2. Physical testing fundamentals for model updating

Finite element model updating is only as trustworthy as the experimental data it consumes. Before any correlation metric, sensitivity matrix, or residual is computed, the test itself must produce frequency response functions (FRFs) that are physically meaningful, correctly referenced, and traceable to known excitation and response locations. This chapter establishes the measurement foundation: how modal testing is organized, what an FRF actually contains, and how each element of the measurement chain — exciter, transducer, conditioner, digitizer — shapes the data that later drives the updating problem.

What Modal Testing Delivers to the Updating Problem

Modal testing identifies the modal parameters of a structure — natural frequencies, mode shapes, and damping ratios — from measured excitation and response signals. For model updating, the deliverables are more specific than a generic modal report:

  • Resonance frequencies with quantified uncertainty, used as eigenvalue residuals.
  • Mode shapes at degrees of freedom (DOFs) that correspond to FE mesh nodes, used as eigenvector residuals after pairing and normalization.
  • Damping estimates, which are usually updated with lower confidence because damping models in FE codes are approximate.
  • FRF data sets retained in raw form, so that alternative identification methods or frequency bands can be revisited without repeating the test.

The critical constraint is spatial correspondence. A mode shape is only useful for updating if each measured DOF can be mapped to a specific FE node and direction. This mapping is decided during test planning, not after measurement.

The Measurement Chain and Where Errors Enter

An FRF is a ratio of response to excitation, but both quantities pass through a chain of physical and electronic transformations. Each stage contributes gain, phase, and noise that must be either calibrated out or bounded.

Stage Typical hardware Dominant error source Mitigation
Excitation Instrumented hammer, electrodynamic shaker Force drop-out, double hits, shaker–structure interaction Windowed impact, coherence check, low-mass stingers
Transduction Piezoelectric accelerometers Mass loading, transverse sensitivity, mounting resonance Select low-mass sensors, wax/stud mounting, calibrate on shaker
Conditioning Charge amplifier, IEPE supply Drift, incorrect sensitivity scaling, cable noise Verify sensitivity constants, low-noise cabling, AC coupling
Digitization DAQ with anti-alias filters Aliasing, quantization, trigger jitter Set fs ≥ 2.5 × fmax, enable anti-alias filter, stable trigger
Processing FFT analyzer / software Leakage, window bias, averaging errors Force/exponential windows, sufficient averages, overlap

Two conventions must be fixed before any FRF is interpreted. First, the sign convention for response direction: positive acceleration is defined along the sensor's sensitive axis, and this must match the FE coordinate system. Second, the FRF definition itself — receptance (displacement/force), mobility (velocity/force), or accelerance (acceleration/force). Mixing these across a test campaign produces apparent mode shape sign reversals that are artifacts, not physics.

Excitation Methods: Impact versus Shaker

Impact testing uses an instrumented hammer to apply a short force pulse. It is fast, requires no attachment, and excites a broad frequency band in a single hit. The usable bandwidth is set by the hammer tip: a hard steel tip concentrates energy at high frequency, while a soft rubber tip extends low-frequency content but limits the upper band. The force spectrum rolls off above the tip-dependent cutoff, so coherence drops there. Practical checks include inspecting the force pulse for a clean single peak (no double hit) and confirming that the coherence function stays above roughly 0.9 across the band of interest.

Shaker testing applies controlled, repeatable excitation — typically random, swept sine, or stepped sine. It offers better signal-to-noise ratio and allows averaging, but introduces two complications. The shaker and its stinger add mass and stiffness to the structure, shifting resonances; and the attachment point constrains the excitation to a single DOF. Swept-sine excitation gives the cleanest FRFs at the cost of longer test time, while random excitation is faster but requires more averaging to reduce variance.

For model updating, impact testing is often preferred for its speed and lack of mass loading, provided the structure is lightly damped and the frequency band of interest is well covered by the chosen tip. Use shaker testing when high modal density or significant damping demands better SNR.

From Time Signals to FRFs

The FRF is computed as the ratio of the cross-spectrum to the auto-spectrum of the force:

H(f) = S_fx(f) / S_ff(f)

where Sfx is the cross-spectral density between force and response, and Sff is the auto-spectral density of the force. This estimator (H1) assumes noise is present only on the response, which is the usual case for impact testing. When noise dominates the input, the H2 estimator (Sxx/Sxf) is used instead. The coherence function

γ²(f) = |S_fx(f)|² / (S_ff(f) · S_xx(f))

quantifies how linearly the response is related to the excitation at each frequency. Values near unity indicate a reliable FRF; low coherence signals leakage, nonlinearity, noise, or a double hit.

Measurement Chain: From Excitation to FRF Excitation Hammer / Shaker Force transducer Structure Test article Response DOFs Transducer Accelerometer Mass loading Conditioner Charge amp / IEPE Sensitivity scaling Digitizer / DAQ Anti-alias filter, f_s ≥ 2.5 f_max Trigger, averaging FFT Processing H(f) = S_fx / S_ff Coherence γ²(f) FRF Data Set Receptance / Accelerance → Modal identification Error Watchpoints • Double hit → low coherence • Mass loading → shifted f_n • Wrong sensitivity → gain error • Aliasing → false peaks • Leakage → biased damping • Sign convention mismatch → apparent mode reversal

Practical Test Setup and Verification

A reproducible modal test for model updating follows a defined sequence. The example below uses a simple cantilever beam to illustrate the workflow; the parameters are representative and intended for teaching, not as measured results.

  1. Define the DOF grid. Mark measurement points along the beam at, say, 10 equally spaced locations. Each point corresponds to a vertical (bending) DOF that maps to an FE node. Record the coordinate system: x along the beam, y vertical, positive upward.
  2. Select the accelerometer. A 5 g piezoelectric accelerometer with 100 mV/g sensitivity is appropriate for a light beam. Mount it with wax for quick surveys or a stud for permanent installations. Note the mounting resonance — wax typically limits usable bandwidth to about 5 kHz.
  3. Choose the hammer tip. For a first bending mode near 50 Hz and interest up to 1 kHz, a medium-hard plastic tip provides adequate energy across the band. Verify by inspecting the force spectrum.
  4. Rove or fix the sensor. In roving-hammer testing, the accelerometer stays at a fixed reference DOF while the hammer moves. This keeps the response chain constant and simplifies mass-loading correction. The alternative — roving accelerometer with fixed hammer — requires mass correction at every point.
  5. Acquire and average. Take at least 5 hits per point. Reject any hit with a double pulse or coherence below 0.9 in the band of interest. Apply a force window (to zero the post-pulse noise) and an exponential window to the response (to reduce leakage).
  6. Verify reciprocity. For a linear structure, the FRF measured with excitation at A and response at B should equal that with excitation at B and response at A. Checking one pair of off-diagonal FRFs is a powerful sanity check on the entire chain.
Do not proceed to modal identification or model updating if coherence is poor across the target band. Low coherence at resonance often indicates excessive damping or nonlinearity; low coherence between resonances usually indicates noise or leakage. Fixing the measurement is far cheaper than compensating for bad data in the updating algorithm.

Linking Test Data to the FE Model

The final step before updating is establishing the test–analysis correspondence. This means:

  • Pairing each measured mode with a predicted FE mode, typically by MAC (Modal Assurance Criterion) values above 0.7–0.8.
  • Scaling measured mode shapes to match the FE normalization (usually mass normalization).
  • Documenting the frequency band used for identification, since modes outside that band may be missed or poorly estimated.
  • Recording the FRF units (accelerance vs. receptance) and the sign convention, so that residuals are computed consistently.

With a clean FRF set, verified coherence, and a documented test–analysis mapping, the updating problem has a solid experimental foundation. The next chapter addresses how to formulate the residual and select parameters for updating.

← Previous 📖 Contents Next →