Experimental Modal Analysis of Business Jet Fuselage Tail Section Sub-Assemblies

by

Ian Donaldson

A thesis submitted to the Department of Mechanical and Materials Engineering

In conformity with the requirements for

the degree of Master of Applied Science

Queen’s University

Kingston, Ontario, Canada

(October, 2020)

Copyright © Ian Donaldson, 2020

 Abstract

Experimental modal testing is a technique through which the dynamic response characteristics of

a system can be found. Parameters such as the natural frequencies and mode shapes of a system can be

extracted through experimentation, and these results can be used to confirm computational models, and/or

provide insights from which improvements can be made to improve dynamic response characteristics.

This thesis provides an overview of experimental modal analysis performed on two half scale aircraft

fuselage subassemblies using shaker excitation. Prior to experimentation, relevant literature was

reviewed. This provided the necessary foundation of theory and prior experiments which helped to form

the experimental methodology presented. This methodology including the construction of each structure,

data acquisition parameters, and validity checks, is covered in detail. Linearity and repeatability checks

were used to validate the testing methodology in accordance with ISO guidelines. Additional validity

checks were performed to improve the test setup and further increase the level of confidence in the

experimental results. The natural frequencies found were compared to the computational model, and

where necessary, recommendations for future modelling improvements were made. In the case of the first

sub-assembly, discrepancy in natural frequency for certain modes was reduced from 12 % to less than

2%. An acceptable level of correlation was found between experimental and computational results for the

second fuselage structure, with a discrepancy of less than 10 % observed for most natural frequencies.

Certain natural frequencies showed variance above 30 %, and potential reasons for this discrepancy are

discussed. The experimental results also provided a basis of comparison for the mode shapes predicted in

a computational model constructed by other researchers, acting to further validate computational results.

With a detailed experimental methodology formed, and reliable experimental results produced, the project

can now progress to the next phase which involves the construction and testing of a complete half scale

tail section.

ii
 Acknowledgements

First and foremost, I would like to extend my gratitude to my supervisor Dr. Chris Mechefske for

his continued technical guidance and support throughout this project. The completion of this work would

not be possible without his commitment.

To all my colleagues in the Systems Dynamics Research group, I would like to thank you for

your technical support and friendship over the past two years. A special thank you to Diego Chamberlain

for forming the groundwork of this project, and for providing a comprehensive introduction to

experimental modal analysis. Additionally, I would like to thank Braden Warwick and Chris Lam who in

addition to providing computational data, assisted me throughout the various phases of the project.

I would like to thank Stephen Colavincenzo and Reza Madjlesi of Bombardier Aerospace for

their continued financial and technical support. Additionally, for providing me the opportunity to work at

Bombardier, the experience from which will prove invaluable in the future.

Finally, I would like to thank my close friends and family for their unwavering support

throughout my time at Queen’s. To my partner, Madeleine, thank you for your continuous support

throughout my academic career. To my parents, thank you for everything. Your continuous support

throughout my academics and beyond is appreciated more than you know.

iii
 Table of Contents

Abstract ......................................................................................................................................................... ii
Acknowledgements ...................................................................................................................................... iii
List of Figures .............................................................................................................................................. vi
List of Tables ............................................................................................................................................... ix
List of Abbreviations .................................................................................................................................... x
Chapter 1 Introduction .................................................................................................................................. 1
1.1 Background ......................................................................................................................................... 1
1.2 Motivation ........................................................................................................................................... 2
1.3 Project Scope ...................................................................................................................................... 2
1.4 Thesis Outline ..................................................................................................................................... 4
Chapter 2 Literature Review ......................................................................................................................... 5
2.1 Literature Review................................................................................................................................ 5
2.2 Modal Analysis Theory..................................................................................................................... 13
2.3 Chapter Summary ............................................................................................................................. 25
Chapter 3 Fuselage Sub-Assembly Design and Construction..................................................................... 27
3.1 Overview ........................................................................................................................................... 27
3.2 Component Design Overview ........................................................................................................... 28
3.2.1 Skin Sections .............................................................................................................................. 29
3.2.2 Center Panels ............................................................................................................................. 33
3.2.3 Frame Sections ........................................................................................................................... 36
3.2.4 Stringers ..................................................................................................................................... 38
3.3 Sub-assembly Design and Construction ........................................................................................... 39
3.3.1 Sub-assembly 1 Version 1.......................................................................................................... 39
3.3.2 Sub-assembly 1 Version 2.......................................................................................................... 43
3.4 Sub-assembly 2 ................................................................................................................................. 44
3.5 Full Tail Section Assembly ............................................................................................................... 48
3.6 Chapter Summary ............................................................................................................................. 52
Chapter 4 Experimental Methodology ........................................................................................................ 53
4.1 Overview ........................................................................................................................................... 53
4.2 Boundary Conditions ........................................................................................................................ 53
4.3 Excitation Methodology.................................................................................................................... 58
4.4 Input and Output Measurements ....................................................................................................... 62

iv
 4.5 Data Acquisition ............................................................................................................................... 68
4.6 Signal Processing .............................................................................................................................. 69
4.7 Data Analysis .................................................................................................................................... 72
4.8 Chapter Summary ............................................................................................................................. 73
Chapter 5 Results and Correlation .............................................................................................................. 74
5.1 Overview ........................................................................................................................................... 74
5.2 Validity Checks................................................................................................................................. 74
5.2.1 Linearity Check .......................................................................................................................... 74
5.2.2 Repeatability Check ................................................................................................................... 79
5.2.3 Driving Point Measurement ....................................................................................................... 81
5.2.4 Mass Effects Study..................................................................................................................... 84
5.2.5 Response Measurement Placement Validation .......................................................................... 86
5.2.6 Stinger Length Test .................................................................................................................... 88
5.3 Correlation Overview........................................................................................................................ 89
5.4 Sub-assembly One Natural Frequency Correlation........................................................................... 90
5.5 Sub-assembly One Mode Shape Correlation .................................................................................... 91
5.6 Sub-Assembly Two Natural Frequency Correlation ......................................................................... 97
5.7 Sub-assembly Two Mode Shape Correlation.................................................................................... 99
5.8 Chapter Summary ........................................................................................................................... 104
Chapter 6 Conclusions and Future Recommendations ............................................................................. 106
6.1 Summary and Conclusions.............................................................................................................. 106
6.2 Future Recommendations ............................................................................................................... 109
Appendix A Sub-assembly One Experimental Mode shapes ................................................................... 110
Appendix B Sub-assembly Two Experimental Mode Shapes .................................................................. 116
References ................................................................................................................................................. 129

v
 List of Figures

Figure 2.1: Bode diagram for a SDOF dynamic system (Avitable, 2018). ................................................. 15
Figure 2.2: Co-Quad diagram containing real and imaginary parts of an FRF from a SDOF dynamic
system (Avitable, 2018). ............................................................................................................................. 16
Figure 2.3 A generic mobility plot for a MDOF system. (Ewins, 2000) .................................................... 23
Figure 2.4:Typical driving point measurement for sub-assembly 1. .......................................................... 24
Figure 3.1: Schematic for the flat pattern of a truncated cone used for the skin of the first sub-assembly. 29
Figure 3.2: First sub-assembly skin design (first iteration)......................................................................... 30
Figure 3.3: First subassembly skin flat pattern (final iteration). ................................................................. 31
Figure 3.4: First sub-assembly skin final design......................................................................................... 31
Figure 3.5: Schematic of the second sub-assembly skin, with pylon cut-out detail. .................................. 32
Figure 3.6: The result of bending the flat pattern in Figure 3.5 into the final conical shape. ..................... 33
Figure 3.7: Schematic of the first sub-assembly aft bulkhead. ................................................................... 34
Figure 3.8: Final schematic for the front engine support frame (FESF) including engine pylons. ............. 35
Figure 3.9:Final schematic for the rear engine support frame (RESF) including engine pylons. ............... 36
Figure 3.10: Universal cross section schematic used for all frame sections from both sub-assemblies. .... 37
Figure 3.11: Frame section 1066 from sub-assembly 1. ............................................................................. 38
Figure 3.12: The schematic of a stringer from sub-assembly 1. ................................................................. 39
Figure 3.13: A side view of the full tail section showing where sub-assembly 1 originates. ..................... 40
Figure 3.14: Isometric view of the full tail section showing where sub-assembly 1 originates. ................ 40
Figure 3.15: An image showing the completed first version of sub-assembly 1. ....................................... 42
Figure 3.16: An image highlighting the poor mechanical connection between sections of skin on the first
version of the first subassembly. ................................................................................................................. 42
Figure 3.17:An image highlighting the improvement in mechanical connection between sections of skin
on the first version of the first subassembly. .............................................................................................. 43
Figure 3.18: A side view of the full tail section showing where sub-assembly 2 originates ...................... 44
Figure 3.19: Isometric view of the full tail section showing where sub-assembly 2 originates. ................ 44
Figure 3.20: Initial Stages of sub-assembly 2 construction. In this image 3 of 4 frame sections including
both engine support frames have been attached to one panel of skin. ........................................................ 46
Figure 3.21: The second sub-assembly nearing completion. The slots in the skin form clearance holes for
the engine pylons which were added later. ................................................................................................. 47
Figure 3.22: The completed second sub-assembly. .................................................................................... 48

vi
Figure 3.23: The rectangle highlights the frame added to the front of the bulkhead to facilitate the
transition between cylindrical and conical skin sections. ........................................................................... 50
Figure 3.24: Side view of the completed design of the full tail section. ..................................................... 50
Figure 3.25: Schematic of the internal structure of the full tail section. ..................................................... 51
Figure 3.26: Completed design of the full tail section sub-assembly. ........................................................ 51
Figure 4.1: The boundary condition setup used for sub-assembly 1........................................................... 55
Figure 4.2: The support structure used for sub-assembly 2. ....................................................................... 57
Figure 4.3: Boundary condition setup for sub-assembly 2. ........................................................................ 57
Figure 4.4: Shaker setup during sub-assembly 2 modal testing .................................................................. 59
Figure 4.5: Typical drill rod stinger setup implemented during preliminary testing of sub-assembly 1. ... 62
Figure 4.6: PCB Piezotronics impedance head 288D01. ............................................................................ 63
Figure 4.7: Sub-assembly 1 bulkhead DOF placement. .............................................................................. 66
Figure 4.8: DOF placement for the FESF of sub-assembly 2. .................................................................... 67
Figure 4.9: DOF placement for the RESF of sub-assembly 2..................................................................... 67
Figure 4.10: Typical FRF and coherence measurement. Notice the dips in coherence in relation to the
anti-resonances of the FRF (Avitable, 2018). ............................................................................................. 71
Figure 5.1: Linearity check from DOF 2 on the bulkhead of sub-assembly 1. (Orange 2N, Blue 7N and
Green 15N peak input) ................................................................................................................................ 76
Figure 5.2: Linearity check from DOF 2 on the bulkhead of sub-assembly 1. (Orange 2N, Blue 7N and
Green 15N peak input) ................................................................................................................................ 77
Figure 5.3: Linearity check from DOF 44 on the FESF of sub-assembly 2. (Orange 3N, Blue 10N and
Green 20N Peak Input) ............................................................................................................................... 78
Figure 5.4: Linearity check from DOF 144 on the RESF of sub-assembly 2. (Orange 3N, Blue 10N and
Green 20N peak input) ................................................................................................................................ 79
Figure 5.5: Repeatability check of DOF 141 on the bulkhead of sub-assembly 1. ..................................... 80
Figure 5.6: Repeatability check of DOF 62 on the RESF of sub-assembly 2. ............................................ 81
Figure 5.7: Driving point measurement for sub-assembly 1 (Magnitude) .................................................. 82
Figure 5.8: Driving point measurement for sub-assembly 1 (Imaginary) ................................................... 83
Figure 5.9: Driving point measurement for sub-assembly 2 (Magnitude) .................................................. 83
Figure 5.10: Driving point measurement for sub-assembly 2 (Imaginary) ................................................. 84
Figure 5.11: Overlayed FRF curves for bulkhead of sub-assembly 1 (No Dummy Masses) ..................... 85
Figure 5.12: Overlayed FRF curves for bulkhead of sub-assembly 1 (With Dummy Masses) .................. 86
Figure 5.13: Aft Bulkhead AutoMAC before (Left) and after (Right) points were added. ........................ 87
Figure 5.14: Sub-assembly 1 AutoMac for all modes................................................................................. 87
vii
Figure 5.15: Sub-assembly 2 AutoMAC for the FESF (Left) and RESF (Right) ....................................... 88
Figure 5.16: Stinger length comparison for DOF 42 of sub-assembly 1. (Orange: 10.5 cm, Green: 15 cm,
Blue: 25 cm)................................................................................................................................................ 89
Figure 5.17: Repeated modes of axisymmetric structures (Ewins, 2000). .................................................. 92
Figure 5.18: “6 Point Star Skin” FEM mode shape (CMP 10 at 172.78 Hz) (Lam & Mechefske, 2020) .. 93
Figure 5.19: “6 Point Star Skin” Experimental mode shape (CMP 10 at 170 Hz) (Lam & Mechefske,
2020) ........................................................................................................................................................... 94
Figure 5.20: Experimental mode eight at 138 Hz (highlighted in yellow in Table 1) ................................ 95
Figure 5.21: Computational Modes eight and 9 (highlighted in yellow in Table 1) (Lam & Mechefske,
2020) ........................................................................................................................................................... 95
Figure 5.22: POC computation for 11 correlated modes of sub-assembly one........................................... 96
Figure 5.23: Third FESF experimental mode shape at 41.6 Hz ................................................................ 100
Figure 5.24: Third FESF computational mode shape at 30.24 Hz (Lam & Mechefske, 2020). ............... 100
Figure 5.25: Fifth RESF experimental mode shape at 89 Hz ................................................................... 101
Figure 5.26: Fifth RESF computational mode shape at 85.05 Hz (Lam & Mechefske, 2020). ................ 101
Figure 5.27: POC computation for 15 correlated modes from the FESF of sub-assembly 2.................... 102
Figure 5.28: Similarity between FESF mode 11 (Left) and FESF mode 12 (right) found experimentally.
.................................................................................................................................................................. 103
Figure 5.29: Predicted FESF modes 11 (Left) and 12 (Right) (Lam & Mechefske, 2020). ..................... 103
Figure 5.30: POC computation for 10 correlated modes from the RESF of sub-assembly 2. .................. 104

viii
 List of Tables

Table 5.1: Comparison of sub-assembly 1 natural frequencies .................................................................. 91

Table 5.2: Comparison of sub-assembly 2 natural frequencies .................................................................. 97

ix
 List of Abbreviations

Sound Pressure Level - SPL

Experimental Modal Analysis - EMA
Fast Fourier Transform - FFT
Electro-Optic Holography - EOH
Scanning Laser Doppler Vibrometer - SLDV
Finite Element Model - FEM
Multiple Input Multiple Output - MIMO
Ground Vibration Test - GVT
Computational Modal Analysis - CMA
Correlated Mode Pair - CMP
Single Degree of Freedom - SDOF
Frequency Response Function - FRF
Multiple Degrees of Freedom - MDOF
Degree of Freedom - DOF
Unit Modal Mass - UMM
Modal Assurance Criterion - MAC
Pseudo Orthogonality Check - POC
System Equivalent Reduction Expansion Process - SEREP
Computer Aided Machining - CAM
Computer Numerical Control - CNC
Engine Support Frame - ESF
Front Engine Support Frame - FESF
Rear Engine Support Frame - RESF
Rigid Body Mode - RBM
Complex Mode Indicator Function - CMIF
Multivariate Mode Indicator Function - MMIF

x
 Chapter 1

Introduction

1.1 Background
With regards to business jets, there is a significant premium placed on aircraft that

provide a luxurious flying experience. Noise in aircraft is an ever-present issue which negatively

affects the comfort and overall flying experience for passengers. Jet engines produce a significant

portion of the noise which is transmitted to the interior along airborne or structure borne paths

(Mixson & Kearny, 1978). This noise is generated from vibrations induced from mechanical

imbalances in the engines, and other more complex mechanisms such as exhaust interactions with

the outside of the plane. For aircraft with wing mounted engines the airborne transmittance of

sound is the primary concern, and for aircraft with fuselage mounted engines, structure borne

noise is more prevalent (Wilby, 1995). This research aims to reduce the sound pressure level

(SPL) in a range of business jets with fuselage mounted jet engines produced by Bombardier

Aerospace, by first obtaining an understanding of the dynamic characteristics of a generic

representation of the tail section of these aircraft. In future this understanding will guide structural

modifications to alter the dynamic response of the fuselage tail section and reduce interior cabin

noise.

Experimental modal analysis (EMA) is often used to understand the vibrational

characteristics of structures. Modal parameters found using EMA such as the natural frequencies

and mode shapes are often compared to a computational model for validation. Once it is shown

that the computational model can accurately represent the dynamic properties of a structure,

improvements can be proposed, tested computationally, and finally tested experimentally. These

improvements can benefit a system in a variety of ways, and for this project they aim to reduce

the dynamic response of the system resulting from an excitation from the primary engine tone of

1
117 Hz. This methodology significantly reduces the amount of physical testing required, and the

cost associated with making multiple structural modifications.

1.2 Motivation
The primary motivation behind this research is to improve passenger comfort for those

flying in the newest generation of business jet. These aircraft are designed according to multiple

design objectives including performance, fuel economy and passenger comfort. The new

generation of business jets offer faster cruising speed, and longer overall range; however they

struggle to provide a quiet environment associated with a luxurious flying experience. In this

highly competitive sector of the market, it is important to gain an advantage wherever possible.

Previous research has been performed to relate passenger discomfort to interior vibration and

sound pressure level. Dempsey et al. have shown that passenger discomfort is linearly related to

vibration acceleration level and varies logarithmically with interior sound pressure level

(Dempsey et al. 1978). To gain an advantage with respect to comfort, reducing the interior sound

is paramount. As such there is a strong motivation to understand the dynamic response of

business jet fuselage structures, such that improvements can be designed, validated, and

implemented. These improvements will aim to reduce noise with a minimal impact on other

design objectives. The first step, however, is to understand the dynamic response of a generic

aircraft fuselage to form a foundation of knowledge that will be used to guide these future

improvements.

1.3 Project Scope

A prominent Canadian aircraft manufacturer has shown interest in reducing noise and

vibration in the passenger cabin for a range of their business focused aircraft. To accomplish this,

the dynamic response of the fuselage of the aircraft must be understood, such that measures can

be implemented to mitigate structure borne noise and vibration. A significant portion of the

interior SPL was found to be generated by the fuselage mounted jet engines. This is in accordance
2
with research performed by Wilby which found that aircraft with fuselage mounted engines were

far more susceptible to structure borne noise propagation through the fuselage into the passenger

cabin (Wilby, 1995).

Interest has been focused on the tail section of the aircraft where both jet engines are

mounted, with previous research performed by Chamberlain on individual half scale fuselage

components. This research was focused on understanding the dynamic response of these

components using EMA and provided a source of validation for computational work performed

by Warwick (Chamberlain, 2018) (Warwick, Mechefske, & Kim, 2018). This work formed the

first stage of a bottom up approach which will be implemented across multiple research projects.

Chamberlain also devoted time to design a complete half scale fuselage tail section. This model

was designed to be generic in nature, such that the results from both the experimental test and

computational model would indicate general trends, and the findings from this half scale model

will be applicable to a wider range of fuselage tail sections.

This thesis aims to satisfy two primary objectives. The first of which is to provide

detailed and reliable experimental results for the purposes of computational model validation. The

second objective is to form a comprehensive experimental methodology which will be applicable

throughout the next phase of the project on a larger fuselage structure.

To satisfy these objectives, the research presented in this thesis focuses on expanding the

work performed on individual components, to half scale fuselage tail section sub-assemblies. It

will provide an intermediate step between the tests performed previously, and the full half scale

tail section experiments to be performed in future. In addition to his work modelling individual

components, Warwick et al. have also investigated these components as part of larger sub-

assemblies (Warwick, Kim, & Mechefske, 2019). As a result, experimental data was required to

form a basis of comparison to validate the computational models of fuselage sub-assemblies. To

satisfy this requirement, this thesis focuses on the experimental testing of two fuselage tail section

3
sub-assemblies. The research presented will help validate Warwick’s work and inspire updating

techniques to improve the accuracy of other computational models created by Lam et al. (Lam &

Mechefske, 2020). The experiments performed during this project will also form a testing

methodology suitable for EMA on aircraft fuselage sub-assemblies.

1.4 Thesis Outline

This thesis provides a complete overview of the research, experimental methodology, and

results associated with the experimental modal analysis of aircraft fuselage sub-assemblies. To

accomplish this, it has been divided into 6 chapters. Chapter 2 provides an overview of modal

analysis theory relevant to this project. Additionally, a review of previous research performed

with regards to modal analysis in aerospace is presented. Chapter 3 provides a detailed

description of both sub-assemblies. This includes an overview of their design, as well as an in-

depth description of the manufacturing methods chosen for the completion of both fuselage

sections. Finally, recommendations are made for future construction of the complete half-scale

tail section assembly. Chapter 4 describes the experimental methodology used to extract the

modal parameters from both sub-assemblies. The experimental setup, excitation method,

measurement location, and data acquisition parameters are covered in detail, as well as a detailed

account of the checks used to validate the experimental data. The results from these experiments

are presented in chapter 5. This chapter also discusses the correlation between the experimental

data and the computational predictions generated by Lam; whose work is cited where appropriate.

The natural frequencies as well as the mode shapes are compared, and discrepancies between the

two data sets are discussed. Future testing improvements, as well as a detailed description of

future work are outlined in chapter 6. This chapter also presents conclusions and a summary of

the work presented. Any figures deemed unsuitable for placement in the body of this thesis are

placed in the appendices and are referenced where appropriate.

4
 Chapter 2

Literature Review

2.1 Literature Review

The era of modern modal analysis can be dated back to the 1960’s, when the field began

to branch out from purely research applications into commercial avenues. This expansion was

predicated by developments such as the Fast Fourier Transform (FFT) developed by Cooley and

Tukey in 1963, as well as advances in sensors and data acquisition technology (Brown &

Allemang, 2007).These innovations helped make experimental modal analysis (EMA) accessible

to a wide range of industries looking to understand the dynamic characteristics of mechanical

systems. Soon after this transition into industry, EMA found its way into aerospace, and ever

since aircraft manufacturers have used experimental modal analysis to develop their product.

Many examples of such experiments can be found in the literature, and the following section will

provide a review of this content. Previous work has shown the applicability of EMA on a wide

range of aircraft fuselage components and sub-assemblies and modal analysis has also been

implemented as part of aircraft certification procedures in the form of ground vibration tests

(Goge, Boswald, Fullekrug, & Lubrina, 2007). EMA has also acted as a source of validation for

computational models and has inspired improvements in modelling techniques (Lam &

Mechefske, 2020). Additionally, measurement techniques used in EMA are constantly improving,

and these developments have been implemented in aerospace to integrate modal analysis

techniques as part of structural fault detection systems in aircraft.

An early implementation of EMA documented by Gordon et al. is used to investigate the

modal parameters of lightweight aircraft structures (Gordon, Wolfe, & Talmadge, 1977). The

primary objective behind this research was to obtain a more complete understanding of the

fatigue life of aerospace components due to dynamic loading under operating conditions. To

5
accomplish this, it was deemed necessary that the dynamic properties of the components in

question be understood, such that new failure modes and high stress locations could be

discovered. Another motivation of this project was to evaluate the effectiveness of a novel testing

method (for the time) which involved the use of an impact hammer, and a digital approach to data

analysis and manipulation. This research is documented as an early example of the use of digital

methods including a minicomputer which facilitated the use of the recently discovered Fast

Fourier Transform. As such, it was important to prove its viability through comparison with well-

established methods.

During the experiment, three aircraft panels were tested in two separate trials. The first

used a well-established analog testing methodology which employed a speaker for excitation

using a swept sine input signal, and accelerometers to measure the dynamic response. The digital

methodology used an impact hammer to supply a random excitation, and a minicomputer in

conjunction with an FFT algorithm to collect and transform the data into the frequency domain.

The results from both tests were compared to evaluate discrepancies between analog and digital

test methods, and modal parameters including the natural frequencies and modes shapes of each

panel were found. It was shown that the digital modal analysis provided consistent results with

those produced by the previously well-established analog techniques. Also noted by (Gordon,

Wolfe, & Talmadge, 1977). were the improvements in testing associated with the impact hammer

methodology. These included the vastly expedited set-up and test time, the reduction of hazards

associated with noise, and the increased flexibility of data storage and manipulation associated

with digital output. Limitations of the method are also indicated and include the limited range of

excitation, but improvements to the impact hammer technology, as well as developments with

modal shakers would remedy these issues.

Other experiments have been performed on stiffened aircraft panels by Fleming et al.

involving the extraction of modal parameters using electro-optic holography (EOH) as well as a

6
scanning laser doppler vibrometer (SLDV) on a full-scale fuselage panel. The panel constructed

was designed as a general representation of an aircraft fuselage skin section, with stringers and

frames attached to simulate stiffening structures present on aircraft. One of the main issues with

the EOH measurement method is the corruption of high frequency mode shapes from large

amplitude rigid body modes (Fleming & Buerhle, 1998). The development of image processing

algorithms to allow for a free-free boundary condition in the experimental setup is presented in

this work, as well as the validation of finite element models (FEM) using the experimental data

from both measurements.

One of the main focuses of this work was to improve the quality of interferogram output

from the EOH measurement. The initial measurements proved unsuccessful, as they did not

provide a clear image of the various mode shapes, with issues including poor contrast and

distortion caused by unrelated rigid body motion. To improve image quality, enhanced

interferograms were constructed using ensembles of 64 interferograms at each resonant mode

shape. An algorithm was then employed to guide this construction, using regions within the

ensemble having increased contrast and brightness. This method proved to be successful in

providing a clear image of each mode shape, while minimizing distortion from rigid body motion.

The enhanced results from the EOH measurements were compared to those generated by the

SLDV method, and the natural frequencies and mode shapes from both sets showed strong

correlation.

Modal parameters including the natural frequencies and mode shapes collected from both

experimental data sets were used to guide modifications to the FEM to improve correlation. Three

separate models using different element type configurations to construct the skin stiffeners were

evaluated. These configurations included using beam elements, plate elements, then a hybrid of

the two to account for varying stiffener geometry. The configuration that provided the best

correlation with the experimental results was the plate stiffener model, which most closely

7
followed the trends in natural frequency found in the experimental data. The computational model

still underpredicts natural frequency, and work is ongoing to improve correlation.

Another example of EMA being used to determine the modal characteristics of a

stiffened aircraft panel was documented by Wyen et al. (Wyen, Schoettelkotte, Perez, & Eason,

2017).This work formed a portion of a detailed testing methodology developed by the Air Force

Research Laboratory. The objective of this project was to improve the design process of

hypersonic aircraft components subject to complex loading conditions, and integral to this was

the determination of the dynamic properties of an aircraft fuselage panel under different operating

conditions. Two experimental approaches were used over four separate modal tests to evaluate

changes in modal parameters due to the addition of instrumentation and structural members. A

preliminary test utilized a roving impact hammer technique to find a preliminary set of natural

frequencies, mode shapes and damping values of the fuselage panel with structural members

called load introduction fittings. This test also provided a preliminary basis of comparison for a

FEM, and the discovered mode shapes in conjunction with the computational model helped guide

the placement of response measurement locations for future experiments. The second experiment

used three modal shakers to provide excitation, and laser doppler vibrometers to measure the

response of the same structure in three dimensions. The results from these tests were compared,

and minimal discrepancies were noted, indicating that both methodologies provided reasonable

results. The third test used the same multiple input multiple output (MIMO) methodology to test

the changes in modal characteristics resulting in the removal of load introduction fittings from the

fuselage panel. These fittings had a stiffening effect on the perimeter of the structure, and as

expected local modes near this part of the panel occurred at lower frequencies during this test.

The final test was conducted using a roving impact hammer on the fuselage panel with attached

load introduction fittings, as well as a full suite of other instrumentation necessary for other

experimentation involved with the project. This test was performed to provide an insight to the

8
effects of the attached instrumentation on the modal properties of the fuselage panel in operation.

It was found that the attached instrumentation increased the damping factor and decreased the

natural frequency of all modes. In summary, the work presented provides a detailed account of

two modal test methodologies used to determine the modal properties of a stiffened fuselage

panel. The experimentally determined properties were then used to update the computational

model, to improve its representation of the fuselage panel under operating conditions.

The applicability of EMA has also expanded outside the field of research and

development, into aircraft certification procedures in the form of ground vibration tests (GVT).

The primary purpose of a GVT is to identify “flutter” modes of an aircraft which can have

catastrophic structural effects on the aircraft during operation. “Aeroelastic flutter involves the

unfavorable interaction of aerodynamic, elastic, and inertia forces on structures to produce an

unstable oscillation that often results in structural failure.” (Kehoe, 1995)

To accomplish this modal shakers and hundreds of response transducers are often

required, resulting in a complex experimental procedure that can often take weeks to complete.

The results from these tests are then used to update the computational model, which once

validated, can predict flutter critical speeds. Therefore, the production of accurate results from the

GVT is essential to provide accurate predictions of flutter during flight. Recent developments in

the experimental procedure with regards to GVT’s are documented by (Goge, Boswald,

Fullekrug, & Lubrina, 2007). In addition to recommendations regarding implementation of

modern transducers (wireless accelerometers for example), a new process for the verification of

aeroelastic models was proposed to further expedite the experimental process. Procedural

modifications included testing critical components and substructures in parallel with aircraft

development. This would allow for the modal parameters of critical structures to be defined prior

to a full GVT. Additionally, the experimental results from these preliminary tests would provide

data for computational model validation. Preforming this work beforehand could allow for a

9
reduced GVT on the completed aircraft during the certification process, thus reducing overall

production time.

Changes in the dynamic characteristics of components can often indicate the presence of

a structural fault in aircraft, otherwise undetectable through other investigation techniques. Modal

analysis procedures are often used to detect faults which would manifest as a change in the modal

properties of a component or structure. The integration of accelerometers or other transducers

often used in EMA, is often not possible for structures in operation. A measurement system

which can be easily integrated into aircraft structures is desirable to accurately indicate changes in

structural integrity. (Cusano, et al., 2006) describe the integration of Bragg grating sensors into a

model wing for the purposes of EMA. The objective of their experiment is to compare the data

output from these fiber optic strain sensors with data output from sensors more conventionally

used in EMA to demonstrate the applicability of this measurement method.

To evaluate the validity of this new transducer arrangement, EMA was performed on two

identical composite wing structures. These wings formed approximations of what might be found

on unmanned aircraft. One was constructed to be tested using accelerometers, and the other

integrated fiber Bragg grating sensors within the main structural member of the wing. EMA was

performed on both wings over a 0-140Hz bandwidth, using a roving impact hammer for

excitation. As opposed to a free-free boundary condition, a rigid support was implemented on one

end of each wing to simulate the structural support provided by the fuselage of the aircraft. It was

found that both measurement techniques provided a complete set of mode shapes over the

frequency range of interest, and minimal discrepancies in the natural frequency of each mode

were found. As such, it has been shown that this “resident health monitoring system” provides a

complete and accurate set of results with respects to EMA. Next steps could include the

integration of this measurement system into a wide range of composite structures for damage

detection during operation using the core principles of EMA.

10
 Prior to the work presented in this thesis, EMA was conducted on multiple half scale

fuselage components, and this work is presented in detail by Chamberlain et al. at Queen’s

University (Chamberlain, 2018). The primary objective of this project was to understand the

dynamic response of fuselage structures, such that structural modifications could be proposed to

reduce interior sound pressure level on business aircraft. The work presented focuses on the

pressurized bulkhead and engine support frame panels. These components were selected due to

their proximity between the fuselage mounted jet engines and the passenger cabin, and their

direct effect on the structure borne noise propagation through the tail of the aircraft. During the

design process, general simplifications were made to enable the construction of such components

with the available resources. As a result, the experimental results would not provide an exact

representation of the dynamic properties of components on production aircraft. Instead general

trends in the dynamic response could be discovered, and in future design modifications will be

proposed and evaluated. Additionally, the experimental results provided in this study serve to

validate computational models and where necessary, provide guidance for model updating

techniques.

To complete this study, three pressurized bulkheads with different stiffener arrangements

and two engine support frames were designed and assembled using general aviation practices for

inspiration. Implementing different stiffener geometries on the pressurized bulkhead panels

provided an indication of the effects of bending stiffness on natural frequencies and mode shapes

of the panel. Creating three different bulkheads also provided a larger set of experimental results

to aid with model updating and validation. The engine support frames form the most direct

structural connection between the jet turbines and the fuselage, and as such, a detailed

understanding of their respective dynamic properties was required, before integrating them into a

full tail section model.

11
 To obtain the modal characteristics of each component, impact hammer testing was

implemented in accordance with modal testing best practices. Validity checks including

reciprocity, linearity, and repeatability checks were found to be essential for ensuring the

reliability of data output from an impact hammer modal test (Chamberlain, 2018). Other

guidelines found in ISO 7626 including the averaging procedures used during a modal test were

also recognized and followed rigorously (International Organization for Standardization, 1994).

The adherence to EMA best practices during this study provided high quality experimental results

and a reliable source of validation for the computational model.

A proposed outline of future work is also provided and describes potential next steps for

the project. These steps include expanding experimental modal testing on more complex fuselage

sub-assemblies to form an intermediate step between the individual components and the full tail

section model. Additional proposed steps involve investigating developments in test

methodology, including the implementation of a modal shaker for excitation, and expanding the

frequency bandwidth from ~0-90 Hz to 0-200 Hz. These recommendations provided a framework

on which the experimental methodology described in chapter 4 was created, and a defined

objective for the research presented in this thesis.

Although this research is not directly involved with computational modelling of dynamic

systems, the results presented provide a source of validation for such models. Recent work with

computational modal analysis (CMA) was performed by Warwick et al. to investigate the effects

of stiffener configurations on the modal properties of an aircraft pressurized bulkhead (Warwick,

Mechefske, & Kim, 2018). This paper outlines a methodology by which computational and

experimental results can be compared and correlated. This methodology includes the qualitative

correlation of mode pairs, the comparison of natural frequency of the correlated mode pairs

(CMP), and the quantification of correlation using the modal assurance criterion and pseudo

orthogonality checks. Using this methodology, the computational model was validated using

12
experimental results provided by (Chamberlain, 2018), and then developed to include operational

boundary conditions through the implementation of a frame structure around the bulkhead

perimeter. Both sets of results showed strong correlation and indicated that the stiffener

configuration had a significant effect on natural frequencies and modal density over certain

frequency bandwidths. Low frequency modes were more effected by the mass of added stiffeners,

while the frequency of modes higher in the bandwidth was increased as a result of the stiffness

from the added components. Finally, the author notes the importance of implementing this

validation methodology before progressing to more complex models stating “If the fundamental

nature of the component and the effect of basic modifications to the component are not well

understood, it is impossible to perform advance analysis techniques.” (Warwick, Kim, &

Mechefske, 2019)

2.2 Modal Analysis Theory

The following subsection aims to provide an overview of EMA theory relevant to the

research performed for the completion of this thesis. It will primarily focus on theory required to

perform a successful modal experiment, and where necessary will touch on the theory

surrounding dynamic systems. Two textbooks were referenced during the research phase of this

project. “Modal Testing- A Practitioner’s Guide” by Peter Avitable provides an in-depth

description of the procedures to follow for a wide range of modal tests, as well as a brief review

of EMA theory. “Modal Testing Theory Practice and Application” by David Ewins provides a

more in-depth review of the theory and is of special interest when considering the methodology

used to correlate experimental and computational results. Of course, only the necessary sections

of theory are presented below, and for a more complete review of EMA theory, the

aforementioned text should be reviewed.

To begin, the case of undamped forced vibration should be considered, as it provides a

simple mathematical approximation of a structure’s dynamic response to a certain input. As with

13
most terms in dynamic equations, the force input will be written in the harmonic form

f(t) = Feiωt as will the solution x(t) = Xeiωt . We find that the equation for the undamped

response of a structure subject to a harmonic input is represented by:

(−𝐦𝝎𝟐 + 𝐤)𝐗𝐞𝐢𝛚𝐭 = 𝐅𝐞𝐢𝛚𝐭 (𝟐. 𝟏)

Equation 2.1 for a single degree of freedom (SDOF) system can then be re-arranged to

form a ratio of the output response over the input force, which yields the fundamental form of the

frequency response function (FRF) for an undamped system.

𝑿 𝟏
𝑯(𝝎) = = (𝟐. 𝟐)
𝑭 (−𝐦𝝎𝟐 + 𝐤)

To represent a dynamic system more completely, a viscous damping term is added to this

formula in the form of a complex term in the denominator. This equation forms the general

formulation for the frequency response function of a SDOF dynamic system.

𝑋 1
𝐻(𝜔) = = (2.3)
𝐹 2
(−m𝜔 + k + i(ωc))

In its current form, Equation 2.3 represents the receptance of a structure, or the ratio of

the response displacement to the harmonic input force. This ratio is complex, and as such it

provides both magnitude and phase information. There are however other types of FRF’s which

will result from different measurement methods used during a modal test. The first of which is the

mobility FRF which is the ratio of response velocity to input force, and the common output of

SLDV measurements. The third FRF parameter known as inertance or accelerance is the ratio of

response acceleration to input force. This parameter is the output from modal tests using

accelerometers for response measurement. This was the method of choice for the experiments

presented in this thesis, and as such all experimental results presented are in the form of
𝑚⁄
𝑠2
acceleration / force, or .
𝑁

14
 Graphing these complex value functions is not a trivial process. There are three

parameters which must be accounted for, including the amplitude, frequency, and phase angle.

These values cannot be displayed on a standard x-y graph, and as such there are a multitude of

displays to choose from depending on the desired information. Of these, three formats were used

when performing data analysis for this project. The Bode plot provides magnitude vs frequency

along with phase angle vs frequency in two graphs oriented as shown in Figure 2.1. Notice the

phase change that occurs at resonance, as the inertial force surpasses the driving force resulting in

a loss of 1800. The bode diagram is the most common form of frequency response plot used

during the data analysis phase of EMA (Avitable, 2018).

Figure 2.1: Bode diagram for a SDOF dynamic system (Avitable, 2018).

Two other forms of FRF plot are commonly used during analysis and modal parameter

extraction. The real and imaginary components of the FRF are plotted against frequency in most

cases and provide a clear picture of the resonance peaks in relation to frequency. It is for this

15
reason that the imaginary part of the FRF was used during modal parameter extraction, a decision

which was inspired by work done previously by Chamberlain. (Chamberlain, 2018). The

coincident quadrature or “Co-Quad” plot provides information from the real (coincident) and

imaginary (quadrature) parts of the FRF in two graphs oriented in a similar fashion as the phase

and magnitude plots in the Bode diagram. An example of a SDOF Co-Quad plot is shown in

Figure 2.2. Note how the real part of the function goes to zero at resonance while the imaginary

part is a maximum. For this research, modal parameters were extracted using the diagrams

described in this section. Other diagrams for displaying EMA data including the Nyquist plot are

available, and these are outlined in both the referenced texts.

Figure 2.2: Co-Quad diagram containing real and imaginary parts of an FRF from a SDOF
dynamic system (Avitable, 2018).

16
 Equation 2.1 can also be expanded to represent multiple degrees of freedom (MDOF) by

implementing matrices to represent multiple mass and stiffness terms, along with vectors to

account for motion of multiple bodies. The matrices have dimension N X N, while the vectors are

N terms in length where N is the number of degrees of freedom (DOF) in the system. For

simplicity, an undamped free vibration case will be considered, in which case the force term is

zero, however if the external force on the system is known, the math remains largely the same.

(−[M]𝜔2 + [k]){X}eiωt = {F}eiωt = {0} (2.4)

There are two solutions to equation 2.4 for a dynamic system experiencing free vibration.

The “trivial” solution represents a system at rest, or in equation form {X} = 0. For the “non-

trivial” solution, Equation 2.5 can be used to find values of 𝜔 for each natural frequency of the

system.

det|−[𝑀]𝜔2 + [𝐾]| = 0 (2.5)

The determinant of this equation is a polynomial of order 2N and the roots of this

polynomial are the square of the system natural frequencies (𝜔𝑅 2 ). These values can then be

substituted into equation 2.4 to find the mode shape vector {𝜓𝑅 } which is equivalent to the {𝑋}

term in the equation. As [M] and [K] form N X N matrices, it can be seen that the number of

solutions to this equation is equivalent to the number of system DOF’s. It should also be noted

that while the vector of natural frequencies (eigenvalues) is unique, the matrix of mode shapes

(eigenvectors) is not, as the mode shapes are subject to some indeterminate scaling factor (Ewins,

2000). In other words, each natural frequency is a fixed value, but each mode can oscillate at any

amplitude while maintaining their characteristic shape.

An important property of linear modal analysis is the orthogonal nature of different mode

shape vectors. That is, each mode shape vector used to describe the motion of a MDOF dynamic

system will be orthogonal to one another. This property is explained mathematically by

17
manipulating the equation for free vibration of a MDOF system as shown in the following set of

equations. First the equation of motion is evaluated for a particular mode to find equation 2.6.

(−[M]𝜔𝑅 2 + [k]){𝜓}𝑅 = {0} (2.6)

where 𝜔𝑅 and {𝜓}𝑅 are the natural frequency and mode shape for mode R respectively. Then pre-

multiply this equation by the transpose of a different eigenvector to find equation 2.7.

{𝜓}𝑇𝑆 (−[M]𝜔𝑅 2 + [k]){𝜓}𝑅 = 0 (2.7)

For mode S the following equation can be written.

(−[M]𝜔𝑆 2 + [k]){𝜓}𝑆 = {0} (2.8)

This equation can then be transposed, and post multiplied by {𝜓}𝑅 .

{𝜓}𝑇𝑆 (−[M]𝑇 𝜔𝑅 2 + [k]𝑇 ){𝜓}𝑅 = 0 (2.9)

Since [M] and [k] are generally symmetric and therefore identical to their transposes, equations

2.7 and 2.8 can be combined to form equation 2.10a. It can be seen that this equation is only

satisfied under the condition shown in 2.10b.

(𝜔𝑅 2 − 𝜔𝑆 2 ){𝜓}𝑇𝑆 [M]{𝜓}𝑅 = 0 (2.10a)

{𝜓}𝑇𝑆 [M]{𝜓}𝑅 = 0; 𝑅 ≠ 𝑆 (2.10b)

For the case where R = S the following two equations apply; where 𝑚𝑟 and 𝑘𝑟 represent

the modal mass and stiffness respectively. These equations also represent a transformation from

the physical space to the modal space, which results in diagonal mass, damping and stiffness

matrices (Ewins, 2000). The same transformation can be performed on the damping matrix

assuming it is proportional to either the mass or stiffness matrices (ideal damping) as shown in

2.13 where 𝛽 is some proportionality constant.

{𝝍}𝑻 [𝑴]{𝝍} = [𝒎𝒓 ] (𝟐. 𝟏𝟏)

{𝜓}𝑇 [𝐾]{𝜓} = [𝑘𝑟 ] (2.12)

{𝜓}𝑇 [𝐶]{𝜓} = 𝛽[𝑘𝑟 ] (2.13)

18
 Since the mode shape vectors are subject to some indeterminant scaling factor, the modal

mass and stiffness are not unique values. In most cases the mode shape vectors are scaled such

that the largest element has a value of one (Ewins, 2000). Mass normalization is often used as a

scaling technique for normalization of the mode shape vectors, and was the technique

implemented when studying mode shape correlation. Mass normalized mode shape vectors are

denoted as {𝜑} and when substituted into equations 2.11 and 2.12 the results are as follows where

[I] denotes the identity matrix.

{𝜑}𝑇 [𝑀]{𝜑} = [𝐼] (2.14)

{𝜑}𝑇 [𝐾]{𝜑} = [𝜔𝑟2 ] (2.15)

The orthogonality property can also be used as a check to determine if two eigenvectors

are identical. This is done be comparing the transpose of one vector with the original form of

another to determine their orthogonality. The expected result from this check is shown in the

equations below, where unit modal mass (UMM) normalized modes shape vectors are compared.

{𝜑}𝑖 𝑇 [𝑀]{𝜑}𝑗 = 0 𝑓𝑜𝑟 𝑖 ≠ 𝑗 (2.16)

= 1 𝑓𝑜𝑟 𝑖 = 𝑗

{𝜑}𝑖 𝑇 [𝐾]{𝜑}𝑗 = 0 𝑓𝑜𝑟 𝑖 ≠ 𝑗 (2.17)

= [𝜔𝑟2 ]𝑖 𝑓𝑜𝑟 𝑖 = 𝑗

The principle of orthogonality is used to validate the mode shapes found both

computationally and experimentally. The modal assurance criterion (MAC) was implemented to

determine the confidence with which mode shapes could be experimentally determined as a

function of number of DOF’s. The equation used to compute the MAC is shown below.
2
|{𝜑}𝑇𝑖 {𝜑}𝑗 |
𝑀𝐴𝐶(𝑖, 𝑗) = (2.18)
({𝜑}𝑇𝑖 {𝜑}𝑖 )({𝜑}𝑇𝑗 {𝜑}𝑗 )

The ideal result for two sets of mode shapes would be the identity matrix, however during

experimental testing there are many sources of error which may contaminate the results. These

19
include non-linearities in the test structure, noise in the data and inappropriate selection of

measurement DOF’s (Ewins, 2000). This third issue is of most interest for the data presented, as

the MAC was primarily used to determine if the placement of DOF’s was sufficient during

testing. This was done by performing an AutoMAC computation, in which an identical set of

experimental mode shape vectors are input into equation 2.18. The MAC was also useful for

providing an indication of the level of correlation between the experimental and computational

mode shapes; however it does not confirm correlation between two sets of eigenvectors. Ewins

introduces many improved versions of the MAC, but the method used to compare two different

sets of mode shape vectors was the pseudo orthogonality check (POC) (Ewins, 2000). The POC

confirms orthogonality by comparing the transpose of the UMM normalized computational mode
𝑇
shape vectors [𝜑]𝐴 against the UMM normalized experimental mode shape vectors [𝜑]𝐸 . Once

again, the expected result from this computation is the identity matrix, however it is expected that

sources of error will be present in both the computational predictions and experimental data. It is

generally accepted that diagonal terms above 0.8 and off diagonal terms below 0.2 indicate that

two sets of mode shape vectors show acceptable correlation and having diagonal terms above 0.9

and off diagonal terms bellow 0.1 would indicate a strong level of correlation (Ewins, 2000).

{𝜑}𝐴 𝑇 [𝑀]{𝜑}𝐸 = [𝐼] (2.19)

Experimental mode shape vectors contain one DOF for each measurement location on the

test structure. The placement of measurement locations is guided by multiple factors including

preliminary checks such as the AutoMAC described earlier. Using this method ensures that each

mode shape in the bandwidth of interest can be properly characterized. Test feasibility is another

factor that must be considered to avoid unreasonably long testing periods which can introduce

experimental error. As such, there is a practical limit to the number of degrees of freedom used

during an experimental test. The number of DOF’s in a computational model is determined by the

size of the mesh, with 6 degrees of freedom used to represent each node in the model. As a result,

20
the size of the mass matrix and computational mode shape vector are often many orders of

magnitude larger than the size of the experimental mode shape vectors. This discrepancy in size

must be addressed before the calculation of the POC or MAC, because both mode shape vectors

as well as the mass matrix must have the same dimension to perform these calculations. That is, if

the experimental mode shape vector has N terms corresponding to the number of DOF’s, then the

analytical mode shape vector must have N terms and the mass matrix must have size N x N.

Reducing the analytical mode shape vector is simply a matter of selecting the nodes in the model

that most closely match the experimental measurement locations, and removing the rest from the

vector. The reduction of the mass matrix for the POC calculation was performed using the system

equivalent reduction expansion process (SEREP) by Lam. This method and others are

documented by Warwick when discussing the reduction of computational mode shape vectors and

mass matrices in his work modelling the modal response of a half scale aircraft tail section sub-

assembly. “The SEREP provides an exact means of mapping between the analytical DOFs and

the experimentally active DOFs.” (Warwick, Kim, & Mechefske, 2019)

The principle of orthogonality discussed above indicates that the response of a structure

can be expressed as a linear combination each eigenvector. This principle is known as the modal

expansion theorem as shown in 2.20 below, where {𝜓}𝑅 is the mode shape vector for mode R and

𝑞𝑅 (𝑡) is the modal amplitude for mode R.

{𝑥(𝑡)} = {𝜓}1 𝑞1 (𝑡)+{𝜓}2 𝑞2 (𝑡)+…{𝜓}𝑛 𝑞𝑛 (𝑡) (2.20)

As can be determined from this equation, the contribution of each mode will depend on

the mode shape vector and the modal amplitude of each mode. For example, if the system is

excited at or near a certain resonant frequency {𝜔𝑅 }, then the mode shape of the system will be

dominated by the mode shape vector for mode R. Generally, for experimental modal analysis

wide frequency ranges are excited, such that the overall response represents a summation of the

mode shapes within the bandwidth of interest. This is achieved through selection of test

21
parameters such as hammer tips for impact testing, or excitation signal generation in the case

where modal shakers are used. There are some situations in which not all modes in a bandwidth

are excited, and these are discussed below.

To perform a successful modal experiment, the relationship between the modal

parameters of a system and the resulting FRF’s must be understood. The following set of

equations provide a mathematical basis which guide the placement of response measurement

locations, and facilitate the understanding of FRF functions produced during EMA. Equation 2.21

forms the fundamental expression of an FRF and indicates that modal properties can be found by

evaluating the response characteristics of a system (Ewins, 2000).

𝑁
𝑋𝐽 𝐴𝑅𝑗𝑘
𝐻𝑗𝑘 (𝜔) = =∑ 2 (2.21)
𝐹𝐾 𝜆𝑅 − 𝜔 2
𝑅=1

The numerator (𝐴𝑅𝑗𝑘 ) represents the residue of mode R at DOF J with excitation at DOF

K. It is constructed of the mode shape vector for mode R {𝜑}𝑅 where {𝜑𝐽 }𝑅 is the relative

displacement of point J as a result of resonance at the Rth natural frequency. 𝜆2𝑅 is the eigenvalue

for mode R and contains both the natural frequency and damping ratio. The residue for mode R

can also be defined by the following equation when considering UMM normalized eigenvectors.

𝐴𝑅𝑗𝑘 = {𝜑𝐽 }𝑅 {𝜑𝐾 }𝑅 (2.22)

It is this equation which carries an important implication for EMA as clearly the residue

is directly dependent on the eigenvector value for both point J and K. The magnitude of the FRF

is therefore dependent on the input and output DOF’s, so while FRF peaks will align in terms of

natural frequency, their magnitude will vary. This phenomenon has a direct impact on the

selection of excitation and response locations. If for example one was to place the point of

excitation on a node of a certain mode A, then {𝜑𝐾 }𝐴 would be a vector with all terms equal to

zero. Therefore, regardless of response measurement location no resonance peak will occur, and

any response at this natural frequency will be entirely dictated by off resonance contributions

22
from nearby modes. As such careful consideration must be taken when selecting the excitation

location prior to a modal experiment, such that the driving point is not placed on a node of any

modes. There are different methods available to predict the expected mode shapes of a structure

prior to performing a modal test, and the method used for this experiment will be discussed in

Chapter 4.

The residues also provide information about the overall shape of an FRF. In general, the

FRF curve will represent a summation of the individual terms in the FRF series as per the modal

expansion theorem. At or near each resonant peak it is found that the total FRF curve very closely

matches each resonant peak, however discrepancies are found in the areas in between each peak.

This phenomenon is shown in Figure 2.3 and is explained by the phase relationship between two

resonant peaks in an FRF.

Figure 2.3 A generic mobility plot for a MDOF system. (Ewins, 2000)

The mathematical reasoning behind this phenomenon is discussed in detail by Ewins and

can be reduced to some generic statements. If the residues of two consecutive modes have the

same sign, then the response at some frequency between them will go close to zero and an anti-

23
resonance is formed. Between two modes of opposite sign, a minima will be found (Ewins,

2000). This becomes important when considering a driving point measurement, for which the

excitation and response locations are the same. In this scenario, it is expected that the residues

will have the same sign throughout the frequency range of the FRF, and as such it is expected that

an antiresonance will be found between every resonance peak. This is in fact the main criteria

used to evaluate the experimental setup and response linearity during a driving point validity

check. A typical driving point measurement is shown in Figure 2.4 below.

Figure 2.4:Typical driving point measurement for sub-assembly 1.

To successfully extract the modal parameters of a structure, the concept of the frequency

response function matrix must be understood. This matrix is composed of a series of FRF

functions each representing a potential measurement pair of excitation and response measurement

locations. As such the size of the FRF matrix is determined during the planning phase of the test,

where the number of DOF’s required to form a complete set of eigenvalues and eigenvectors is

determined. The procedure by which the arrangement of measurement locations is chosen is

discussed in Chapter 4.

24
 One column or one row of the FRF matrix must be completed to provide enough

information regarding the natural frequencies and mode shapes of the system. Each row of the

FRF matrix represents a measurement during which the response measurement location is fixed,

and the excitation location is roved over the entire set of DOF’s. A column represents the inverse,

where the excitation location is kept constant, and the response transducer is roved over the

structure. The diagonal terms in the matrix represent the driving point measurement where the

excitation and measurement locations are the same. For modal tests using a shaker for excitation,

it is far more practical to complete a column of the FRF matrix, and this was done during the

modal tests presented. The two methods stated are interchangeable due to the principle of

reciprocity, which states that the corresponding rows and columns of the FRF matrix will present

the same FRFs. A check to confirm the reciprocity of the structure can be performed by first

exciting the structure at some DOF K, and taking a response measurement at some point J. Then

reverse this arrangement and repeat the measurement, which should produce an identical FRF to

that found in the first phase of the test. This test is important to perform because according to

Avitable “reciprocity is basically the reason why only one row or one column of the frequency

response matrix needs to be measured.” (Avitable, 2018)

2.3 Chapter Summary

This chapter presents a detailed review of previous work done in the field of modal

analysis, with a focus on the aerospace industry. This review spans from fundamental testing

methods implemented by Gordon et al. to better understand the fatigue life of aerospace

components, to a novel integration of fiber bragg grating sensors into the composite structure of a

wing to detect changes in dynamic properties (Gordon, Wolfe, & Talmadge, 1977). Additionally,

a review of research performed by Chamberlain et al. was presented to explain the first step of

this project prior to the research presented in this thesis (Chamberlain, 2018). Finally, the work

25
completed by Warwick et al. was presented as a basis for the correlation procedures which were

implemented for this research (Warwick, Kim, & Mechefske, 2019).

In addition to a review of the literature, a summary of the relevant background theory was

presented. Starting with a mathematical representation of simple SDOF dynamic system,

concepts including the frequency response function and correlation methods for MDOF systems

are presented. It is necessary to understand these concepts in order to complete a successful

modal test, however there is an abundance of extra material which can be found in the referenced

text (Avitable, 2018) (Ewins, 2000). It is now appropriate to move forward to describe the

construction of both fuselage sub-assemblies in Chapter 3, and the experimental methodology

used to test them in Chapter 4.

26
 Chapter 3

Fuselage Sub-Assembly Design and Construction

3.1 Overview
This chapter outlines the design and construction of the two-half scale tail section

fuselage sub-assemblies which form an intermediate step between the individual components and

the full tail section model. It will first provide a summary of the design process implemented on

the full half scale fuselage tail section by Chamberlain. This summary includes a detailed

description of the individual components of the tail section, as well as their functional purpose as

part of the larger structure. Many of the individual components were constructed previously for

testing by Chamberlain, and these structures and the process used to construct them are covered

in detail in (Chamberlain, 2018). This chapter will focus primarily on the integration of these

components into larger sub-assemblies for testing. Multiple challenges were overcome during

construction using non-intuitive solutions, which should be implemented during the construction

of the full tail section. These construction techniques as well as future procedural improvements

will also be covered in detail.

Simplifications were made during the design and assembly of the tail section sub-

assemblies to facilitate construction with the available resources. The other primary difference

between the model and the production aircraft tail section, was the difference in scale. The tail

section model was designed to be a half scale interpretation, due to the amount of laboratory

testing space available. This choice, as well as the other simplifications, were made in accordance

with advice provided by industry experts from the aircraft manufacturer partnered with this

project. It is acknowledged that the model does not exactly represent the production aircraft,

however the objective of this project is primarily to observe trends, such that design

improvements could be recommended, and modelled with confidence.

27
 The following subsections detail the design and construction of two tail section sub-

assemblies. Two iterations of the first sub-assembly were constructed due to an issue with the

first version found during preliminary tests. This assembly was initially constructed with three

skin sections that were joined only to the central frame. This lack of structural connection

between skin sections had a significant effect on the dynamic response of the skin during modal

tests, so a modification was required. For the second version, the mechanical connection between

skin sections was improved, and so was the response of the skin during tests. The construction of

the second sub-assembly proved to be a significant challenge, partly due to the scale of the

structure, but largely due to a lack of knowledge regarding techniques that would prove to be

invaluable for the assembly of this structure. The process of manufacturing this second sub-

assembly is covered in detail, as the knowledge obtained from this process will help to expedite

the assembly of the full tail section model. Finally, recommendations will be made for additional

manufacturing improvements that will improve the construction process, and the quality of the

finished product.

3.2 Component Design Overview

Both sub-assemblies were constructed from a set of components including frames,

stringers, center bulkhead panel(s), and sections of skin. The skin sections and center panels

including both engine support structures and the rear bulkhead, were cut from large 1.6 mm thick

panels of 6061-T6 aluminum. To cut out the required shape, a waterjet was used in conjunction

with a computer aided machining (CAM) software. The frame sections were first formed by

rolling an initially straight section of 6061-T6 aluminum “C” channel to the desired radius. The

necessary cut-outs and holes were cut using a computer numerical control (CNC) mill and the

same CAM software. Finally, the stringers were cut from 6061-T6 “T” channel using the same

technique used for the frame sections, with no rolling required. Specific details regarding the

design process and specific manufacturing challenges for each component are included in

28
subsections 3.2.1-3.2.4 below. To reference their functionality in each sub-assembly, refer to

section 3.3.1-3.4, which describe the completed fuselage assemblies, and their similarities and

differences with production jet fuselage sections.

3.2.1 Skin Sections

To create the truncated conical shape of the aircraft fuselage skin, a flat sheet must be cut

into a shape analogous to the one shown in Figure 3.1, which shows a template for the entire skin

section of the first sub-assembly. This shape was created using multiple functions built into

SolidWorks, however these values can also be determined using a cone calculator, which can be

found online. The values shown below are necessary to create a cone with a large diameter of

2204.8 mm, a small diameter of 1952.73mm, and an inclination angle of 7.5 degrees. The

resulting cone is shown in Figure 3.2, which also illustrates the initial design for the first sub-

assembly skin.

Figure 3.1: Schematic for the flat pattern of a truncated cone used for the skin of the first
sub-assembly.

29
 Figure 3.2: First sub-assembly skin design (first iteration).

After the necessary geometry for the skin panels was established, future iterations were

created which accounted for stringer mounting, as well as available cutting space on the waterjet.

To fit on the waterjet, the skin was divided into three identical sections displayed in Figures 3.3

and 3.4 which show the final design for the first subassembly skin. Mounting holes for the

stringers were also included in the final design, however to save cutting time on the waterjet,

these holes were only pierced then later drilled to final dimension by hand. Another intermediate

iteration was created which did not have the sections of skin connect along a stringer, and this

iteration is discussed in more detail in section 3.3.1.

30
 Figure 3.3: First subassembly skin flat pattern (final iteration).

Figure 3.4: First sub-assembly skin final design.

The skin from the second sub-assembly was manufactured using an identical method to

that employed on the first sub-assembly. However, this structure was far larger, and as a result the

skin was divided into four sections to accommodate the limitations of the waterjet. Additional
31
consideration was required when designing the cut-outs for the engine support pylons which

protrude through the skin, and the design of one of these cut-outs is highlighted in Figure 3.5. In

the flat pattern slots with curved edges of certain radii are required. The simplest way to design

these features was to connect lines A and B at a point, then use this point as the center location

for two circles which have their circumference form the two curved edges of the slot. This will

ensure that when the skin section is bent during construction, the two slots form perfect rectangles

to accommodate the pylons. It should also be noted that unlike sub-assembly one, these skin

sections are not identical due to the placement of the pylon cut-outs, however the design process

for the other sections of skin was identical.

Figure 3.5: Schematic of the second sub-assembly skin, with pylon cut-out detail.

32
 Figure 3.6: The result of bending the flat pattern in Figure 3.5 into the final conical shape.

3.2.2 Center Panels

Bulkhead structures are added to an aircraft fuselage to improve the structural integrity in

certain areas. These structures are often found near the wings, and in the tail section of most

aircraft. Therefore, to accurately represent the tail section of a business jet, these center panels

must be evaluated as part of fuselage sub-assemblies. The center panels from both sub-assemblies

were previously designed and manufactured by Chamberlain, but some small modifications were

required to integrate these components into their respective sub-assemblies. The following

subsection briefly summarizes the design and manufacturing of each component prior to

integration in their respective assembly. For a more detailed description regarding the design

process of these components refer to (Chamberlain, 2018).

The first sub-assembly contained one central panel representative of the aircraft’s aft

bulkhead. This panel forms the final structural member at the back of the plane; not to be

33
confused with the pressurized bulkhead which separates the passenger cabin and the

unpressurized tail section. The aft bulkhead was cut from a sheet of 1.6 mm thick aluminum using

an identical method to that used for the skin panels. This bulkhead includes the necessary holes

for riveting to its corresponding frame, and semicircular cut-outs which are necessary for

clearance with the stringers. The final design of the aft bulkhead is shown in Figure 3.7.

Figure 3.7: Schematic of the first sub-assembly aft bulkhead.

The second sub-assembly forms the section of fuselage tail responsible for supporting

both jet engines. As such, both the front and rear engine support frames (ESFs) are included in

this assembly. The engine support frames are panels which provide an improved structural

connection between the engine pylons and the fuselage, by distributing the engine loads around

their corresponding fuselage frame. As with the aft bulkhead, holes are required to rivet each ESF

34
to its corresponding frame and cut-outs are required for stringer clearance. In addition, holes were

required to facilitate the bolted connection with the engine pylons, and crawl space cut-outs were

added to better represent the components present on the full-scale aircraft. These semicircular cut-

outs are made in the lower half of both ESFs to provide access to the rear section of the plane.

The final design for both ESFs including their corresponding pylons are presented in Figures 3.8

and 3.9.

Figure 3.8: Final schematic for the front engine support frame (FESF) including engine
pylons.

35
 Figure 3.9:Final schematic for the rear engine support frame (RESF) including engine
pylons.

3.2.3 Frame Sections

Aircraft frames form essential structural members within the fuselage of aircraft. They

add rigidity to the skin of the aircraft and provide structurally sound mounting locations for

components including the jet engines, wings, and bulkheads. On operational aircraft, geometry of

the frames varies depending on the loading condition imposed at their location in the fuselage.

This is done primarily to optimize the structural characteristics of the plane relative to other

factors such as weight. For simplicity though, the half scale frames constructed for both sub-

assemblies were of constant cross section. This allowed the frames to be rolled using one set of

dies reducing cost significantly, and this simplification was verified through consultation with

industry partners. The frames were rolled from 6061-T6 aluminum “C-Channel” to the required

radius, and during this process an inclination angle of 7.50 was imposed on the outer face to

accommodate the conical shape of the fuselage tail section. The final design for the frame cross

section is shown in Figure 3.10.

36
 Figure 3.10: Universal cross section schematic used for all frame sections from both sub-
assemblies.
The first step in creating each frame section is to transform the originally straight “C-

Channel” into a curved section of the required diameter, and by using custom dies during this

cold rolling process the 7.50 inclination angle can also be formed. Due to manufacturing

restrictions, each fuselage frame was formed out of two identical 1800 sections, as it was

impossible to create a complete ring using this technique. After being formed to the necessary

shape, the stringer clearance cut-outs and bulkhead mounting holes were cut using a CNC mill to

match up with the corresponding bulkhead panels. A schematic for a frame section from sub-

assembly 1 (FS-1066) is shown in Figure 3.11. The frame sections for sub-assembly 2 shared the

same pattern for rivet holes and stringer clearance cut-outs which was expanded to fit their larger

radius.

37
 Figure 3.11: Frame section 1066 from sub-assembly 1.

3.2.4 Stringers

Stringers in aircraft are implemented to improve the longitudinal rigidity of the fuselage

structure. Much like the fuselage frames, there is a range of different geometries used in aircraft

dependent on load conditions, available clearance, and other factors. For both sub-assemblies, one

uniform cross section was selected to expedite manufacturing. As with the fuselage frame

sections, this cross section was inspired by structures on full scale aircraft and verified through

consultation with industry partners. The choice of cross section was also partially motivated by

what aluminum extrusion was commercially available, and because this “T-Channel” was mass

produced externally, costs and manufacturing time were reduced. Holes were drilled in the

38
stringers to facilitate mounting to their respective skin sections. The spacing of these holes was

chosen based on two factors; the general spacing used in aviation, and what was deemed feasible

with the available resources. Ideally, the spacing between rivets for the half scale models would

be half of what is found on full scale aircraft; however this would result in an unreasonable

number of rivets for either sub-assembly. After consulting with industry experts, the space

between rivets was approximately doubled to expedite the manufacturing process. The final

design for a stringer from the first sub-assembly is shown in Figure 3.12. The stringers for the

second sub-assembly were 1112.3mm long with a hole spacing of 25.7mm.

Figure 3.12: The schematic of a stringer from sub-assembly 1.

3.3 Sub-assembly Design and Construction

3.3.1 Sub-assembly 1 Version 1

The first sub-assembly was used to form a general testing methodology suitable for

testing aircraft fuselage structures. It was necessary that this methodology was applicable to

evaluate the modal properties of major components within different fuselage sections. This first

structure would also be used to verify computational modelling techniques through the

comparison of natural frequencies and mode shapes. As such, it was important that this first sub-

assembly was made up of a central bulkhead structure as well as a section of skin and stringers.

This provided a complete set of results which were used to evaluate the testing methodology and

correlation with the computational predictions. Additionally, due to the trial and error approach

39
used to form certain aspects of the methodology, a small section of the fuselage was desired to

expedite the process. For these reasons, the aft most section of the half scale tail section model

was selected for preliminary experimentation. The first sub-assembly’s place within the full tail

section model is highlighted in Figures 3.13 and 3.14 below.

Figure 3.13: A side view of the full tail section showing where sub-assembly 1 originates.

Figure 3.14: Isometric view of the full tail section showing where sub-assembly 1 originates.

40
 The first sub-assembly consists of the aft bulkhead, its corresponding frame, and three

one-foot long sections of skin and stringers. To fasten these components together, rivets were

used in accordance with general aviation practices. To reduce drag, flush rivets are the primary

fasteners used on production aircraft, however the installation of these rivets is time consuming.

Using blind rivets allows for quicker assembly, resulting in a more feasible project timeline. The

aerodynamic properties of each sub-assembly are not of interest, and after consultation with

industry experts, any structural differences between blind and flush rivets were found to be

outside the scope of this project. Therefore, blind “pop” rivets of two different sizes were used to

fasten these components together. To hold the aft bulkhead to the fuselage frame (FS 1066),

3/16” rivets were used, and to fasten the stringers to the skin, 1/8” rivets were chosen. The

completed assembly is shown in Figure 3.15 and Figure 3.16, and Figure 3.16 highlights an issue

with this version found during initial modal tests. Each skin section was connected to the central

frame, but the panels were not connected to each other, and this lack of mechanical connection

resulted in poor energy distribution around the structure during modal tests. Additionally, the

results from this test were not representative of the modal characteristics of general aircraft

assemblies which have strong mechanical connections between each section of skin. As such a

revision to the design of the skin panels was made, such that they connected along a stringer. This

design formed the second iteration of the first sub-assembly which is described in section 3.3.2.

41
 Figure 3.15: An image showing the completed first version of sub-assembly 1.

Figure 3.16: An image highlighting the poor mechanical connection between sections of skin
on the first version of the first subassembly.
42
3.3.2 Sub-assembly 1 Version 2
The second version of the first sub-assembly used the same frame sections, bulkhead and

stringers as the first iteration, and three new sections of skin. These sections of skin were

designed to facilitate connection between panels along a stringer as highlighted in Figure 3.17.

This modification greatly improved the structural integrity of the skin, and while not a perfect

representation of the lap joints or other methods employed on production aircraft, it provided a

reasonable representation of what is found in industry. The eye bolts in Figure 3.17 were used to

suspend the assembly for modal tests. More details regarding the boundary conditions used can

be found in Chapter 4.

Figure 3.17:An image highlighting the improvement in mechanical connection between

sections of skin on the first version of the first subassembly.
43
3.4 Sub-assembly 2

Figure 3.18: A side view of the full tail section showing where sub-assembly 2 originates

Figure 3.19: Isometric view of the full tail section showing where sub-assembly 2 originates.

Sub-assembly two models the section of fuselage responsible for supporting both jet

engines. Previously, the modal properties of both ESF’s were extracted by Chamberlain, and this

structure was used to evaluate changes in modal properties of the ESF’s as part of a fuselage

structure (Chamberlain, 2018). Additionally, the results from testing this structure will serve to

validate computational models and/or guide model updating where necessary.

44
 This sub-assembly includes the front and rear engine support frame panels, and four

structural frames; two of which facilitate the connection between the skin and engine support

frames. Four skin panels were required to wrap around the structure, and these were connected

along stringers to form a sound structural connection. The engine pylons are bolted directly to the

engine support frames and protrude through the fuselage skin. On production aircraft, fastening

the pylons in a similar manner distributes the loads generated by each engine around the entire

fuselage. To support the significant loads produced by the turbines in flight, significant resources

are devoted to developing these components, however, to expedite the manufacturing process

with the available resources, a simpler design was selected. All design simplifications were made

in consultation with industry partners, and because this project was directed at finding general

trends of dynamic response in fuselage structures, more detailed individual component design

was deemed outside the scope of this research. Figures 3.20, 3.21 and 3.22 show the second sub-

assembly throughout the construction process and provide a look at the internal construction of

the sub-assembly.

45
 Figure 3.20: Initial Stages of sub-assembly 2 construction. In this image 3 of 4 frame
sections including both engine support frames have been attached to one panel of skin.

46
Figure 3.21: The second sub-assembly nearing completion. The slots in the skin form
clearance holes for the engine pylons which were added later.

47
 Figure 3.22: The completed second sub-assembly.

3.5 Full Tail Section Assembly

The next stage of this project is to evaluate the modal properties of the complete half

scale fuselage tail section. To accomplish this, a physical model of the fuselage from the

pressurized bulkhead to the aft-most part of the plane must be constructed. This assembly aims to

provide a complete set of results with respect to the transmission of sound and vibration from the

jet turbines to the passenger cabin through the fuselage structure. These results will help to

validate computational models, and these validated models can then be used to guide future

structural improvements. This assembly can then provide a platform on which component

48
modifications can be evaluated as well as other modifications including the implementation of

sound absorptive materials.

The rough design for this assembly was created by Chamberlain and forms a simplified

half scale model of a fuselage tail section (Chamberlain, 2018). Some modifications were made to

the original design primarily to accommodate manufacturing limitations, and details were added

to complete the design for final assembly. The skin was divided into 15 separate panels such that

each could be cut using the available waterjet machine. These divisions were chosen such that

longitudinal seams lined up with stringers, and seams running around the structure lined up with

frames to ensure a strong mechanical connection between each section of skin. To accommodate

the transition from a cylindrical to conical shape at the pressurized bulkhead, a frame was added

to the front side of the bulkhead. This was done to provide a strong mechanical connection

between two independent sets of skin and avoid the complex process of forming the skin over this

transition. This modification carries the added benefit of making the front end of the tail section

modular, and as a result changing the pressurized bulkhead becomes much simpler. This change

is shown in detail in Figure 3.23 through Figure 3.26 which show the final design for the

completed assembly.

49
Figure 3.23: The rectangle highlights the frame added to the front of the bulkhead to
facilitate the transition between cylindrical and conical skin sections.

Figure 3.24: Side view of the completed design of the full tail section.

50
 Figure 3.25: Schematic of the internal structure of the full tail section (units in mm).

Figure 3.26: Completed design of the full tail section sub-assembly.

Techniques were learned during the manufacturing process of sub-assemblies one and

two that will help expedite the construction of the full tail section assembly. On the first version

of the first sub-assembly, the holes for the stringers were not cut on the waterjet. This was done to

51
reduce cost, as there is a significant financial penalty associated with the time spent using the

waterjet machine. This resulted in a significant amount of time drilling these holes into each skin

section by hand, and although care was taken, the result could have been more precise. For the

subsequent assemblies, the stringer holes were cut on the waterjet, but due to a lack of confidence

in the circularity of the frames, the frame holes were omitted until the final assembly of each

structure. This became a significant problem during the construction of the second sub-assembly,

where it was quite difficult to orient the frames in the correct location. If the holes for the frames

were pre-cut in the skin, this procedure would have been easier. It is therefore recommended that

all necessary mounting holes for each section of skin are pre-cut, and while this adds cost with

regards to time spent on the waterjet, the final result will be easier to assemble and more precise.

3.6 Chapter Summary

In this chapter, a detailed description of both sub-assemblies including their design and

construction was presented. The design of the individual components was provided, and where

necessary, the manufacturing methods used to create them was described. The design revisions

implemented on sub-assembly 1 were also presented, and further recommendations for future

manufacturing processes were included. Finally, the designs for the full tail section assembly was

showcased, including some necessary modifications to facilitate more expedient manufacturing.

With an in-depth review of both test articles (sub-assemblies 1 and 2) completed, it is now

suitable to discuss the experimental methodology used to test them in Chapter 4.

52
 Chapter 4

Experimental Methodology

4.1 Overview
The following chapter describes the experimental setup and methodology used to extract

the modal parameters from both sub-assemblies. Additionally, this chapter describes the data

processing steps used to convert the raw data output into a form which could be compared with

the finite element model (FEM). Multiple sources found in the literature were used to guide the

experimental procedure such that it was in accordance with generally accepted best practices. The

work by Avitable was used as a guideline of the general practices to follow to complete a

successful modal test, and was used during the early phases of test planning (Avitable, 2018). It

should be noted that due to differences in test structures and available equipment, no modal

testing methodology will apply universally. During the preliminary testing of both sub-

assemblies, a certain level of trial and error was used to determine the optimal testing parameters

for the experiments presented. The decisions made during this process were guided in part by

random experimentation, but mostly by other sources found in the literature. These references

provide solutions more specific to this testing application and will be cited where appropriate in

the following sections.

4.2 Boundary Conditions

There are two primary methods available when considering the support of a structure

during a modal test. The first option is to test the component in a rigid support structure which is

often designed to replicate the component’s operating condition. This method is also

implemented when considering components as part of large structures which cannot be moved,

and as such must be tested while fixed in place. Replicating the operational condition of a

component is often an arduous task, and for large components such as the fuselage sections tested

53
in this thesis, construction of a rigid support would not be feasible as a result of multiple factors

including cost and available testing space. It is therefore more common to implement a “free-

free” boundary condition for which the primary goal is to decouple the structure from the

surrounding environment. It is not possible to support the structure in a truly free-free condition,

as its mass mandates the use of some form of support structure. To approximate this condition as

closely as possible, soft springs or similar elements are used to form connection between the test

article and an external support structure. In doing so the structure is supported in equilibrium on

spring elements which allow it to oscillate in space along 6 degrees of freedom. These 3

translational and 3 rotational oscillations form the 6 rigid body modes (RBMs) which according

to Avitable should be separated from the low frequency flexural modes by a factor of 10 to

reduce dynamic coupling between the RBMs and the flexural modes (Avitable, 2018). To reduce

the frequency of the RBMs, the softest possible springs or bungees should be chosen to support

the structure. However, a balance must be found between the frequency reduction of the RBMs,

and the structural requirements imposed by the mass of the component to be supported. This latter

criteria is the first priority, especially when testing heavier objects such as the second sub-

assembly, because a failure of the elastic supports could result not only in severe damage to the

structure an experimental equipment, but also pose a safety hazard to those near the structure.

To support the first sub-assembly, two bungee cords were used to connect the fuselage

section to a steel structure resting on the ground. These bungee cords were connected to the sub-

assembly through the skin-frame interface via eye bolts. This location was chosen because it

provided a structurally sound connection point and held the structure in a suitable orientation to

facilitate shaker connection. This location had the added benefit of being modally inactive over

the frequency range of interest, which acted to further reduce the dynamic coupling between the

RBMs and flexural modes. It was found that adjustability to the height of the structure was

required to allow for better alignment with the shaker. To accomplish this, adjustable length ropes

54
were created and used to connect the eyebolts and the bungee cords. This allowed for a fine level

of adjustment to the height of the structure and thus better alignment with the shaker. Figure 4.1

illustrates the boundary condition setup and support structure used during experimental testing of

sub-assembly 1.

BUNGEE CORD

ADJUSTABLE ROPE
STEEL SUPPORT
STRUCTURE

EYE BOLT

Figure 4.1: The boundary condition setup used for sub-assembly 1

55
 The second sub-assembly was approximately 4 times the size and weight of the first and

required stronger bungee cords attached at four locations around the structure. Due to the larger

size of this structure two steel support frames were used to provide a more stable test setup which

to some extent aided with repeatability, but above anything else provided a safe working

environment. Once again adjustable length ropes were used to connect the structure with the

bungee cords, however instead of using eye bolts, carabiners were fastened to existing holes in

both engine support frames. This was done partly for convenience, but also to minimize any

unnecessary modifications to the structure. As a result of the stiffer bungee cords, and their

attachment to modally active sections of the structure, there was a higher risk of the flexural

modes being affected by the boundary condition. These effects will be discussed in more detail in

chapter 5, where future improvements to this experimental setup are also proposed. Figure 4.2

and Figure 4.3 illustrate the support structure and boundary conditions used to test sub-assembly

2.

56
Figure 4.2: The support structure used for sub-assembly 2.

Figure 4.3: Boundary condition setup for sub-assembly 2.

57
4.3 Excitation Methodology
When considering the excitation of a structure during a modal test, there are multiple

methods available to choose from. In some cases, the operating conditions of a structure are used

as a form of natural excitation, which yields some information regarding the dynamic response of

the structure. However, because the force input is not measured, there is no way to ensure

excitation of all resonant frequencies across the bandwidth of interest. Additionally, with no force

input measurement, a calibrated FRF cannot be obtained, and the mode shapes cannot be scaled

(Avitable, 2018). As such, for structures being tested in isolation, excitation methods for which

the input force can be measured are preferred. Of these there are two techniques available, the

modal impact hammer and a modal shaker. There are many benefits and drawbacks to both

methods, but for the case of testing large structures in a stationary lab setting, the shaker table

was chosen. According to (Avitable, 2018), “Shaker testing allows for a much more uniform

excitation of the structure, which is very important when testing larger structures.” The

experimental setup of the modal shaker will be focused on in this section, however for more

information on the experimental setup with regards to an impact hammer, previous work on

experimental modal analysis of individual aircraft components as documented by Chamberlain

should be referenced (Chamberlain, 2018).

The shaker used was a Labworks ET-139 modal shaker. This device includes a through

hole armature as well as a trunnion base, both of which provide adjustability to make alignment

easier during setup. This shaker could provide more than enough excitation force, with a peak

sine force of 75 lbs and a peak random force of 50lbs. A smaller unit would have been equally

suitable, however due to its availability in the lab, this shaker was chosen. The shakers can either

be clamped to a rigid structure or suspended near the test structure on a stand. Both methods are

capable of delivering excitation to the test article, however clamping the shaker down to a large

relatively immobile structure was found to provide a more consistent test setup. An image of the

58
shaker arrangement used during the modal parameter extraction of sub-assembly 2 is shown in

Figure 4.4 below.

Figure 4.4: Shaker setup during sub-assembly 2 modal testing

There is an array of different excitation signals which can be used in conjunction with a

modal shaker to excite the structure. Each have their benefits and limitations and can be divided

into two general categories: deterministic and non-deterministic signals. Deterministic signals

such as sine sweep and sine chirp increment through the frequency range of interest from user

defined lower and upper bounds. They are defined by a mathematical function, and as such will

provide the same excitation throughout the averaging process. As a result, they are ideal for

determining the response linearity of a structure (Avitable, 2018). The main advantage to sine

chirp is that a complete sweep of the frequency range is completed within one sample interval.

This allows for the response to decay, and in conjunction with a pre-trigger delay, the periodicity

requirement of the FFT can be satisfied without the use of windows. For these reasons, a sine

chirp signal was selected to evaluate the linearity of both structures during the linearity validity

checks.

59
 Non-deterministic signals including pure random and burst random do not conform to

any mathematical function. They therefore provide different excitation throughout the averaging

process and will average out noise and slight non-linearities from the results. However, it is

important to determine the overall response linearity using a deterministic signal, before

proceeding with the assumptions made for linear modal analysis. Both structures were found to

behave in a linear manner, however, slight noise and non-linearities were found as seen in Figures

5.2, 5.3 and 5.4 so a burst random signal was chosen for modal parameter extraction. This signal

has the same benefits as sine chirp, in that the signal is contained within one sample interval so in

most cases no windowing is required, which is why it was chosen over other non-deterministic

signals. To generate the excitation signals for each test, an LDS Dactron Photon 2.0 was used in

conjunction with RT Pro software (LDS-Dactron, 2004) (Bruel & Kajer, 2008).

The shaker was powered by a Labworks PA-141 Linear Power Amplifier. This device

could be used as a highly damped voltage source or as a high impedance current source. There are

benefits to both operational modes, but when using waveforms which are meant to decay in the

sample interval, the voltage source mode is preferred. Using this mode provides resistance to the

armature in the shaker after excitation which helps the input signal decay to zero more quickly.

This resistance gets recorded by the impedance head transducer on the structure, so the results are

not contaminated, and the increased decay rate helps to eliminate the need for windows (Peres,

Bono, & Brown, 2010). This meant that any signal distortion that might come as a result of

windowing was not present in the results produced by this experiment.

Another factor to consider when using an impact shaker is the method of coupling it to

the structure. It is of course essential to measure the input force into the structure to collect

enough information to provide calibrated FRF plots. There is however a complication associated

with this measurement, as typical force transducers or impedance heads used in experimental

modal analysis (EMA) only measure force in one axis. It is therefore essential to excite the

60
structure along this axis, while minimizing all other inputs to the structure which would

contaminate the results. This is done by using a connecting rod or wire known as a stinger, which

directs force along one axis, while being flexible in all other directions as to not impose stiffness

effects on the structure.

There are multiple types of stingers which effectively reduce the out of axis forces

applied to the structure. A piano wire stinger accomplishes this most effectively, as the lateral

stiffness of the wire is minimal, however the wire must be pre-loaded with a force greater than the

applied force to the structure to avoid buckling during testing. This adds more complication to the

test setup, and for structures suspended using a free-free boundary condition a piano wire is often

not feasible. The most commonly used stinger is a thin “drill rod” which negates the need for

applied pre-load and is viewed as an acceptable compromise between performance and usability

(Peres, Bono, & Brown, 2010). As such a drill rod stinger was selected for use during the

experiments presented, however the added lateral stiffness has the potential to reduce the quality

of the results. The effect of the stinger’s lateral stiffness on the results is largely dependent on the

stiffness of the structure as well as the frequency range being tested. For relatively stiff structures

being tested at low frequency ranges (<1000 Hz) the lateral stiffness of the stinger is not as

critical, however this needs to be validated on a case by case basis. According to Peres et al. an

effective way to evaluate the stingers lateral effects on the structure is to vary the length of the

stinger by +/- 10% and observe any changes in the measured frequency response functions (Peres,

Bono, & Brown, 2010). This procedure was included as part of the pre-test validity checks and

the results are presented in Chapter 5. Figure 4.5 illustrates a typical drill rod stinger forming a

mechanical connection between a modal shaker and test structure.

61
 Figure 4.5: Typical drill rod stinger setup implemented during preliminary testing of sub-
assembly 1.

4.4 Input and Output Measurements

To collect FRF curves during a modal test, the input force must be measured. Previously

this was accomplished using a force transducer, however this creates an issue regarding the

driving point measurement (Peres, Bono, & Brown, 2010). Significant error in this measurement

is associated with misalignment of the force transducer and accelerometer. An impedance head

integrates these sensors into one unit thus ensuring perfect alignment and has the added benefit of

providing a driving point measurement for each set of measurements performed throughout the

experiment. To collect the force input and driving point measurements, a PCB Piezotronics

288D01 impedance head was used (PCB Piezotronics, 2020). The impedance head was attached

to the structure by first adhering the supplied adhesive mounting base to the structure, then

attaching the sensor via an intermediate stud. For permanent installations it is also possible to

attach the sensor directly to the structure assuming a hole for the stud can be drilled and tapped in

the appropriate location, however for the experiments performed the adhesive base was sufficient.

A schematic of the impedance head and adhesive base is shown in Figure 4.6. For a more detailed

62
drawing as well as product specifications reference the PCB Piezotronics website (PCB

Piezotronics, 2020).

Figure 4.6: PCB Piezotronics impedance head 288D01.

The selection of a suitable excitation location is essential to the success of a modal test.

Factors to consider when selecting a driving point include accessibility, the ability to excite all

modes in the bandwidth of interest, and the ability to excite the entire structure. The first

consideration is trivial, as the point chosen must be accessible for attachment with the shaker

stinger apparatus, which in this case limits the available options to exterior surfaces of both

63
fuselage sub-assemblies. To excite all modes in the bandwidth, first the location of modal nodes

must be considered. As stated in Chapter 2, excitation at the node of a certain mode N will result

in little to no response, greatly reducing or eliminating the presence of a resonant peak in the FRF

data. The most common way to avoid a node is to excite the structure in a modally active region,

however this can create other issues when considering flexible structures. If the shaker is attached

to the structure where the response displacement is high, it will not be able to force the structure

over its entire range of motion. This will cause a drop in the force spectrum at one or more

resonances, which will result in poor frequency response measurements (Avitable, 2018). In this

situation exciting the structure in a stiffer location will provide more even excitation across the

frequency range and can also assist in exciting the entire structure more evenly. To find a suitable

driving point location, multiple tests were performed on sub-assembly 1. Initially the structure

was excited through DOF 20 of the aft bulkhead, however this location was not stiff enough to

excite the surrounding skin structure. Multiple locations on the skin were also evaluated, however

many of the same problems were encountered. Ultimately, it was determined that exciting the

structure where the bulkhead connected to the skin as shown in Figure 4.7 provided the best

results. The results from this test were considered when selecting the driving point for sub-

assembly 2, which again was located near the connection between the skin and FESF panel as

shown in Figure 4.8.

Accelerometers are most often used to measure the dynamic response of a structure

during EMA. There is one major factor which much be considered when selecting and using

accelerometers, and that is the effect of the accelerometer’s mass on the dynamic response of the

system. There are a couple of ways to overcome this issue the first and simplest of which is to

select lightweight accelerometers. The accelerometers chosen were PCB Piezotronics 352C22,

which had a mass of just 0.5g (PCB Piezotronics, 2020). According to Avitable however, “while

the accelerometer weight may be very small in relation to the wight of the structure, the weight

64
must be considered relative to the effective weight of the structure at the attachment point.”

(Avitable, 2018). To evaluate the effects of the accelerometer mass on the structure, dummy

masses were machined out of aluminum to simulate the weight of the accelerometer on the

structure in locations where accelerometers were not present. The structure was tested with and

without the dummy masses in place, and it was found that the mass of the accelerometers had a

significant effect on the alignment of resonant frequencies as shown in Figures 5.11 and 5.12. To

successfully correlate the FEM results with the experimental results, the accelerometer mass must

be included in the computational model, which is accomplished by adding a lumped mass to the

nodal location of the accelerometer. This simple modification is made complicated when

considering a roving response measurement, during which the accelerometers (and their

associated mass) move around the structure, resulting in a non-constant mass distribution. The

simplest way to mitigate this is to add dummy masses to the DOFs not being measured, such that

the distribution of mass is kept constant regardless of accelerometer placement. As such, dummy

masses were implemented during testing of both sub-assemblies to eliminate the roving mass

effect, which improved the alignment of the resonant peaks due to the global parameters of the

system remaining constant.

Another critical consideration during the setup phase is the selection of response

measurement locations. The objective of this placement is to provide both complete visual and

mathematical representations of each mode shape in the bandwidth of interest. It is also important

to consider which DOFs will be required to form a set of mathematically correlated mode pairs

between the experimental results and computational model. The expected mode shapes of the

structure, and their complexity will for the most part define the number of degrees of freedom

required. The computational models developed by (Lam & Mechefske, 2020). provided an

indication of the mode shapes for both sub-assemblies, which in turn guided the placement of

measurement locations for preliminary tests. The placement of DOFs for the bulkhead of sub-

65
assembly 1 is shown in Figure 4.7 below. Points 1 through 55 were used during preliminary

testing, at which point an AutoMAC (as described in Chapter 2) was used to determine if the

measurement was sufficient. It was found that more points were required and as such points 129

through 146 were added and the measurement process was repeated. The AutoMAC matrices

produced by both tests are presented in Chapter 5 to illustrate the improvement made.

Figure 4.7: Sub-assembly 1 bulkhead DOF placement.

The response measurement locations for sub-assembly 2 were selected using the same

method used for sub-assembly 1. The FEM produced by (Lam & Mechefske, 2020) was used to

guide the placement for preliminary tests, then an AutoMAC was used to provide an indication of

the confidence with which each mode shape could be mathematically differentiated. When

evaluating the AutoMAC created using the preliminary data, it was found that the initial

placement for both the FESF and RESF of the second sub-assembly as shown in Figure 4.8 and

Figure 4.9 was sufficient.

66
Figure 4.8: DOF placement for the FESF of sub-assembly 2.

Figure 4.9: DOF placement for the RESF of sub-assembly 2.

67
4.5 Data Acquisition
The data from the validity checks and full modal experiments was collected using a

National Instruments NI PXI-1031 (National Intruments, 2019). This system can collect data

from 16 channels, however for the experiments performed, only 10 were used. Two channels

were used to collect force and acceleration data from the impedance head, while eight were used

to collect response data from the roving accelerometers. Using this number of accelerometers

provided a compromise between expedient testing, and the ability to monitor individual

accelerometer measurements during the averaging process. Using more accelerometers would

have reduced testing time but would have also increased the probability of sensor error going

unnoticed during testing.

For the first sub-assembly, a sampling rate of 1000 Hz was used to collect data over a 2.5

second sample interval. It was found that 2.5s provided enough time for the structure to be

excited fully by the shaker, then have its response decay back to zero in accordance with the

periodicity requirement associated with FFTs. This yielded a frequency resolution of 0.4 Hz for

the frequency domain plots, and a maximum bandwidth of 500 Hz. This higher bandwidth was

chosen as it was not initially clear what frequency range would be required to confirm correlation

between the experimental results and computational model. Additionally, there were not any

limiting factors regarding the data storage capabilities, so longer sample intervals could be used

to increase the frequency resolution of frequency domain plots. The second sub-assembly used

the same sampling rate, but the sample interval was increased from 2.5s to 5s. This was done to

give the structure more time to reach steady response and decay, which was required due to its

increased mass and response levels at certain DOFs. It also had the added benefit of increasing

the frequency resolution to 0.2 Hz for frequency domain plots. Fifteen averages were taken

during modal parameter extraction of both sub-assemblies, which increased the statistical

68
reliability of the data, and helped to reduce the effects of noise and slight non-linearities due to

the burst random excitation signal (Agilent Technologies, 1997).

Another important factor to consider when configuring the sampling parameters is the

pre-trigger delay. This parameter allows for data before the start of each excitation to be stored,

thus capturing a near zero response of the structure as well as the complete excitation which

otherwise might be missed. Any excitation information that is not collected would likely result in

distortion of the FRF data (Avitable, 2018). A pre-trigger delay is especially important when

performing impact measurements, and shaker measurements using non-deterministic signals, as it

allows for the periodicity requirement to be met for these instances of random input without the

use of windows. For the experiments presented, the sample was triggered by the force transducer

in the impedance head, and 200 pre-trigger samples were collected.

4.6 Signal Processing

The raw analog input and output signals must be processed into useful FRFs which can

later be used for modal parameter extraction. This process begins with the application of

antialiasing filters, which in this case are low pass filters to prevent high frequency content from

contaminating the results in the bandwidth of interest. For most data acquisition systems, the filter

parameters are set based on the sampling frequency chosen. A sampling frequency of 1000 Hz

was used during testing resulting in a frequency bandwidth of 500 Hz in accordance with the

Nyquist requirement. The input and output signals are then digitized using a built-in analog to

digital converter at which point windows could be applied if necessary. For the experiments

presented, no windows were required. The digitized input and output data could then be used to

calculate the linear input and output frequency spectrums using an FFT. The linear spectra were

then converted to the power spectra, at which point the coherence and FRF functions could be

computed. The conversions from linear to power spectra are defined below.

69
 𝑋(𝑓): 𝐿𝑖𝑛𝑒𝑎𝑟 𝐼𝑛𝑝𝑢𝑡 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

𝑌(𝑓): 𝐿𝑖𝑛𝑒𝑎𝑟 𝑂𝑢𝑡𝑝𝑢𝑡 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

𝐺𝑥𝑥 = (𝑋)(𝑋 ∗ ): 𝐼𝑛𝑝𝑢𝑡 𝐴𝑢𝑡𝑜 𝑃𝑜𝑤𝑒𝑟 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

𝐺𝑦𝑦 = (𝑌)(𝑌 ∗ ): 𝑂𝑢𝑡𝑝𝑢𝑡 𝐴𝑢𝑡𝑜 𝑃𝑜𝑤𝑒𝑟 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

𝐺𝑥𝑦 = (𝑋)(𝑌 ∗ ): 𝐶𝑟𝑜𝑠𝑠𝑝𝑜𝑤𝑒𝑟 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

𝐺𝑦𝑥 = (𝑌)(𝑋 ∗ ): 𝐶𝑟𝑜𝑠𝑠𝑝𝑜𝑤𝑒𝑟 𝑆𝑝𝑒𝑐𝑡𝑟𝑢𝑚

Finally, the selection of an appropriate FRF estimator is required to produce accurate

results. The two primary methods most commonly available in industry are H1 and H2 shown in

equations 4.1 and 4.2. The major factor to consider when making the selection is whether there is

more likely to be noise generated in the input or output signal. According to Ewins, noise near

resonance is likely to influence the force signal, while noise at anti-resonance will affect the

response signal(s). H1 will likely be less accurate with noise at the input, while H2 will be more

liable to errors with noise at the output (Ewins, 2000).

𝐺𝑦𝑥
𝐻1 = (4.1)
𝐺𝑥𝑥

𝐺𝑦𝑦
𝐻2 = (4.2)
𝐺𝑥𝑦

More noise was observed near anti-resonances in the FRFs output from the data

acquisition system resulting in the H1 estimator being chosen. This decision is further supported

by previous work done by Chamberlain et al. and in the referenced text (Ewins, 2000)

(Chamberlain, 2018). It should be noted that steps were taken to minimize external vibrations

which had the potential to introduce input noise. This was primarily done through the isolation of

the vacuum pump used to cool the shaker which was accomplished by moving it from the table

on which the shaker was mounted to a thick rubber pad on the floor of the lab.

According to Avitable however, the H1 estimator underpredicts the amplitude of the FRF

while the H2 estimator overpredicts the FRF amplitude (Avitable, 2018). It is therefore necessary
70
to quantify the level of discrepancy between the two estimators which can be accomplished using

the coherence function shown in equation 4.3.

𝐺𝑦𝑥 𝐺𝑥𝑦 𝐻1
𝛾2 = = (4.3)
𝐺𝑥𝑥 𝐺𝑦𝑦 𝐻2

The coherence functions are also monitored to ensure the repeatability of the FRF data

being collected. A coherence measurement of one would indicate no discrepancy between the two

estimators, and serves as a general indicator of high measurement quality. Coherence values will

in theory remain between 0 and 1, and for acceptable measurement should remain between 0.9

and 1 over the frequency range. One exception to this requirement is near anti-resonances, where

the output response is very low. At these frequencies the signal to noise ratio is also very low, and

it is expected for the coherence to drop (Avitable, 2018). An example of a typical FRF

measurement and coherence function is shown in Figure 4.10.

Figure 4.10: Typical FRF and coherence measurement. Notice the dips in coherence in
relation to the anti-resonances of the FRF (Avitable, 2018).
71
4.7 Data Analysis
ME’Scope VES 4000 software was used to extract the modal parameters from the raw

data collected during testing. This software provides a curve fitting procedure which applies

mathematical expressions to the experimental data, such that natural frequencies, damping values

and residues can be extracted. The first step of curve fitting is to identify the position of the

resonant peaks in the frequency range of interest. There are two options available to accomplish

this: the complex mode indicator function (CMIF) or the multivariate mode indicator function

(MMIF). Of these the CMIF proved to be most effective at identifying resonance peaks in the

frequency range. The mode indicator function serves two purposes, to provide a single curve for

counting the number of resonance peaks, and to limit the frequency and damping curve fitting

methods to the data surrounding each resonant peak (Vibrant Technology, 2015).

The next step is to curve fit the data which in this case was done using a global

polynomial method. According to Vibrant Technologies, the global polynomial method may be

used when the resonant peaks from each DOF show good alignment with respect to frequency,

which was found to be the case for the data extracted from both sub-assemblies. This method

provides one frequency and one damping value for each resonance in the frequency band which

act as global properties for the system (Vibrant Technology, 2015). One factor that allowed this

method to be implemented was the use of dummy masses during testing. This ensured that the

physical properties of the system remained constant regardless of accelerometer placement, which

resulted in closer peak alignment between FRFs. The results from the dummy mass experiment

which illustrate their effect on the results are presented in Chapter 5.

The final step is the determination of the residues of each mode in the frequency range.

As with curve fitting there are two methods to choose from. According to the Vibrant

documentation, the “peak” method is most appropriate for FRF data in which the peaks do not

align closely. However, for FRFs which contain closely coupled modes, the peak method is not

72
suitable, and a MDOF method which can simultaneously estimate the modal parameters of

closely coupled modes should be used. The polynomial method uses all of the real and imaginary

data to estimate the numerator polynomial of each FRF using the least squared error curve fitting

process (Vibrant Technology, 2015). The denominator is estimated using the frequency and

damping terms from each mode found during curve fitting. After the residues are calculated, a

shape table containing the frequency and damping terms as global properties, and the magnitude

and phase information for each DOF and every mode shape is created. The frequency values are

used during the first stage of correlation with the computational model, and the magnitude and

phase information is used when correlating mode shapes using the pseudo orthogonality check.

4.8 Chapter Summary

A review of the experimental setup including the boundary conditions, excitation

methodology and data acquisition parameters was presented in this chapter. Additionally, the

techniques used to extract the natural frequencies and mode shapes from the raw data was

presented. The sensors used to collect data, and their placement on the structure was also

described in detail, in conjunction with the AutoMAC technique used to validate their position

during testing. Testing considerations including the driving point location which are specific to

the experiments presented were also described for the purposes of progressing this project onto

larger structures. In each instance, the choices made were supported by previous work found in

the literature, or where applicable, by the technical support resources produced in industry. This

chapter aims to facilitate a better understanding of the experimental setup and procedure, as well

as the data acquisition and analysis methods. It is now suitable to review the results produced by

these experiments and their correlation with the computational model in Chapter 5.

73
 Chapter 5

Results and Correlation

5.1 Overview
The following chapter presents the experimental results collected, and their correlation

with computational models produced by (Lam & Mechefske, 2020). The validity checks used to

confirm the quality of the experimental results are described, and the results from these checks

are discussed. These checks include linearity, repeatability and driving point measurements as per

ISO guidelines, as well as additional measurements inspired by the literature. These

measurements include a mass effects study and a test to evaluate the effects of stinger length on

excitation quality. The results from these validity checks show that the assumptions made

regarding linear modal analysis are upheld, however minor issues in the results from these checks

were found and are discussed.

The final experimental results including the natural frequencies and mode shapes of both

sub-assemblies are presented. Additionally, the techniques used to evaluate the correlation

between the experimental and computational results are discussed, and the results of the

correlation studies are presented. In general, a strong level of correlation was found between the

FEM and experimental natural frequencies and mode shapes for both sub-assemblies. Finally,

recommendations are made for both future experimental testing improvements, as well as further

model enhancements to improve correlation further.

5.2 Validity Checks

5.2.1 Linearity Check

One of the primary assumptions associated with the experimental modal analysis

techniques implemented, is that the structure behaves in a linear manner. According to Ewins

“Most of the theory upon which modal testing and FRF measurement is founded, relies heavily
74
on the assumption that the test structure's behavior is linear.” (Ewins, 2000). In other words, the

response of the structure is linearly related to the force input, and as a result the FRF

measurements will remain constant regardless of input amplitude. Additionally, computational

modelling performed for this project assumed the structure was linear, further confirming the

need for this check. The linearity check evaluates this behavior by repeating a measurement at

certain input and output measurement locations at varying amplitude and evaluating the resultant

FRF measurements. Non-linear behavior would cause discrepancies in the frequency response

plots especially near resonance, which would be exhibited by a change in amplitude of a

resonance. A lack of any significant discrepancy in the FRFs produced from this test indicates

that the structure is behaving linearly within the range of excitation used.

The procedure used to evaluate response linearity was identical for both sub-assemblies.

As stated in Chapter 4, a deterministic signal should be used to evaluate linearity, and as such a

sine chirp signal was selected. The literature states that one input-response pair should be used to

evaluate linearity, but because eight accelerometers were already connected, eight points were

tested on both structures. This was especially important for the second sub-assembly which could

potentially exhibit linear behavior in the FESF but non-linear behavior in the RESF due to the

amount of mechanical connections between excitation and response measurement locations. It is

also recommended that the excitation force be varied by a factor of 10 where possible, however

there were limitations in the experimental setup which prevented this from being possible

(International Organization for Standardization, 1994). Particularly with the second sub-

assembly, there was a certain threshold excitation below which the entire structure was not

properly excited. This is not surprising due to the structure’s size, however ten times this

threshold excitation would risk damage to the stinger, and potentially the impedance head. For

this reason, the excitation signal was only varied by a factor of approximately 6 for the linearity

check of the second sub-assembly.

75
 For the first sub-assembly, four accelerometers were placed on the center bulkhead at

points 2, 12, 22 and 42; the schematic for which can be found in Figure 4.7 and four were

distributed around the skin at points 72, 82, 92 and 112. The structure was excited through the

driving point used for the rest of the experimental testing, and all other factors including the

boundary conditions also remained constant. The excitation level was varied from 2N to 7 N then

finally to 15 N peak input. The resultant FRFs produced by the linearity check of sub-assembly

one are shown in Figures 5.1 and 5.2. The following figures show the results for only one DOF

per graph for ease of viewing, but it should be noted that all DOFs from sub-assembly 1 showed

similar results.

Figure 5.1: Linearity check from DOF 2 on the bulkhead of sub-assembly 1. (Orange 2N,
Blue 7N and Green 15N peak input)

76
 Figure 5.2: Linearity check from DOF 2 on the bulkhead of sub-assembly 1. (Orange 2N,
Blue 7N and Green 15N peak input)

As exhibited by Figures 5.1 and 5.2, the variance of input excitation force from 2 to 15 N

resulted in minimal discrepancies with regards to the resonance peaks. At higher excitation levels,

more noise was seen near minima and anti-resonances, which indicates the presence of slight

non-linearities. However, due to the consistency of the resonant peak amplitudes, and the

elimination of this noise through the implementation of a burst random excitation signal, these

noisy regions are not of concern. It can therefore be concluded that sub-assembly 1 shows linear

dynamic response, and the assumptions made regarding linear modal analysis are upheld.

To evaluate the linearity of the second sub-assembly, four accelerometers were placed on

the FESF at points 4, 40, 44, and 128, and on the RESF at points 64, 84, 110, and 144 the

schematics for which can be found in Figures 4.8 and 4.9 respectively. Once again 3 excitation

77
levels were used: 3N, 10 N and 20 N peak input. A sample of the resultant FRFs are shown in

Figures 5.3 and 5.4, and once again only 1 DOF from both the FESF and RESF are showcased for

ease of viewing.

Figure 5.3: Linearity check from DOF 44 on the FESF of sub-assembly 2. (Orange 3N, Blue
10N and Green 20N Peak Input)

78
 Figure 5.4: Linearity check from DOF 144 on the RESF of sub-assembly 2. (Orange 3N,
Blue 10N and Green 20N peak input)

Once again minimal discrepancies are found between the various tests indicating that the

structure is behaving in a mostly linear nature. It was found that at higher excitation levels the

plots became slightly noisier, as shown by the green and blue traces, however throughout modal

parameter extraction, this noise was averaged out using a non-deterministic signal. Once again,

the peaks show strong alignment both in frequency and amplitude, and as such, it can be

concluded that sub-assembly 2 behaves linearly.

5.2.2 Repeatability Check

The repeatability check is used to evaluate the consistency of the experimental setup and

procedure, and the reliability of the resultant data. As per ISO guidelines, at least one

measurement should be repeated before and after modal parameter extraction (International

79
Organization for Standardization, 1994). Multiple points on both sub-assemblies were tested

multiple times throughout the testing procedure. In addition to performing the repeatability

checks in accordance with the ISO recommendations, measurements were taken each time the

sub-assembly was suspended from the support structure. This was done to ensure that any minor

change in the supporting boundary conditions did not have a significant effect on the dynamic

response of the sub-assembly. For most repeatability checks performed, the data was found to be

consistent between trials, and the experimental setup was deemed acceptable. There were some

rare instances where discrepancies were noted, however in each case this was found to be a result

of setup errors including stinger misalignment or sensor connectivity issues. Once corrections

were made, the repeatability check was performed again, and the expected results were found.

The following Figures 5.5 and 5.6 show typical results found during the repeatability checks of

sub-assembly 1 and 2 respectively.

Figure 5.5: Repeatability check of DOF 141 on the bulkhead of sub-assembly 1.

80
 Figure 5.6: Repeatability check of DOF 62 on the RESF of sub-assembly 2.

5.2.3 Driving Point Measurement

The driving point measurement is performed by collecting input and output response

from the same point on the structure. As described in Chapter 2, the expected result is to see no

phase changes in the measured FRF, which can be visualized using two different presentations of

the FRF data. Phase changes are associated both with minima in between resonance peaks and

with a change from positive to negative (or vice versa) in the imaginary FRF content. When

looking at an FRF magnitude plot, it is therefore expected that every resonance peak is followed

by an anti-resonance. When looking only at the imaginary content, all resonant peaks should form

on one side of the frequency axis. Any deviation from these expected results would indicate an

issue with the experimental setup, most likely with the excitation apparatus or transducers. This is

most often the case for impact measurements, where the excitation location is slightly variable,

81
and thus not perfectly aligned with the measurement location. Using a shaker and impedance

head greatly reduces the probability of discrepancies in the driving point measurement, as near

perfect alignment is far easier to achieve. The driving point measurements for both sub-

assemblies are presented in Figures 5.7 through 5.10 below. During modal parameter extraction

these measurements were checked on multiple occasions, and the expected patterns regarding the

anti-resonances in the magnitude plot, and the unidirectional nature of the data in the imaginary

domain were found.

Figure 5.7: Driving point measurement for sub-assembly 1 (Magnitude)

82
Figure 5.8: Driving point measurement for sub-assembly 1 (Imaginary)

Figure 5.9: Driving point measurement for sub-assembly 2 (Magnitude)

83
 Figure 5.10: Driving point measurement for sub-assembly 2 (Imaginary)

5.2.4 Mass Effects Study

To evaluate the effects of the mass imposed by the accelerometers on the dynamic

response of each sub-assembly, a mass effects study was performed. The accelerometer mass,

while small, can have a significant effect on the natural frequencies of the system. This is

especially true when considering the case of eight accelerometers closely grouped during some

phases of modal parameter extraction. To obtain a better understanding of the influence of

transducer mass on the response of both structures, sub-assembly one was tested under two

conditions. First only using accelerometers roving around each DOF, then with the dummy

masses in place at each DOF without an accelerometer present. Without dummy masses, the mass

distribution of the structure changes depending on the accelerometer placement during any given

measurement. This has the potential to effect certain natural frequencies, depending on the mode

shapes of the structure, which can lead to misalignment of peaks in the FRF curve. This

characteristic was noticed particularly at the resonance at 160 Hz where a frequency variance of 3

84
Hz was found. Figure 5.11 illustrates the FRF curves from the bulkhead of sub-assembly one

overlayed to highlight this misalignment.

Figure 5.11: Overlayed FRF curves for bulkhead of sub-assembly 1 (No Dummy Masses)

Dummy masses were then added to the structure and the data was collected again. This

time, a noticeable improvement with regards to the alignment of FRF peaks was observed and is

shown in Figure 5.12. The most severe misalignment was again seen on the peak near 160 Hz;

however, it was reduced from 3 Hz to 1.4 Hz. The other resonances in the frequency range also

showed closer alignment, further emphasizing the benefits of using dummy masses in subsequent

experiments. It should also be noted that each natural frequency was reduced by approximately 1

Hz, which was not a surprising result given the additional system mass. This added mass was

accounted for in the FEM, and as such was not a concern during correlation between

experimental and computational results. If a more precise modal data is required from structures

with closely grouped DOF locations, then non-contact measurement devices such as laser

vibrometers should be considered.

85
Figure 5.12: Overlayed FRF curves for bulkhead of sub-assembly 1 (With Dummy Masses)

5.2.5 Response Measurement Placement Validation

The preliminary DOF placement for both structures was guided by the predicted mode

shapes determined using the computational model. There is a certain level of trial and error

associated with this process, so a method of quantifying the quality of results produced by a

certain DOF layout was required. The MAC provides an indication of eigenvector orthogonality,

and therefore reveals the confidence with which mode shapes can be differentiated. The POC is

often regarded as the most rigorous correlation method, however an AutoMAC computation

function was integrated into ME’Scope. This allowed for DOF layouts to be evaluated more

quickly, with less data conversions and transfers required, so the AutoMAC was selected at this

stage. An auto orthogonality matrix was produced in conjunction with the POC used to evaluate

mode shape correlation later in the experiment, which acted as further validation for the final

DOF placement. Figures 5.13 and 5.14 below show the AutoMAC results for sub-assembly one

before and after points 129-146 were added to the aft bulkhead of sub-assembly 1. Figure 5.15

shows the AutoMAC results for sub-assembly 2. Additional experiments were performed with

86
additional DOF locations on both the FESF and RESF, however negligible improvements were

observed, so the initial placement of DOF locations was used for subsequent experiments. The

higher diagonal terms found in the first four modes of the FESF and the first three modes of the

RESF will be explained when discussing mode shape correlation in section 5.7.

Figure 5.13: Aft Bulkhead AutoMAC before (Left) and after (Right) points were added.

Figure 5.14: Sub-assembly 1 AutoMac for all modes after points were added.

87
 Figure 5.15: Sub-assembly 2 AutoMAC for the FESF (Left) and RESF (Right)

5.2.6 Stinger Length Test

The final check performed prior to modal parameter extraction evaluated the effects of

stinger length on the experimental results. There are two primary concerns associated with stinger

length. If the stinger is to short, lateral stiffness could be imposed on the structure, thus

contaminating results, and a stinger which is too long could introduce additional peaks into the

FRF curves caused by stinger resonance (Peres, Bono, & Avitable, 2012). Once again, sub-

assembly one was used in conjunction with 8 accelerometers distributed around the structure. As

with previous validity checks, four accelerometers were placed on the center bulkhead, and four

were placed on the skin. Three measurements were performed using a similar procedure to that

used for the repeatability check, except between trials, the stinger length was varied. Three

different stinger lengths were used ranging from 10.5cm to 15 cm and finally 25 cm.

Unfortunately, due to geometric limitations with the skin overhanging from both sub-assemblies,

it was not possible to test shorter stingers. The results from this test are presented in Figure 5.16

which shows an overlaid plot of the three FRFs from DOF 42 of the first sub-assembly. Over the

200 Hz frequency range, negligible discrepancies are found both at resonance and lower

amplitudes. The longest stinger did produce slightly noisier results, so the use of shorter stingers

88
over this frequency range is recommended for future testing. It can be concluded however, that

for the experiments presented, the length of the stinger is shown to have little effect on the natural

frequencies or mode shapes. Moving forward a stinger length of approximately 11cm was used

during the modal parameter extraction of both sub-assemblies.

Figure 5.16: Stinger length comparison for DOF 42 of sub-assembly 1. (Orange: 10.5 cm,
Green: 15 cm, Blue: 25 cm)

5.3 Correlation Overview

As a result of the numerous validity checks implemented prior to testing, there is now an

acceptable level of confidence in the experimental results collected. The experimental data

collected included the natural frequencies and mode shapes, which were then compared with the

results produced computationally. Where applicable, recommendations were made regarding

future modelling improvements, and in the case of the first sub-assembly, a significant

improvement in natural frequency correlation was produced. The methods used to compare mode

shapes including both visual and mathematical techniques are also discussed. In some instances,

the results indicated that more DOFs were required, which was found to be the case for the

bulkhead of sub-assembly one. Once the necessary improvements to both the experimental setup
89
and computational model were made, a strong level of correlation was found between the sets of

data.

5.4 Sub-assembly One Natural Frequency Correlation

Fourteen mode shapes were predicted computationally, and thirteen modes were found

experimentally in the 200 Hz bandwidth and visually correlated. The natural frequencies between

the two sets were compared as shown in Table 5.1. A skin mode and bulkhead mode (highlighted

in yellow) were too closely coupled to separate. This meant that correlation with the

computational model for these modes was not possible. Highlighted in green are the mode shapes

found in the skin sections of sub-assembly one, while the rows highlighted in blue are the

bulkhead modes. When evaluating the percentage difference between the preliminary model and

experimental natural frequencies, a significant discrepancy between the natural frequencies

related to the skin modes of the structure was observed. It was theorized that by bending each

skin section around the center frame, a pre-stress was imposed effectively stiffening the skin. As

a result, the initial computational model underpredicts the skin mode natural frequencies by 10-14

%. A significant amount of work was performed by (Lam & Mechefske, 2020) to evaluate the

most accurate way to stiffen the skin in the computational model. Multiple updating techniques

were evaluated to find a technique that added stiffness to the skin without contaminating the

bulkhead results or mode shape data. The most successful methodology involved increasing the

modulus of elasticity of the stringers by 100 %. This technique along with many other trialed are

described in detail by (Lam & Mechefske, 2020). The final model was able to predict natural

frequencies within 6 % and it can therefore be concluded that the updated model of sub-assembly

one is successfully predicting natural frequency.

90
 Table 5.1: Comparison of sub-assembly 1 natural frequencies (Skin modes in orange,
bulkhead modes in blue)

Revised
Experimental Preliminary Preliminary Model Final Model
Mode Description CMP Model
(Hz) Model (Hz) Discrepancy (%) Discrepancy (%)
(Hz)
Oval Skin 2 1 31.3 27.2 -13.22 30.84 -3.2
Bulkhead Drum 2 41.1 43.6 6.00 43.55 5.9
Triangle Skin 3 69.7 61.4 -11.89 70.05 0.5
Triangle Skin 2 4 71.6 62.4 -12.90 71.35 -0.4
Bulkhead Halves 5 83.3 86.5 3.83 86.59 3.9
Bulkhead Halves 2 6 86.2 87.8 1.84 87.82 1.9
Square Skin 1 7 105 93.0 -11.39 106.32 1.3
Five Point Skin 138 140.35 1.7
Bulkhead Quarters 138 140.00 1.5
Bulkhead Quarters
2
8 142 142.1 0.07 142.24 0.2
Bulkhead Double
Drum
9 155 162.6 4.87 162.67 4.9
6 Point Star Skin 10 170 152.9 -10.07 172.78 1.6
6 Point Star Skin 2 11 172 153. -10.95 174.07 1.2

5.5 Sub-assembly One Mode Shape Correlation

Of the fourteen predicted mode shapes, thirteen were found experimentally and visually

correlated, and eleven were mathematically correlated. It was found experimentally that a skin

mode and bulkhead mode both resonated at 138 Hz. The computational model predicted a slight

separation in natural frequency between these mode shapes and as a result, these two mode

shapes could not be mathematically related. Additionally, experimental differentiation of one

mode was not possible due to the closely coupled nature of axisymmetric modes. As stated by

Ewins, axisymmetric structures can often undergo dynamic behavior resulting in repeated modes

as illustrated in Figure 5.17. For perfectly symmetric structures at a given natural frequency, one

mode will be found as a combination of all possible resonant shapes (Ewins, 2000). In reality, the

first sub-assembly is only nearly axisymmetric, resulting in multiple closely coupled modes. In

the case of the “square skin” modes, the two mode shapes predicted by the model were too

closely coupled to separate in the experimental data. As a result, the second “square skin” mode

shape was also not correlated with the model. The nearly axisymmetric property of sub-assembly

91
1 in addition to the closely coupled bulkhead and skin mode discussed earlier, meant that only 11

of 14 predicted modes could be correlated using the following techniques.

Figure 5.17: Repeated modes of axisymmetric structures (Ewins, 2000).

The first stage of mode shape correlation involves performing a qualitative evaluation of the

shapes found experimentally, with those predicted computationally. To produce a graphical

representation of each experimental mode shape, ME’Scope uses the amplitude and phase data

extracted experimentally from each DOF to generate animations of the structure at resonant

frequencies. These were then compared with the graphical mode shape output from the FEM, at

which point a list of correlated mode pairs was generated. Thirteen of the fourteen predicted

modes were visually correlated, and an example of this visual correlation is shown in Figures

5.18 and 5.19. The overlapping experimental modes at 138 Hz and their computational

counterparts are shown in Figures 5.20 and 5.21 to illustrate the issue which presented itself while

trying to mathematically correlate modes.

92
Figure 5.18: “6 Point Star Skin” FEM mode shape (CMP 10 at 172.78 Hz) (Lam &
Mechefske, 2020)

93
Figure 5.19: “6 Point Star Skin” Experimental mode shape (CMP 10 at 170 Hz) (Lam &
Mechefske, 2020)

94
 Figure 5.20: Experimental mode eight at 138 Hz (highlighted in yellow in Table 5.1)

Figure 5.21: Computational Modes eight and 9 (highlighted in yellow in Table 5.1) (Lam &
Mechefske, 2020)

95
 Once the mode shapes have been visually correlated between the two sets of results, the

quality of correlation must be quantified. The POC was used to compute the level of correlation,

with good correlation being indicated by diagonal terms above 0.8 and off diagonal terms below

0.2. The results from the POC implemented on the results from sub-assembly one are shown in

Figure 5.22. It is found that all diagonal terms in the POC matrix are above 0.94 and all off

diagonal terms are below 0.2 with the exception of the term from the comparison of experimental

mode 11 with computational mode 10 which had a value of 0.21. It can be concluded that for the

eleven modes presnted below, there is a strong level of correlation. This confirms both that the

correct natural frequencies are being compared, and that the computational model is correctly

predicting the overall dynamic behaviour of sub-assembly one.

Figure 5.22: POC computation for 11 correlated modes of sub-assembly one.

96
5.6 Sub-Assembly Two Natural Frequency Correlation
Due to the modal density of the second sub-assembly in the 200 Hz bandwidth, the

results from both engine support frames were separated prior to analysis and correlation. The

procedure to evaluate the correlation between experimental and computational results followed

the same steps as used for the first sub-assembly where mode shapes were visually related, their

respective natural frequencies compared, at which point the POC was used to determine the

quality of mode shape correlation for the mode shapes compared. The comparison of natural

frequencies for sub-assembly two is presented in Table 5.2.

Table 5.2: Comparison of sub-assembly 2 natural frequencies

CMP FESF Experimental Results (Hz) Computational Results (Hz) Percent Difference (%)
1 21 15.17 -27.74
2 33.8 20.78 -38.52
3 41.6 30.24 -27.32
4 46.8 43.35 -7.38
5 66.5 60.83 -8.52
6 79.1 76.99 -2.67
7 96.5 94.27 -2.29
8 115 105.82 -7.98
9 127 124.22 -2.19
10 137 133.31 -2.69
11 151 138.29 -8.42
12 158 149.01 -5.69
13 163 151.89 -6.81
14 170 168.90 -0.65
15 188 175.67 -6.56

CMP RESF Experimental Results (Hz) Computational Results (Hz) Percent Difference (%)
1 25.5 17.66 -30.75
2 32.5 20.61 -36.58
3 41.6 36.37 -12.57
4 75.8 66.43 -12.36
5 89 85.05 -4.44
6 136 119.71 -11.98
7 153 141.87 -7.28
8 157 147.11 -6.30
9 177 159.02 -10.16
10 197 176.13 -10.59

97
 From the results in Table 5.2, there is clearly an issue with the experimental frequencies

of CMPs 1 - 3 of the FESF and the first two CMPs of the RESF. When evaluating the natural

frequency comparison of the first sub-assembly, there was a relatively consistent magnitude of

error between computational and experimental results, however this was not the case for the

second sub-assembly. There is an acceptable level of error for most CMPs, but the first three

natural frequencies of the FESF in addition to the first two CMPs of the RESF show a far greater

discrepancy. It is theorized that the boundary altered the natural frequencies of the first three

modes. This is based on previous research where boundary conditions were shown to have a

significant effect on natural frequencies (Chamberlain, 2018). To best approximate the free-free

boundary condition, soft elastic members are recommended to support the test article, however to

support the weight of the structure, relatively stiff bungees were required. The second sub-

assembly was supported through the pylons of both engine support frames which only show

modal activity for the first three modes of both the RESF and the FESF. The experimental natural

frequencies for these modes indicate that the bungees are imposing a stiffening effect on the

structure. It is therefore recommended that the support locations be moved from the engine

pylons to the sections of frame surrounding both engine support frames, which are modally

inactive over the bandwidth of interest. Testing the sub-assembly using this configuration will

confirm or disprove this theory with the latter scenario indicating an issue with the computational

model.

It should also be noted that the computational model underpredicts every natural

frequency in the bandwidth. This was not the expected result, as computational models generally

overpredict stiffness and as a result, natural frequency. This discrepancy is most likely due to the

outward tension imposed on both engine support frames by stress in the sections of skin. This

theory is further corroborated by the larger discrepancies found in the RESF natural frequencies.

The RESF has a smaller radius than the FESF, and thus the skin section must be bent further to

98
conform to its shape. This results in more pre-stress in the skin, and therefore more outward

tension on the RESF. Recommendations to improve this involve either pre-forming the skin to the

required radii, or using aluminum of lower young’s modulus for the skin sections.

5.7 Sub-assembly Two Mode Shape Correlation

The same procedures used to correlate the mode shapes found with sub-assembly one,

were implemented when comparing mode shapes found for sub-assembly two. That is, a

graphical representation of each mode shape was generated using Me’Scope, and these were then

visually matched to those generated computationally. The model predicted 25 mode shapes over

the 200 Hz bandwidth, and all 25 were visually correlated with the mode shapes found

experimentally. Figures 5.23 and 5.24 present an example of the visual correlation found for

CMP 3 of the FESF and Figures 5.25 and 5.26 present CMP 5 of the RESF.

99
 Figure 5.23: Third FESF experimental mode shape at 41.6 Hz

Figure 5.24: Third FESF computational mode shape at 30.24 Hz (Lam & Mechefske, 2020).

100
 Figure 5.25: Fifth RESF experimental mode shape at 89 Hz

Figure 5.26: Fifth RESF computational mode shape at 85.05 Hz (Lam & Mechefske, 2020).
101
 Once again, the POC was used to quantify the quality of correlation between the

computational and experimental mode shapes. For the 15 FESF mode shapes predicted, 14

showed acceptable correlation, with all diagonal terms above 0.91, and off diagonal terms below

0.2 except for the correlation value from mode 13 and experimental mode 11 at 0.292. There

were other high off diagonal terms including those found between modes 1 and 3, and modes 2

and 4. Due to the similarity in mode shapes, these off diagonal terms are not of concern,

especially when considering the magnitude of the diagonal values related to CMPs 1-4. CMP 12

was found to be erroneous with a diagonal term of 0.675 and an off diagonal term between

computational mode 11 and experimental mode 12 of nearly the same magnitude. There was a

relatively low density of accelerometers in the area of resonance, and the lack of resonant motion

elsewhere in the structure meant that slight discrepancies in mode shape data were exaggerated.

The POC matrix for the FESF is shown in Figure 5.27, while Figures 5.28 and 5.29 illustrate the

issues regarding the erroneous correlation of CMP 12.

Figure 5.27: POC computation for 15 correlated modes from the FESF of sub-assembly 2.

102
 Figure 5.28: Similarity between FESF mode 11 (Left) and FESF mode 12 (right) found
experimentally.

Figure 5.29: Predicted FESF modes 11 (Left) and 12 (Right) (Lam & Mechefske, 2020).

The computational model predicted 10 mode shapes for the RESF, and 10 were found

experimentally. All of these showed acceptable levels of correlation with all diagonal terms above

0.82 and most off diagonal terms below 0.3. Once again, the first three modes showed significant

off diagonal terms, but due to the similarity in mode shapes, these were not of significant

concern. In general, the correlation of RESF modes was not as strong as the correlation found for

the FESF mode shapes. It does however show strong enough correlation to confirm that the

correct natural frequencies are being compared. Figure 5.30 shows the POC matrix produced

from the comparison of RESF mode shapes.

103
 Figure 5.30: POC computation for 10 correlated modes from the RESF of sub-assembly 2.

5.8 Chapter Summary

This chapter provided an in-depth overview of the results produced by various validation

trials, as well as those used for comparison with the computational model. A rigorous set of

validation tests was implemented, including those recommended by the literature, and further

experimentation to determine the optimal experimental setup for the tests performed. These tests

confirmed the linearity of the test articles and repeatability of the experimental setup and

procedure. Additionally, the validity checks helped to determine the best response measurement

layout and confirmed that other experimental setup parameters would provide accurate results.

Once the test setup and procedure were validated, modal parameters including the natural

frequencies and mode shapes could be extracted with a high level of confidence. These

parameters were compared with those produced by the computational model formed by (Lam &

Mechefske, 2020). In the case of the first sub-assembly, it was found that modifications were

required to accurately model the dynamic behavior of the skin. Once this change was made to the
104
computational model, a strong level of correlation between the natural frequencies and mode

shapes was found. The second sub-assembly provided an acceptable level of correlation for

natural frequencies and mode shapes, however slightly larger discrepancies were noted. It was

determined that some incongruity was likely due to the computational model not accounting for

the pre-stress in the skin panels, however a significant amount was likely due to the boundary

conditions. A future amendment to the experimental setup was proposed to remedy this and

testing with the altered boundary conditions will likely improve correlation. With this phase of

the project complete, recommendations for future testing improvements can be made in chapter 6.

In addition, the next phases of the project including the expansion from sub-assemblies to a

complete half scale tail section will be discussed.

105
 Chapter 6

Conclusions and Future Recommendations

6.1 Summary and Conclusions

Noise within the cabin of aircraft negatively effects the comfort and overall flying

experience for passengers during flight. For business aircraft with jet engines mounted directly to

the fuselage, structure borne noise contributes the majority of that experienced in the passenger

cabin. Therefore, to mitigate the interior sound pressure level, it is first necessary to determine the

dynamic characteristics of these structural transmission paths at the rear of the aircraft. To

accomplish this, an extensive testing methodology was formed involving the experimental testing

and computational modelling of individual components before expanding to sub-assemblies, then

to the creation of a complete tail section model. The experimental testing presented in this thesis

was directed at obtaining the modal parameters from fuselage sub-assemblies, thus forming an

intermediate step between previous work performed on individual components and future tests on

a full tail section assembly

Prior to experimentation, a detailed literature review was performed to obtain the

necessary information pertaining to EMA of aerospace structures. This uncovered a multitude of

potential testing options, from which the final methodology was chosen. Additionally, relevant

background theory was reviewed, and provided the necessary education to perform a successful

modal experiment. Finally, this review uncovered information regarding data interpretation,

which would prove invaluable when analyzing and correlating the experimental data. Additional

pre-test work involved the construction of two fuselage sub-assemblies, the designs of which

were formed as sections of a full tail section model designed by (Chamberlain, 2018). The design

of the first assembly included a section of skin and a central bulkhead, such that a complete

experimental methodology could be formed. The second sub-assembly was created in accordance

106
with recommendations made by Chamberlain that a “Secondary larger subassembly consisting of

at least two separate components connected by stringers and skin such as the RESF and/or frame

rings.” be tested (Chamberlain, 2018). For both sub-assemblies, a review of the individual

components was presented, and recommendations for future manufacturing considerations were

provided.

The structures tested in this thesis represented a significant increase in size and

complexity over the components tested during the first phase of the project. As a result, it was

deemed necessary that a detailed experimental methodology be formed around the testing of

complex fuselage structures to obtain modal parameters from both sub-assemblies and in

preparation for the next stage of the project. The transition from impact hammer to modal shaker

introduced new testing parameters, some of which were selected using findings from the

aforementioned literature review. Others including the choice of stinger length and the decision to

use dummy masses were specific to this thesis and needed to be validated during preliminary

testing. Other validity checks including repeatability and linearity measurements ensured that

both the test setup and structures behaved as expected, and that the experimental results would be

reliable.

The results from the first sub-assembly indicated an issue with the computational model

when considering the skin modes of the system. It was found that the computational model

produced by (Lam & Mechefske, 2020). Under predicted the natural frequencies of these modes

by an average of 11.74 %. Multiple model updating techniques were trialed, and the discrepancies

between natural frequencies of the skin modes were reduced to an average of 1.17 %. The

bulkhead modes also showed an acceptable level of discrepancy in both preliminary and updated

versions of the computational model lending further confidence to the results produced. When

comparing both sets of mode shapes, a strong level of correlation was found for each CMP, thus

confirming that the correct natural frequencies were compared. It is therefore evident, that the

107
experimental results were essential to improving the computational model of sub-assembly one,

and after modifications were made, a strong level of correlation was found.

With regards to the second sub-assembly, natural frequencies and mode shapes of both

engine support frames were compared. Of the 25 CMPs 20 showed acceptable correlation in

natural frequency, and five were found to be erroneous. It was theorized that the boundary

conditions used to support the structure were imposing a stiffening effect on certain modes thus

raising their natural frequency. It was proposed that the boundary conditions be modified to

remedy this, and the testing involved to evaluate this change is included in the future work.

Twenty four of the twenty-five predicted mode shapes were found in the experimental data and

showed acceptable mathematical correlation. While not as strong as the correlation found with

sub-assembly 1, the results from the POC still indicated that the correct natural frequencies were

being compared. Finally, trends in the discrepancies between natural frequencies were discussed,

and potential model improvements were proposed to further reduce discrepancies between

experimental and computational results.

In conclusion, two primary objectives for this project were established, including the

formation of a comprehensive testing methodology and the production of detailed and reliable

experimental results were outlined. The experimental methodology described was successfully

implemented to extract the modal parameters including natural frequencies and mode shapes from

aircraft fuselage sub-assemblies. The extensive validity checks performed as part of this process

lent a high level of confidence to the results obtained. These results were then used as a point of

comparison for computational models, and where applicable, modifications were made to reduce

discrepancy. At this stage, the methodology formed can be implemented on larger fuselage

structures as described in the following section. It should be noted that with every modal test

comes a unique set of challenges to overcome, however the procedure outlined along with some

minor modifications if necessary, will provide acceptable experimental results.

108
6.2 Future Recommendations
The next stage of this project involves extending the test structure to a complete half

scale model of the tail section as shown in Figure 3.26. This would complete the direct structure

borne vibration paths between the jet engines and passenger cabin and provide a more

comprehensive picture of the dynamic behavior of this section of fuselage. Additionally, this

structure will facilitate the simulation of other noise generating components often placed in the

tail section, including hydraulic pumps and the auxiliary power unit. In addition to providing

experimental results for comparison with computational models, this structure would also

facilitate the evaluation of novel sound and vibration mitigation techniques, including those

proposed by (Warwick, Mechefske, & Kim, 2019).

Work in the more immediate term involves the modification to the boundary conditions

of sub-assembly 2 which involves moving the supporting bungee cords from the pylons to the

section of structure where the skin connects to the engine support frames, followed by a test of

this structure to evaluate the effects. It is expected that the erroneous natural frequencies will shift

as a result and improve correlation for the first three CMPs of the FESF and the first two CMPs

of the RESF. If this is not the case however, then it must be assumed as a result of the extensive

experimental validity checks, that the computational model to some extent is erroneous, but as a

result of the strong correlation found with the other CMPs of sub-assembly 2 and the correlation

found with sub-assembly 1, this is not likely to be the case. Future work with respects to the

computational side of this project are described in detail by (Lam & Mechefske, 2020).

109
 Appendix A
Sub-assembly One Experimental Mode shapes

Figure A. 1: Mode 1 – Oval skin @ 31.2Hz

Figure A. 2: Mode 2 – Bulkhead single drum @ 41.3 Hz

110
Figure A. 3: Mode 3 – Triangle skin 1 @ 69.6 Hz

Figure A. 4: Mode 4 – Triangle skin 2 @ 71.6 Hz

111
Figure A. 5 Mode 5 – Bulkhead halves 1 @ 83.5 Hz

Figure A. 6: Mode 6 – Bulkhead halves 2 @ 86.3 Hz

112
 Figure A. 7: Mode 7 – Square skin @ 105 Hz

Figure A. 8: Mode 7 – Five point skin / Bulkhead quarters 1 @ 138 Hz

113
 Figure A. 9: Bulkhead quarters 2 @ 142 Hz

Figure A. 10: Bulkhead double drum @ 156 Hz

114
Figure A. 11: 6 point skin 1 @ 170 Hz

Figure A. 12: 6 point skin 2 @ 172 Hz

115
 Appendix B
Sub-assembly Two Experimental Mode Shapes

Figure B. 1: FESF Mode 1 @ 21.0 Hz

Figure B. 2: FESF Mode 2 @ 33.8 Hz

116
Figure B. 3: FESF Mode 3 @ 41.6 Hz

Figure B. 4: FESF Mode 4 @ 46.8 Hz

117
Figure B. 5: FESF Mode 5 @ 66.3 Hz

Figure B. 6: FESF Mode 6 @ 78.8 Hz

118
Figure B. 7: FESF Mode 7 @ 96.2 Hz

Figure B. 8: FESF Mode 8 @ 115 Hz

119
 Figure B. 9: FESF Mode 9 @ 126 Hz

Figure B. 10: FESF Mode 10 @ 136 Hz

120
Figure B. 11: FESF Mode 11 @ 151 Hz

Figure B. 12: FESF Mode 12 @ 157 Hz

121
Figure B. 13: FESF Mode 13 @ 163 Hz

Figure B. 14: FESF Mode 14 @ 169 Hz

122
Figure B. 15: FESF Mode 15 @ 188 Hz

123
Figure B. 16: RESF Mode 1 @ 25.5 Hz

Figure B. 17: RESF Mode 2 @ 32.6 Hz

124
Figure B. 18: RESF Mode 3 @ 41.6 Hz

Figure B. 19: RESF Mode 4 @ 75.8 Hz

125
Figure B. 20: RESF Mode 5 @ 89.1 Hz

Figure B. 21: RESF Mode 6 @ 137 Hz

126
Figure B. 22: RESF Mode 7 @ 153 Hz

Figure B. 23: RESF Mode 8 @ 157 Hz

127
Figure B. 24: RESF Mode 9 @ 177 Hz

Figure B. 25: RESF Mode 10 @ 197 Hz

128
References

Agilent Technologies. (1997). The Fundamentals of Modal Testing. Agilent Technologies.

Avitable, P. (2018). Modal Testing A Practitioners Guide. Hoboken: John Wiley and Sons.

Brown, D. L., & Allemang, R. J. (2007). The Modern Era of Experimental Modal Analysis.

Cincinati: University of Cincinati.

Bruel & Kajer. (2008). Photon+ Dynamic Signal Analyser. Bruel and Kejer.

Chamberlain, D. A. (2018). Experimental Modal Analysis of a Half Scale Model Rear Twin-Engine

Mounted Aircraft Fuselage Section. Kingston: Queen's University.

Cusano, A., Capoluongo, P., Campopiano, S., Cutolo, A., Giordano, M., Felli, F., . . . Caponero,

M. (2006). Experimental Modal Analysis of an Aircraft Model Wing by Embedded Fiber

Bragg Grating Sensors. IEEE Sensors Journal Bolume 6, 67-77.

Dempsey, T. K., Leatherwood, J. D., & Clevenson, S. A. (1978). Developement of Noise and

Vibration Ride Comfort Criteria. Hampton, Virginia: National Aeronautics and Space

Administration.

Ewins, D. (2000). Modal Testing Theory, Practice and Application. Philedelphia: Research Studies

Press Ltd.

Fleming, G. A., & Buerhle, R. D. (n.d.). Modal Analysis of an Aircraft Fuselage Panel using

Experimental and Finite-Element Techniques. Hampton, Virginia: National Aeronautics

and Space Administration.

Goge, D., Boswald, M., Fullekrug, U., & Lubrina, P. (2007). Ground Vibration Testing of Large

Aircraft- State-of-the-Art and Future Perspectives. IMAC .

129
Gordon, R. W., Wolfe, H. F., & Talmadge, R. D. (1977). Modal Investigation of Lightweight

Aircraft Structures Using Digital Techniques. Dayton, Ohio: Air Force Flight Dynamics

Laboratory .

International Organization for Standardization. (1994). ISO 7626 Vibration and Shock-

Experimental Determination of Mechanical Mobility. Geneva: International Orginization

for Standardization.

Kehoe, M. W. (1995). A Historical Overview of Flight Flutter Testing. National Aeronautics and

Space Administration.

Lam, C., & Mechefske, C. (2020). COMPUTATIONAL MODAL ANALYSIS OF HALF SCALE

GENERIC BUSINESS JET SUBSTRUCTURES. International Design Engineering

Technical Conferences. St. Louis: American Society of Mechanical Engineers.

LDS-Dactron. (2004). RT Pro Dynamic Signal Analysis User Guide. Fremont, CA: LDS Test and

Measuremetn Ltd.

Mixson, J. S., & Kearny, B. C. (1978). Investigation of Interio Nois in a Twin-Engine Light Aircraft.

New York, N.Y: National Aeronautics and Space Administration; Columbia University.

National Instruments. (2017, December 14). National Instruments. Retrieved from Measuring

Vibration with Accelerometers: http://www.ni.com/white-paper/3807/en/#toc4

National Intruments. (2019). Specifications PXI-1031/PXI-1031DC. National Instruments.

PCB Piezotronics. (2020). Model 288D01. (PCB Piezotronics) Retrieved February 6, 2020, from

https://www.pcb.com/products?model=288D01

PCB Piezotronics. (2020). Model 352C22. (PCB Piezotronics) Retrieved January 14, 2020, from

https://www.pcb.com/products?model=352C22

130
Peres, M. A., Bono, R. W., & Avitable, P. (2012). Effects of shaker, stinger and transducer

mounting on measured frequency response functions. International Conference on Noise

and Vibration Engineering, (pp. 1475-1490). Leuven.

Peres, M., Bono, R., & Brown, D. (2010). Practical Aspects of Shaker Measurements for Modal

Testing. International Conference on Noise and Vobration Engineering. Leuven.

Vibrant Technology. (2015). ME'ScopeVES Reference Manual Basic Operations. Centennial, Co:

ibrant Technology.

Warwick, B. T., Kim, I. Y., & Mechefske, C. K. (2019). Substructuring verification of a rear

fuselage mounted twin-engine aircraft. Aerospace Science and Technology, 93.

Warwick, B. T., Mechefske, C. K., & Kim, I. Y. (2018). Effect of Stiffener Configuration on

Bulkhead Modal Parameters. International Design Engineering Technical Conferences.

Quebec City, Canada: The American Society of Mechanical Engineers.

Warwick, B. T., Mechefske, C. K., & Kim, I. Y. (2019). Topology optimization of a pre-stiffened

aircraft bulkhead. Structural and Multidisciplinary Optimization, 60(4), 1667-1685.

Wilby, J. F. (1995). Aircraft Interior Noise. Calabasas, CA: Academic Press Limited.

Wyen, T. A., Schoettelkotte, J. J., Perez, R. A., & Eason, T. (2017). Experimental Modal Analysis

of an Aircraft Fuselage Panel. Shock and Vibration, Aircraft/Aerospace, Energy

Harvesting, Acoustics and Optics, Volume 9 (p. 12). Dayton Ohio: Society For

Experimental Mechanics.

131