Sensors, FFT & Modal Testing

1. Introduction to Experimental Modal Analysis (EMA)

Experimental Modal Analysis (EMA) is the cornerstone of modern Noise, Vibration, and Harshness (NVH) engineering. While Finite Element Analysis (FEA) provides predictive simulations of structural dynamics, EMA serves as the ultimate arbiter of truth, validating numerical models, diagnosing noise and vibration path issues, and guiding structural optimization.

At its core, EMA is based on the system identification paradigm. By applying a known dynamic force excitation $f(t)$ to a structure and measuring its subsequent acceleration, velocity, or displacement response $x(t)$, we can identify the transfer characteristics of the system. In the frequency domain, this relationship is expressed as:

\[ Y(\omega) = H(\omega) X(\omega) \]

where:

•

$X(\omega)$ is the Fourier transform of the input excitation force.

•

$Y(\omega)$ is the Fourier transform of the output response (typically acceleration).

•

$H(\omega)$ is the Frequency Response Function (FRF), which contains the modal parameters:

•

natural frequencies (poles),

•

modal damping ratios (decay rates),

•

and mode shapes (residues).

In automotive development, EMA is performed at multiple stages:

•

Component level (e.g., testing suspension control arms, brackets),

•

Sub-assembly level (e.g., testing dressed engines, subframes, axles),

•

Body-in-White (BIW) level (e.g., testing the bare body structure),

•

and Trimmed Body or Full Vehicle level.

As we progress from component level to full vehicle level, the boundary conditions change from free-free (suspended by soft elastic cords) to fully constrained, and the structural damping increases from extremely low values ($0.1\%$ to $0.5\%$) to much higher levels ($2\%$ to $5\%$) due to joints, trim, and dampeners.

2. Transducer and Sensor Technology

Accurate modal testing begins at the physical boundary, where physical mechanical variables are transduced into electrical signals. The quality of the final modal model is fundamentally limited by the performance, selection, and placement of these sensors.

2.1 Accelerometers

Accelerometers are the most common sensors used to measure structural response in NVH. They measure the acceleration of a point on the structure, which is easily converted to velocity or displacement in the frequency domain by dividing by $i\omega$ or $-\omega^2$, respectively.

The two primary architectures utilized are:

•

Piezoelectric Accelerometers,

•

and MEMS (Micro-Electro-Mechanical Systems) Accelerometers.

Piezoelectric Accelerometers

Piezoelectric accelerometers rely on the piezoelectric effect of certain crystalline materials such as quartz, lead zirconate titanate (PZT), or bismuth titanate. These materials generate an electrical charge $q$ when subjected to mechanical strain, which is caused by the inertial force of a seismic mass $m_s$ acting on the crystal:

\[ q = d \cdot F = d \cdot m_s \cdot a \]

where:

•

$d$ is the piezoelectric charge constant.

•

$F$ is the dynamic force acting on the crystal.

•

$m_s$ is the seismic mass inside the accelerometer housing.

•

$a$ is the base acceleration.

Piezoelectric sensors exist in two main electrical configurations:

Charge Mode:

•

The sensor outputs raw charge directly from the crystal.

•

It requires highly insulated low-noise cables

and a high-impedance charge amplifier to convert the charge to a usable voltage.

•

While complex and susceptible to triboelectric cable noise

(electrical noise generated by cable flexing), charge mode sensors can operate at extremely high temperatures (up to $350^\circ\text{C}$ or more) because they contain no internal electronics.

IEPE (Integrated Electronics Piezo-Electric):

•

Also known by proprietary names like ICP,

these sensors feature internal miniature microelectronics (charge-to-voltage converters) directly behind the crystal.

•

They output a low-impedance voltage signal over standard coaxial cables,

making them highly robust against cable movement and electrical noise.

•

They are powered by a constant-current excitation (typically $2$ to $20\text{ mA}$)

superimposed on the signal line.

•

However, the internal electronics limit their maximum operating temperature

to approximately $120^\circ\text{C}$ to $180^\circ\text{C}$.

*Piezoelectric Design Styles:*

•

Compression Mode:

•

The crystal is sandwiched between the seismic mass and the base,

preloaded by a spring bolt.

•

This design is highly rigid but sensitive to base strain

(bending of the mounting surface) and thermal transients.

•

Shear Mode:

•

The crystal is subjected to shear deformation between a central post

and the seismic mass.

•

This design is highly isolated from base bending strain

and thermal expansion, making it the industry standard for high-accuracy modal testing.

MEMS Accelerometers

MEMS accelerometers consist of micro-machined silicon structures where a proof mass is suspended by silicon springs. Acceleration causes displacement of the proof mass, which is measured capacitively using differential finger structures:

\[ C = \epsilon \frac{A}{d_0 - x} - \epsilon \frac{A}{d_0 + x} \approx \frac{2 \epsilon A x}{d_0^2} \quad (\text{for } x \ll d_0) \]

MEMS accelerometers are capable of DC response (measuring static acceleration down to $0\text{ Hz}$), making them essential for:

•

low-frequency vehicle dynamics,

•

active suspension control,

•

and tilt sensing.

Their drawbacks include a higher noise floor (noise density) and lower high-frequency bandwidth compared to premium piezoelectric sensors.

Sensor Mounting Dynamics

The mounting interface acts as a spring in series with the accelerometer, forming a local mass-spring system. The stiffness of the mount $k_m$ and the accelerometer mass $m_s$ define a mounted resonance frequency:

\[ f_m = \frac{1}{2\pi} \sqrt{\frac{k_m}{m_s}} \]

The accelerometer behaves accurately only well below this resonance (typically up to $f_m / 3$). Different mounting techniques change $k_m$, drastically altering the sensor's usable frequency range:

Mounting Method	Typical High Frequency Limit (Hz)	Characteristics & Practical Considerations
:---	:---	:---
Stud Mounting	10,000 - 20,000	Best performance. Requires drilling/tapping the structure. Maximum contact stiffness.
Cyanoacrylate (Superglue)	5,000 - 8,000	Excellent high-frequency coupling. Fast and semi-permanent. Risk of sensor damage during removal.
Beeswax	3,000 - 5,000	Highly convenient for clean laboratory testing. Softens at elevated temperatures, destroying high-frequency transmission.
Magnetic Base	1,500 - 2,500	Convenient for steel/iron structures. The magnetic mass acts as an adder, lowering the mounted resonance.
Handheld Probe	< 500 - 1,000	Poor repeatability. Low and variable contact stiffness. Highly operator-dependent.

Transverse Sensitivity

An accelerometer should ideally respond only to acceleration along its primary axis. However, due to manufacturing tolerances and crystal misalignment, it will exhibit a small response to cross-axis excitation, known as transverse sensitivity (typically expressed as a percentage, $1\%$ to $5\%$). In multi-axis modal testing, high transverse sensitivity introduces cross-talk errors, leading to distorted mode shape estimations.

Sensor Calibration

Sensors must be calibrated regularly. The standard method is the comparison method (or back-to-back calibration) according to ISO 16063-21. The sensor under test is mounted directly on top of a reference standard accelerometer with a traceable calibration. Both are excited on a calibration shaker, and the ratio of their outputs yields the sensitivity of the test sensor. Alternatively, a handheld calibrator (e.g. producing $1\text{ g}$ RMS at $159.15\text{ Hz}$) can be used for quick field checks.

2.2 Impact Hammers

An impact hammer is an excitation device that provides a transient force input to the structure. It contains a force transducer (load cell) built into the hammer head behind the strike tip.

Excitation Bandwidth and Tip Selection

The force impulse has a duration $\Delta t_c$ (contact time) governed by the contact stiffness between the hammer tip and the structure, and the mass of the hammer. The frequency spectrum of this impulse is flat at low frequencies and rolls off, having its first zero-crossing at approximately:

\[ f_0 \approx \frac{1.5}{\Delta t_c} \]

To control this excitation bandwidth, engineers swap the hammer tip:

•

Hard Tip (Steel, Ceramic):

•

High contact stiffness $\implies$ short contact duration.

•

Wide frequency bandwidth (excitation up to $10\text{ kHz}$).

•

Used for small, stiff steel components.

•

Medium Tip (Plastic, Hard Rubber):

•

Medium contact stiffness $\implies$ medium bandwidth (excitation up to $1.5\text{ kHz}$).

•

Soft Tip (Soft Rubber):

•

Low contact stiffness $\implies$ long contact duration.

•

Narrow frequency bandwidth (excitation concentrated below $200\text{ Hz}$).

•

Essential for large, compliant structures (like a fully dressed BIW or composite panels)

to avoid structural damage and to inject enough energy at low frequencies.

Hammer Double Hitting

Double hitting occurs when the hammer rebounds and strikes the structure a second time within a single measurement window. This creates a comb-filter effect in the frequency domain, resulting in zero-energy nodes in the excitation spectrum. Any frequency where the input force spectrum has a notch will have an extremely low Signal-to-Noise Ratio (SNR), corrupting the FRF at that frequency. Double hits must be rejected by the data acquisition system.

Force and Exponential Windowing

To improve signal quality in transient testing, specific windows are applied:

•

Force Window:

•

A rectangular-like window centered on the impact.

•

It retains the force signal during the impact

and sets the rest of the signal to absolute zero, eliminating background electrical noise or hammer swing noise post-impact.

•

Exponential Window:

•

Applied to the response channel.

•

Since light damping in structures can cause the response to ring

past the end of the acquisition time window $T$, an artificial exponential decay $e^{-\beta t}$ is applied to force the signal to zero at $t = T$, preventing leakage.

•

The added artificial damping must be mathematically subtracted

from the identified modal damping during post-processing:

\[ \zeta_{actual} = \zeta_{measured} - \frac{\beta}{\omega_n} \]

3. Inline Technical Diagrams

Figure 12.1: Internal structural layout of an inertial piezoelectric sensor under vertical base excitation.

Figure 12.2: How multiplying a non-periodic time capture by a Hanning window prevents spectral leakage by tapering the boundary discontinuities to zero.

4. Digital Signal Processing Fundamentals

Experimental data captured by sensors is continuous and analog. Before performing digital analysis, the data must undergo digitization and signal conditioning.

4.1 The Shannon-Nyquist Sampling Theorem

To reconstruct an analog signal accurately without distortion, the sampling frequency $f_s$ must be strictly greater than twice the highest frequency component $f_{max}$ present in the signal:

\[ f_s > 2 f_{max} \]

The limit $f_{Nyq} = f_s / 2$ is the Nyquist Frequency. If the signal contains energy at a frequency $f_{high} > f_{Nyq}$, this energy is falsely folded (aliased) back into the measurement band at an apparent frequency:

\[ f_{alias} = |2 \cdot k \cdot f_{Nyq} - f_{high}| \]

To prevent aliasing, hardware analyzers employ Anti-Aliasing Filters (AAF). These are analog low-pass filters (typically steep Chebyshev or Elliptic filters) that attenuate all frequencies above the Nyquist limit before the analog-to-digital converter (ADC). Modern analyzers utilize high-speed delta-sigma ADCs with extremely high oversampling, allowing for highly efficient digital filtering to prevent aliasing.

4.2 Spectral Resolution & Leakage

The relationship between sample parameters is governed by the Fourier transform structure:

\[ T = N \cdot \Delta t = \frac{1}{\Delta f} \]

•

$\Delta f$: The frequency resolution (spacing between spectral bins).

•

$T$: The time length of the signal record (block size).

•

$N$: The number of time samples.

If the signal contains a frequency component that does not match an integer multiple of $\Delta f$, it will not fit perfectly within the time record $T$. When the DFT periodic extension is applied, a sharp step discontinuity appears at the boundaries. In the frequency domain, this discontinuity spreads energy across multiple adjacent bins. This phenomenon is known as spectral leakage.

4.3 Windowing Functions

To reduce spectral leakage, the time history is multiplied by a windowing function $w(t)$ that smoothly tapers the signal to zero at the boundaries:

Hanning (Hann) Window:

\[ w(t) = 0.5 \left[ 1 - \cos\left( \frac{2\pi t}{T} \right) \right] \]

•

*Characteristics*: Excellent for general continuous vibration signals,

offering a good balance between amplitude accuracy and frequency resolution. Sidelobe roll-off is $60\text{ dB/octave}$.

Hamming Window:

\[ w(t) = 0.54 - 0.46 \cos\left( \frac{2\pi t}{T} \right) \]

•

*Characteristics*: Similar to Hann,

but optimizes the height of the first sidelobe at the expense of subsequent roll-off.

Flat-Top Window:

\[ w(t) = a_0 - a_1 \cos\left(\frac{2\pi t}{T}\right) + a_2 \cos\left(\frac{4\pi t}{T}\right) - a_3 \cos\left(\frac{6\pi t}{T}\right) \]

where:

•

$a_0 = 0.21557895$

•

$a_1 = 0.41663158$

•

$a_2 = 0.277263158$

•

$a_3 = 0.083594737$

•

*Characteristics*: The mainlobe is wide and flat.

It provides extremely high amplitude accuracy (error $<0.01\text{ dB}$) at the expense of frequency resolution. Excellent for calibration.

Force/Exponential Windows:

•

Used in impact testing.

•

The force window isolates the impact,

and the exponential window forces the structural response to decay to zero within the record time.