figure 43: Case 9: MCCOH for all DOFs at 401.5 Hz 57 43 +Y 45 +Y Percent Difference Between MCCOH and the Maximum MCOH at f = 512 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 44: Case 9: MCCOH for all DOFs at 512 Hz Percent Difference Between MCCOH and the Maximum MCOH at f = 632.5 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y Figure 45: Case 9: MCCOH for all DOFs at 632.5 Hz 58 43 +Y 45 +Y Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 305.6667% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 300 250 200 150 100 50 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 0 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y Figure 46: Case 9: MCCOH for all DOFs at 800 Hz 59 43 +Y 45 +Y FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 10 FRF FRF FRF FRF -1 13 +Y,100 +X 15 -Y,100 +X 13 +Y,101 +Z 15 -Y,101 +Z -2 M R 10 -3 10 15 20 25 30 35 40 45 Frequency (Hz) Multiple Coherence for Points 13 and 15 directions +Y and -Y. 1 0.8 MCOH 13+Y MCOH 15-Y 0.6 H O C M 0.4 0.2 0 0 100 200 300 500 600 700 800 Multiple Combined Coherence for Point 13 and 15 directions +Y and -Y. 1 0.8 H O C C M 400 Combined Coherence 2 Frequencies above 10 Hz. with 5% Improvement 0.6 0.4 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 47: Case 9 FRF and coherence functions showing the 21.5 Hz combined coherence improvements over multiple coherence 60 FRF Magnitude vs. Frequency for Various Input/Output Combinations FRF FRF FRF FRF 10 )f bl / n(i e d ut i n g a M F R F 10 10 48 +Y,100 +X 44 +Y,100 +X 48 +Y,101 +Z 44 +Y,101 +Z -3 -4 -5 45 50 55 60 65 Frequency (Hz) 70 75 80 Multiple Coherence for Points 44 and 48 directions +Y and +Y. 1 0.8 0.6 H O C M 0.4 MCOH 44+Y MCOH 48+Y 0.2 0 0 50 100 150 200 250 300 Multiple Combined Coherence for Point 44 and 48 directions +Y and +Y. 1 0.8 H O C C M 0.6 0.4 Combined Coherence 3 Frequencies above 10 Hz. with 5% Improvement 0.2 0 0 50 100 150 Frequency (Hz) 200 250 300 Figure 48: Case 9 FRF and coherence functions showing the 45 Hz combined coherence improvements over multiple coherence 61 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 )f bl / n(i e d ut i n g a M F R F 10 10 10 FRF FRF FRF FRF -2 46 +Z,100 +X 44 +Z,100 +X 46 +Z,101 +Z 44 +Z,101 +Z -3 -4 -5 480 490 500 510 520 530 Frequency (Hz) 540 550 560 570 Multiple Coherence for Points 44 and 46 directions +Z and +Z. 1 0.8 0.6 H O C M 0.4 0.2 0 MCOH 44+Z MCOH 46+Z 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 44 and 46 directions +Z and +Z. 1 0.8 H O C C M 0.6 0.4 0.2 0 Combined Coherence 5 Frequencies above 10 Hz. with 5% Improvement 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 49: Case 9 FRF and coherence functions showing the 512 Hz combined coherence improvements over multiple coherence 62 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 )f bl / n(i e d ut i n g a M F R F 10 10 10 FRF FRF FRF FRF -2 46 +X,100 +X 47 +X,100 +X 46 +X,101 +Z 47 +X,101 +Z -3 -4 -5 300 320 340 360 380 400 420 Frequency (Hz) 440 460 480 500 Multiple Coherence for Points 46 and 47 directions +X and +X. 1 0.8 0.6 H O C M 0.4 MCOH 46+X MCOH 47+X 0.2 0 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 46 and 47 directions +X and +X. 1 0.8 H O C C M 0.6 0.4 Combined Coherence 84 Frequencies above 10 Hz. with 5% Improvement 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 50: Case 9 FRF and coherence functions showing the 400 Hz combined coherence improvements over multiple coherence 63 10 FRF Magnitude vs. Frequency for Various Input/Output Combinations -1 FRF Magnitude vs. Frequency for Various Input/Output Combinations 10 10 )f bl / n(i e d ut i n g a M F R F 10 -2 10 )f bl / n(i e d ut i n g a M F R F -3 10 10 10 FRF FRF FRF FRF -1 51 +Y,100 +X 52 -Y,100 +X 51 +Y,101 +Z 52 -Y,101 +Z -2 -3 -4 -4 360 370 380 390 400 410 420 Frequency (Hz) 430 440 450 460 580 600 620 640 660 Frequency (Hz) 680 700 720 740 Multiple Coherence for Points 51 and 52 directions +Y and -Y. 1 0.8 0.6 H O C M 0.4 0.2 0 MCOH 51+Y MCOH 52-Y 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 51 and 52 directions +Y and -Y. 1 0.8 H O C C M 0.6 0.4 0.2 0 Combined Coherence 73 Frequencies above 10 Hz. with 5% Improvement 0 100 200 300 400 Frequency (Hz) 500 600 700 Figure 51: Case 9 FRF and coherence functions showing the 400 and 635 Hz combined coherence improvements over multiple coherence 64 800 10 10 )f bl / n(i e d ut i n g a M F R F 10 10 FRF Magnitude vs. Frequency for Various Input/Output Combinations -1 FRF FRF FRF FRF 44 +Y,100 +X 44 +Z,100 +X 44 +Y,101 +Z 44 +Z,101 +Z -2 -3 -4 700 710 720 730 740 750 760 Frequency (Hz) 770 780 790 800 Multiple Coherence for Points 44 and 44 directions +Y and +Z. 1 0.8 0.6 H O C M 0.4 0.2 0 MCOH 44+Y MCOH 44+Z 0 100 200 300 400 500 600 700 800 Multiple Combined Coherence for Point 44 and 44 directions +Y and +Z. 1 0.8 H O C C M 0.6 0.4 Combined Coherence 57 Frequencies above 10 Hz. with 5% Improvement 0.2 0 0 100 200 300 400 Frequency (Hz) 500 600 700 800 Figure 52: Case 9 FRF and coherence functions showing the 800 Hz combined coherence improvements over multiple coherence 65 Cases 10-11: Effect of Spectral Averaging Spectral averaging with random excitation is known to reduce the effects of nonlinearities, sometimes making it possible to even get good estimates of FRF’s at frequencies affected by nonlinearities. The setup for Cases 10-11 is the same as Case 9, except the number of spectral averages has been varied and cyclic averaging has been eliminated. Table 20 below gives details of the entire setup with major changes highlighted. The purpose of this test is to see if reducing the number of spectral averages increases the ability of combined coherence to spatially locate the nonlinearities. Table 20: Test Setup for Cases 10-12 Excitation Type Random Test Setup Description Accelerometer Distribution Peak Force Shaker #2 Location Concentrated, Figure 11 10 lbf Pt. 101 Input Voltage 3V Shaker #1 Location Pt. 100 Case 10: 25 # of Cyclic Averages 0 # of Spectral Averages Case 11: 100 # of Cyclic Averages 0 # of Spectral Averages Attachment Method: Three vehicle body mounts have been removed and the square frame is supported by shock cord at each of those corners. The one original body mount that remains is not modified in any way. See Figure 40 and Figure 41. The 10 greatest percentage differences between combined coherence and coherence are given in Table 21 and Table 22 below. Greater differences are seen with less averaging, indicating that some of the effect of nonlinearities may be averaged out. Combined coherence improvements for all DOF combinations are shown in Figure 53 through Figure 54 (638 Hz) and in Figure 55 through Figure 56 (800 Hz). An FRF showing the change in “smoothness” of the FRF and COH plots at 638 Hz and 800 Hz is given in Figure 57. For the most part, the MCCOH improvements still seem to be isolated primarily to DOF combinations either within the square frame or within 66 the H-frame. Thus the effect of averaging on the ability of combined coherence to spatially locate nonlinearities is minimal. The MCCOH improvements over MCOH were completely restricted to the square frame at the 638 and 800 Hz antiresonance areas (Figure 55 and Figure 56). These spatial trends were not affected by spectral averaging. Table 23 below summarizes the similarities in features for the 638 Hz and 800 Hz antiresonance areas. Table 21: 10 Greatest Differences between MCCOH and MCOH for Case 10 (25 Spectral Averages) # Pt 1 Dir 1 Pt 2 Dir 2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference 1 2 3 4 5 6 7 8 9 10 44 43 47 47 52 43 47 48 47 44 +Z +X +X +X +Z +X +X +X +X +Z 46 47 45 45 45 47 45 44 45 46 +Z +X +X +X +Z +X +X +X +X +Z 800.00 638.25 638.25 634.75 800.00 634.75 638.00 800.00 638.50 799.00 0.5334 0.6408 0.8545 0.8053 0.3343 0.7723 0.9156 0.7882 0.8470 0.7383 0.1052 0.1862 0.1609 0.2788 0.1167 0.0969 0.2916 0.1777 0.3277 0.2923 0.1220 0.1609 0.2733 0.1719 0.0769 0.2788 0.3363 0.3020 0.2658 0.2241 337.2935 244.1921 212.6885 188.8842 186.4567 177.0420 172.2692 160.9655 158.4943 152.5539 Table 22: 10 Greatest Differences between MCCOH and MCOH for Case 12 (100 Spectral Averages) # Pt 1 Dir 1 Pt 2 Dir 2 Freq (Hz) MCCOH MCOH1 MCOH2 % Difference 1 2 3 4 5 6 7 8 9 10 44 44 47 47 43 43 43 44 44 43 +Y +Z +X +X +X +X +X +Y +Z +X 46 46 45 45 47 47 47 46 46 47 +Y +Z +X +X +X +X +X +Y +Z +X 800.00 799.75 638.00 638.25 638.00 633.00 634.75 799.75 799.50 638.25 0.7604 0.5018 0.8696 0.8533 0.7735 0.8327 0.8382 0.7338 0.5813 0.7629 0.1587 0.1768 0.3107 0.3206 0.2343 0.2692 0.2873 0.2811 0.2430 0.2403 0.2379 0.1775 0.3073 0.3159 0.3107 0.3427 0.3466 0.3066 0.2297 0.3206 Table 23: Similarities Highlighted Between the 638 Hz and 120 Hz MCCOH Improvements Frequency 638 Hz 800 Hz Frequency Band More Broad (>10 Hz) More Broad (>10 Hz) Structures Affected Square Frame Only Square Frame Only Percent Improvement of MCCOH over MCOH Large (~250% max) Large (~330% max) Trends Hold for Both 25 and 100 Spectral Averages Yes Yes Description of FRF at Frequency Band of Interest Ratty but distinct antiresonance valley Ratty but distinct antiresonance valley 67 219.5513 182.6535 179.8830 166.1219 148.9619 142.9733 141.8571 139.3334 139.1734 137.9200 Percent Difference Between MCCOH and the Maximum MCOH at f = 638.25 Hz. Plot Min/Max = -0.05% and 244.1921% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 200 150 100 50 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y 0 Percent Difference Between MCCOH and the Maximum MCOH at f = 638.25 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 53: Case 12: MCCOH for all DOFs at 638.25 Hz (25 spectral averages) 68 Percent Difference Between MCCOH and the Maximum MCOH at f = 638.25 Hz. Plot Min/Max = -0.05% and 166.1219% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 160 140 120 100 80 60 40 20 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y 0 Percent Difference Between MCCOH and the Maximum MCOH at f = 638.25 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 54: Case 12: MCCOH for all DOFs at 638.25 Hz (100 spectral averages) 69 Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 441.6431% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 400 350 300 250 200 150 100 50 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y 0 Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 55: Case 12: MCCOH for all DOFs at 800 Hz (25 spectral averages) 70 Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 297.2975% 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 250 200 150 100 50 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y 0 Percent Difference Between MCCOH and the Maximum MCOH at f = 800 Hz. Plot Min/Max = -0.05% and 5% 5 45 +X 45 +Z 45 +Y 44 +X 44 +Z 44 +Y 43 +X 43 +Z 43 +Y 47 +X 47 +Z 47 +Y 46 +X 46 +Z 46 +Y 15 -Y 15 -X 15 +Z 13 +Y 13 +X 13 +Z 54 -X 54 +Y 54 +Z 52 -Y 52 +Z 52 +X 48 +X 48 +Z 48 +Y 5 +Y 5 +X 5 +Z 3 -Y 3 -X 3 +Z 53 -X 53 +Y 53 +Z 51 +Y 51 +Z 51 -X 101 +Z 100 +X 100 +X 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 53 +Z 5 +Z 52 +X 13 +Z Output DOF #1 46 +Y 43 +Y 45 +Y Figure 56: Case 12: MCCOH for all DOFs at 800 Hz (100 spectral averages) 71 10 e d ut i n g a M F R F 10 10 10 0 Overlay Plot of FRF's and Multiple Coherence for DOF 43 +X for Various Tests 25 Avg: H2(43 +X,100 +X) 25 Avg: H2( 43 +X,101 +X) 100 Avg: H2(43 +X,100 +X) 100 Avg: H2(43 +X,101 +X) -2 -4 -6 550 600 650 700 750 800 1 0.8 e d ut i n g a M H O C 0.6 0.4 0.2 0 550 25 Avg: MCOH(43 +X) 100 Avg: MCOH(43 +X) 600 650 700 750 800 Frequency (Hz) Figure 57: Overlay of FRF and Coherence Plots Showing the Effect of a Spectral Averaging for DOF 43 +X 72 Discussion The theoretical lumped parameter study shows that combined coherence improves over ordinary coherence for two DOFs with a nonlinear connection between them. In contrast, when the nonlinearity was not directly between the two DOFs being combined, combined coherence did not improve over the ordinary coherence of both of the original DOFs. When this was tested for a real system with distributed parameters, Case 1 showed that even a nominally linear system had large improvements at resonant peaks (Table 4, p. 22). Case 2 showed that for the linear structure the dominant combined coherence improvements all disappeared when steps were taken to eliminate leakage errors (Table 6, p. 27). This showed up not only as improvements in MCOH but in a lack of improvement in MCCOH over MCOH at the frequencies which previously had been affected. There were still some much smaller improvements in multiple combined coherence over multiple coherence. These no longer occurred at resonant peaks but instead at antiresonances, suggesting that combined coherence may be affected by leakage errors. Cases 3-5 showed the effect of varying the input force level without cyclic averaging. While there was some change in the dominant frequencies showing improvement and in what part of the structure was affected the most, the largest changes for both high and low forcing levels occurred between DOFs which had no possible nonlinearity in the most direct path between them. Examples of this are the improvements between points 12 and 21 for low forcing levels (Table 9, p. 33) and the improvements between points 44 and 45 for high forcing levels (Table 11, p. 35). 73 Case 6 was the application of cyclic averaging and burst random input to reduce leakage errors. As with the linear structure, everywhere that MCOH greatly improved (indicating a reduction in leakage), MCCOH was no longer greater than MCOH (Figure 30 and Figure 32 through Figure 35). Unlike the nominally linear structure, though, there were frequency bands where the large drops in coherence did not seem to be affected by leakage (Figure 31 through Figure 35). At those frequencies combined coherence still improved over regular coherence. As in Case 2, it seemed that leakage was having an effect on the ability of combined coherence to detect nonlinearities. Case 7 examined the earlier noted pattern of large improvements in combined coherence occurring between DOFs with very linear direct paths between them by changing the density of the accelerometers to see if the effect still occurred with even more direct linear paths between the DOFs. It was discovered that there was basically no effect from doubling the density of the accelerometers in certain areas in question (Figure 11, p. 19). The combined coherence improvements seemed to affect all of the DOFs in the upper structure at certain frequencies instead of primarily affecting those DOFs nearest to the nonlinear connection point, making process of locating a nonlinearity using combined coherence alone impossible in this case. Case 8 looked at how changing the location of the input forces would affect the ability of combined coherence to detect and spatially locate nonlinearities. Moving one of the shakers to the square frame eliminated any combined coherence improvements in the upper structure. Where before the largest improvements in combined coherence occurred between DOFs on the 74 square frame away from the inputs, now there were no improvements seen (Table 17, p. 49). Figure 39 shows that this does not mean that all the FRF’s on the upper frame were clean and that multiple coherence never dropped. For FRF’s relating the response of the square frame to the input on the H-frame, there were still very large drops in MCOH and ratty FRF’s, indicating that a nonlinearity was still affecting the structure there. However, the dominant FRF was linear and combined coherence would no longer improve. Case 9 examined whether reducing the number of nonlinear paths would have an effect on combined coherence. It was discovered that by removing three of the connection points between the H-frame and the square frame that the number of MCCOH improvements actually increased relative to other similar test setups (e.g., Case 7). Thus, it appeared that reducing the number of paths between the two structures did little to improve the ability of MCCOH to spatially locate the nonlinearities. In fact the number of MCCOH improvements and their magnitude both increased between DOFs with very linear direct paths between them (Table 19, p. 56). One thing that was not considered was the structure’s interaction with the supporting shock cord. If energy was being transmitted through them, then the number of nonlinear paths may not have been sufficiently reduced at affected frequencies. Another thing that was not considered was whether the energy distribution was sufficient to excite the nonlinearities. Cases 10 and 11 looked at the effect of spatial averaging to see if the nonlinearities were getting averaged out at all. It was seen, when comparing cases with 25 and 100 spectral averages and no cyclic averaging, that there did seem to be a slight increase in the magnitude of MCCOH improvements when fewer averages were done, but there was no qualitative improvement in the 75 ability of combined coherence to detect and locate nonlinearities. The same parts of the structure saw improvements in MCCOH regardless of the number of averages (Table 21 and Table 22, p. 67). Cases 9-10 also highlight the fact that MCCOH improvements were primarily at antiresonances. This highlights the fact that noise can be significant in these parts of FRF. In retrospect, a test should have been done to determine whether there truly were nonlinearities being excited in those frequency ranges. Assuming that nonlinearities are the cause of the improvements in combined coherence, MCCOH seems unable to distinguish between energy transmitted through a nonlinearity some distance from two DOFs located adjacent to each other and energy transmitted through a nonlinearity very close to or even between the two adjacent DOFs. Often there even seems to be more of an improvement in MCCOH among DOFs located in the upper structure where all of the energy comes through a nonlinearity rather than across the nonlinear connection itself where some energy comes directly from the shaker along a linear path and some comes through the nonlinear connection. 76 Summary and Conclusions In conclusion, it does not seem that combined coherence, as it is currently applied, is particularly useful in spatially locating nonlinearities in a structure for a variety of reasons. Several observations are highlighted below. 1) Leakage errors seem to give false improvements and throw off the calculation whether in a nominally linear structure or a nominally nonlinear structure (Case 2 vs. Case 1 and Case 6 vs. Case 4). 2) Varying the magnitude of force input seems to vary the types of nonlinearities detected, but does not affect the ability of combined coherence to spatially locate nonlinearities (Cases 3-5). 3) Doubling the density of the response accelerometers does not seem to improve the ability of combined coherence to spatially locate nonlinearities (Case 7 vs. Case 6). 4) There seems to be a great sensitivity to the location of the input and the amount of force coming from a direct linear path (Case 8 vs. Case 7). 5) Reducing the number of nonlinear paths does not seem to improve the ability of MCCOH to spatially locate nonlinearities. In fact, it can get worse (Case 9 vs. Case 7). 6) Spectral averaging does not seem to qualitatively affect the ability of combined coherence to spatially locate nonlinearities, although overall larger improvements are seen with fewer averages (Cases 10-11). In addition one unexpected and unusual result noted was that sometimes the greatest improvement in combined coherence came when combining DOFs at the same point, but in orthogonal directions (e.g., 44Y and 44Z in Figure 46, p. 59). In some ways this might be 77 considered an extreme example of Case 7 changing the density of the responses. Off-axis errors in the accelerometers may also be the cause of some of these MCCOH improvements over MCOH. Scaling the motions of the DOFs by their respective masses was not done and cannot be easily done for this distributed parameter case. Perhaps if correct mass scaling factors were obtained and used, the greatest improvements would be seen across the nonlinear connections rather than between DOFs on the same linear structure. It is doubtful whether scaling the masses would produce the desired results, though, due to the fact that even when efforts were made to even out the differences in mass scaling by placing accelerometers on the brackets on either side of the nonlinearity, MCCOH improvements were still not seen between those DOFs. Whether or not a nonlinear situation was even excited is another question. Other test procedures were not used which would tell definitively whether or not a nonlinear situation was excited, and if so, where the nonlinearities were located. It was not determined whether a nonlinearity was sufficiently excited and if so where it was (e.g., which mount, tire behaved nonlinearly). 78 Future Work Combined coherence, though a useful tool in small lumped parameter models, will be more difficult to apply usefully to real structures. Work still needs to be done to understand its properties when applied to real structures before it is widely used. To that end suggestions for future work are listed below. 1) Before further evaluation of combined coherence on a real structure, testing should be done using other methods (such as swept sine input) to check that nonlinearities are in the expected locations and that they are being excited sufficiently. 2) It would also be useful to test each structure separately prior to connecting them and to perhaps hang each structure vertically. Testing each structure separately before combining them will ensure that any new MCCOH improvements over MCOH are due to the nonlinear connections. Also, hanging the frame will ensure that when a mount is removed, any change in the MCCOH calculations is a result of the missing mount and not the additional shock cord supports. 3) A theoretical model of a continuous system with a full mass matrix should be used to see if there is something inherent in continuous systems that make it difficult or impossible to use combined coherence to detect nonlinearities. 4) The path dependence of combined coherence should be looked into more. One start might be a theoretical lumped mass model with hundreds of interconnected DOFs. Other experiments with real structures would also be useful. 5) There seemed to be a great sensitivity of the structure to the location of the input force. Further testing regarding the sensitivity of the formulation to this aspect, may lead to a testing method that makes use of this aspect to locate nonlinearities. 79 6) A sensitivity analysis of the effect of the spatial density of the responses on MCCOH calculations would be useful as well, since it is unclear how close the responses need to be in order to distinguish between nonlinear internal forces and nonlinear external forces. 7) The apparent sensitivity of MCCOH to leakage needs to be examined in greater detail to see if the MCCOH improvements may have been due to nonlinearities or if they were truly due to leakage. 80 References [1] T. Roscher, Detection of Structural Non-Linearities using the Frequency Response and Coherence Functions, Master’s Thesis, University of Cincinnati, 2000 [2] T. Roscher, R. J. Allemang, A. W. Phillips, A New Detection Method for Structural NonLinearities, Proceedings of the International Conference on Noise and Vibration Engineering, September 13-15, 2000, Katholieke Universiteit, Leuven, Belguim pp. 695702 [3] K. Worden and G. R. Tomlinson, Nonlinearity in Structural Dynamics: Detection, Identification and Modeling, Philadelphia, PA: Institute of Physics Publishing, 2001 [4] D. L. Brown, Excitation Signals for Modal Analysis, University of Cincinnati, 1996 [5] R. J. Allemang, Vibrations: Experimental Modal Analysis, University of Cincinnati, 1999 [6] R. J. Allemang, Nonlinearities: A Modal Analysis Perspective, University of Cincinnati, 2003 [7] J. S. Bendat, Nonlinear System Techniques and Applications, John Wiley and Sons, Inc., New York, 1998 [8] Bendat, Julius S., Robert N. Coppolino, Paul A. Palo, Identification of Physical Parameters with Memory in Non-Linear Systems, International Journal of Nonlinear Mechanics, Vol: 30, Issue: 6, pp. 841-860 [9] D. M. Storer, G. R Tomlinson, Recent Developments in the Measurement and Interpretation of Higher Order Transfer Functions from Non-Linear Structures, Mechanical Systems and Signal Processing, Vol: 7, Issue: 2, March 1993, pp. 173-189 81 [10] Q. Chen, G. R. Tomlinson, A New Type of Time Series Model for the Identification of NonLinear dynamical Systems, Mechanical Systems and Signal Processing (1994) 8(5), pp. 531-54 [11] R. Ghanem, F. Romeo, A Wavelet Based Approach for Model and Parameter Identification of Non-Linear Systems, International Journal of Non-Linear Mechanics, Vol: 36, Issue 5, July 2001, pp. 835-859 [12] Y. H. Chong, M. Imregun, Use of Reciprocal Modal Vectors for Nonlinearity Detection, Archive of Applied Mechanics 70 (2000) pp.453-462 [13] J. S. Bendat, A. G. Piersol, Engineering Applications of Correlation and Spectral Analysis, 2nd Ed., John Wiley and Sons, Inc., New York, 1993 [14] J. S. Bendat, A. G. Piersol, Random Data: Analysis and Measurement Procedures, 2nd Ed., John Wiley and Sons, Inc., New York, 1986 [15] D. L. Brown, Frequency Response Function Estimation Concepts, University of Cincinnati, 1996 82 Appendix 1: Derivation of Confidence Intervals for θˆ(ω ) (i.e., ∠Ĥ (ω ) ) Ĥ (ω ) and The estimate of H after nd averages is given by ˆ Hˆ (ω ) = Hˆ (ω ) e − jθ (ω ) The normalized error term is given by εr = [] σ φˆ where φˆ = Ĥ or φˆ = θˆ φ [ ] The error term for Ĥ is given by ε r Hˆ = [1 − γ ] 2 xy γ xy 2nd Confidence intervals are then given by ( ) ( φˆ 1 − C f σ u ≤ φ ≤ φˆ 1 + C f σ u ) where the C f constant is found by relating the desired confidence level to the area under the standard normal distribution curve (e.g., C f =2 for 95% confidence levels and C f =1.64 for 90% confidence levels). Also, σ u is the estimate of the confidence interval, not the estimate of the standard deviation of the magnitude or phase (according to Doebling and Farrar, though Bendat and Piersol do not seem to differentiate between the two from my reading of them). σ u is given below for Ĥ and θˆ . 83 [ ] σ u Hˆ = [1 − γ ] Hˆ = ε [Hˆ ]Hˆ 2 xy γ xy 2nd [ ] 1− γ 2 xy −1 ˆ σ u θ = sin γ 2nd xy [] ( [ ]) = sin −1 ε Hˆ Thus the 90% confidence levels for Ĥ and θˆ are given by ( [ ]) ( [ ]) and θˆ(1 − 1.64σ [θˆ]) ≤ θ ≤ θˆ(1 + 1.64σ [θˆ]) Hˆ 1 − 1.64σ u Hˆ ≤ H ≤ Hˆ 1 + 1.64σ u Hˆ u u Using the above equations, the 90% confidence limits are given below for various numbers of averages and coherence values. 84 Upper and Lower bounds on |H| (dB magnitude) and bounds on θ (degrees) using Bendat’s equations γxy2 \ nd 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 16 32 64 128 256 5.44 4.16 3.14 2.33 1.71 -17.70 60.4 -8.29 38.0 -4.96 25.8 -3.19 17.9 -2.13 12.6 3.97 2.98 2.21 1.62 1.18 -7.53 35.4 -4.58 24.2 -2.97 16.9 -1.99 11.8 -1.36 8.3 3.18 2.37 1.74 1.26 0.91 -5.08 26.3 -3.26 18.2 -2.17 12.8 -1.48 9.0 -1.02 6.4 2.64 1.95 1.42 1.03 0.74 -3.81 20.8 -2.51 14.5 -1.70 10.2 -1.17 7.2 -0.81 5.1 2.21 1.62 1.18 0.85 0.61 -2.97 16.9 -1.99 11.8 -1.36 8.3 -0.94 5.9 -0.65 4.2 1.85 1.34 0.97 0.70 0.50 -2.35 13.7 -1.59 9.6 -1.09 6.8 -0.76 4.8 -0.53 3.4 1.51 1.09 0.79 0.56 0.40 -1.83 10.9 -1.25 7.7 -0.87 5.4 -0.60 3.8 -0.42 2.7 1.18 0.85 0.61 0.43 0.31 -1.36 8.3 -0.94 5.9 -0.65 4.2 -0.46 2.9 -0.32 2.1 0.80 0.57 0.41 0.29 0.21 -0.88 5.5 -0.61 3.9 -0.43 2.8 -0.30 2.0 -0.21 1.4 85 Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) The same table taken from the HP literature is given below Upper and Lower bounds on |H| (dB magnitude) and bounds on θ (degrees) from HP’s literature γxy2 \ nd 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 16 32 64 128 256 5.20 3.80 2.80 2.10 1.50 -14.60 54.0 -7.10 34.0 -4.20 23.0 -2.70 16.0 -1.80 11.0 4.20 3.10 2.20 1.60 1.20 -8.40 38.0 -4.80 25.0 -3.00 17.0 -2.00 12.0 -1.40 8.0 3.50 2.60 1.80 1.30 1.00 -6.00 30.0 -3.60 20.0 -2.30 14.0 -1.60 10.0 -1.10 7.0 3.00 2.10 1.50 1.10 0.80 -4.50 24.0 -2.80 16.0 -1.90 11.0 -1.30 8.0 -0.90 5.0 2.50 1.80 1.30 0.90 0.70 -3.50 19.0 -2.20 13.0 -1.50 9.0 -1.00 6.0 -0.70 4.0 2.10 1.50 1.00 0.70 0.50 -2.70 15.0 -1.70 10.0 -1.20 7.0 -0.80 5.0 -0.60 4.0 1.60 1.10 0.80 0.60 0.40 -2.00 12.0 -1.30 8.0 -0.90 6.0 -0.60 4.0 -0.40 3.0 1.10 0.80 0.50 0.40 0.30 -1.30 8.0 -0.80 5.0 -0.60 4.0 -0.40 3.0 -0.30 2.0 86 Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) The differences between these two tables are highlighted below. Diffference Between the Tables compiled from Bendat’s equations and HP literature γxy2 \ nd 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 16 32 64 128 256 1.23 0.82 0.59 0.48 0.32 -7.07 18.6 -2.52 9.8 -1.23 6.1 -0.71 4.2 -0.44 2.7 1.02 0.73 0.46 0.34 0.29 -3.32 11.7 -1.54 6.8 -0.83 4.2 -0.52 3.0 -0.38 1.6 0.86 0.65 0.38 0.27 0.26 -2.19 9.2 -1.09 5.5 -0.60 3.8 -0.43 2.8 -0.29 1.9 0.79 0.48 0.32 0.25 0.19 -1.53 7.1 -0.81 4.2 -0.54 2.7 -0.36 2.1 -0.25 0.8 0.65 0.46 0.33 0.20 0.20 -1.15 5.3 -0.61 3.4 -0.41 2.2 -0.24 1.2 -0.17 0.6 0.59 0.41 0.21 0.14 0.10 -0.87 4.1 -0.45 2.3 -0.33 1.6 -0.20 1.2 -0.18 1.3 0.42 0.25 0.19 0.17 0.09 -0.64 3.7 -0.36 2.1 -0.25 1.8 -0.14 1.1 -0.08 0.9 0.30 0.23 0.09 0.11 0.09 -0.42 2.5 -0.19 1.1 -0.17 1.2 -0.10 1.0 -0.09 0.6 Hupper H θ (degrees) Hlower θ (degrees) 0 < |H| <.1 .1 < |H| <.2 0 < |θ| < .5 .5 < |θ| < 1 Hupper .2 < |H| <.3 1 < |θ| < 2 Hlower θ (degrees) .3 < |H| <.5 2 < |θ| Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Assuming the same process is used for deriving the two tables, the differences seem to arise because of a 2 term in Bendat’s equation for the error of Ĥ which is not used in the [ ] equations HP used to compile their table. If ε r Hˆ = error term, the following table is calculated. 87 [1 − γ ] 2 xy γ xy nd is used as the equation for the Upper and Lower bounds on |H| (dB magnitude) and bounds on θ (degrees) using Bendat’s equations (except [ ] is multiplied by ε r Hˆ γxy2 \ nd 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 2) 16 32 64 128 256 6.97 5.44 4.16 3.14 2.33 -17.70 60.4 -8.29 38.0 -4.96 25.8 -3.19 17.9 5.20 3.97 2.98 2.21 1.62 -14.89 55.1 -7.53 35.4 -4.58 24.2 -2.97 16.9 -1.99 11.8 4.22 3.18 2.37 1.74 1.26 -8.55 38.8 -5.08 26.3 -3.26 18.2 -2.17 12.8 -1.48 9.0 3.53 2.64 1.95 1.42 1.03 -6.06 30.1 -3.81 20.8 -2.51 14.5 -1.70 10.2 -1.17 7.2 2.98 2.21 1.62 1.18 0.85 -4.58 24.2 -2.97 16.9 -1.99 11.8 -1.36 8.3 -0.94 5.9 2.51 1.85 1.34 0.97 0.70 -3.54 19.6 -2.35 13.7 -1.59 9.6 -1.09 6.8 -0.76 4.8 2.07 1.51 1.09 0.79 0.56 -2.71 15.6 -1.83 10.9 -1.25 7.7 -0.87 5.4 -0.60 3.8 1.62 1.18 0.85 0.61 0.43 -1.99 11.8 -1.36 8.3 -0.94 5.9 -0.65 4.2 -0.46 2.9 1.11 0.80 0.57 0.41 0.29 -1.28 7.9 -0.88 5.5 -0.61 3.9 -0.43 2.8 -0.30 2.0 #NUM! #NUM! Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) Hupper Hlower θ (degrees) When this is used, the table matches quite nicely with the table compiled by HP. The differences are given below. As one can see, there are slight differences which tend to get larger as the error terms get larger. Overall, though, the tables match quite nicely. 88 Diffference Between the Tables compiled from modified Bendat’s equations and HP literature γxy2 \ nd 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 16 32 64 128 256 0.00 -0.17 -0.18 -0.11 -0.12 0.29 -1.1 0.43 -1.4 0.38 -1.2 0.27 -0.9 0.19 -0.8 -0.02 -0.08 -0.17 -0.14 -0.06 0.15 -0.8 0.28 -1.3 0.26 -1.2 0.17 -0.8 0.08 -1.0 -0.03 -0.04 -0.15 -0.12 -0.03 0.06 -0.1 0.21 -0.8 0.21 -0.5 0.10 -0.2 0.07 -0.2 0.02 -0.11 -0.12 -0.08 -0.05 0.08 -0.2 0.17 -0.9 0.09 -0.8 0.06 -0.3 0.04 -0.9 Hlower θ (degrees) -0.01 -0.05 -0.04 -0.07 0.00 Hupper 0.04 -0.6 0.15 -0.7 0.09 -0.6 0.09 -0.8 0.06 -0.8 Hlower θ (degrees) 0.03 -0.01 -0.09 -0.09 -0.06 0.01 -0.6 0.13 -0.9 0.05 -0.7 0.07 -0.4 0.00 0.2 -0.02 -0.08 -0.05 -0.01 -0.03 -0.01 0.2 0.06 -0.3 0.04 0.1 0.05 -0.2 0.06 0.1 Hlower θ (degrees) -0.01 0.00 -0.07 -0.01 0.01 Hupper -0.02 0.1 0.08 -0.5 0.01 0.1 0.03 0.2 0.00 0.0 Hlower θ (degrees) Hupper Hupper θ (degrees) Hlower θ (degrees) 0 < |H| <.1 .1 < |H| <.2 0 < |θ| < .5 .5 < |θ| < 1 Hupper .2 < |H| <.3 1 < |θ| < 2 Hlower θ (degrees) .3 < |H| <.5 2 < |θ| Hupper Hlower θ (degrees) Hupper Hupper Hlower θ (degrees) Hupper It seems from the brief overview above that HP is using a slightly different equation from Bendat. The origin of and reason for the 2 term in the equations used by HP is unknown at this point. 89
© Copyright 2026 Paperzz