Linear Static FEA to Strain Gauge Correlation

What This Article Covers

This article is the flagship of the Test & Analysis Correlation knowledge base. It brings together every discipline discussed in the preceding articles — test design, strain gauge fundamentals, gauge placement and orientation, rosette measurement — and combines them with the finite element extraction and comparison methods discussed in the following articles into a single, end-to-end engineering workflow. The goal is to take an engineer who has a linear static FE model on one desk and a strain-gauged test article on another, and to walk them through every step required to produce a meaningful, defensible correlation between the two. The workflow is presented as twenty sequential steps, grouped into phases, with a practical checklist that should be applied to every gauge. This is not a theoretical treatment — it is a working procedure, engineer to engineer, for the real situation you face when the test data and the FE model are on your desk and you need to decide whether the model is good enough to trust.

The Central Principle

Before any step of the workflow, the engineer must internalise the single principle that governs all strain correlation. Every step that follows exists to satisfy this principle, and every shortcut taken away from it degrades the correlation. The principle is stated in capital letters because it is not a suggestion — it is the definition of a valid comparison.

THE MOST USEFUL STRAIN CORRELATION COMPARES THE SAME PHYSICAL QUANTITY, AT THE SAME LOCATION, IN THE SAME DIRECTION, OVER THE SAME LOAD STATE.

The Correlation Hierarchy

Strain correlation is the most local level of a hierarchy. Before the strain at a gauge is interpreted, the levels above it must be checked. The load must be correct, the reactions must balance, the global deflection must match — only then does a local strain comparison become a meaningful test of the model rather than a potential coincidence of compensating errors. The hierarchy is read from the top down, and a failure at any level invalidates the interpretation of the levels below.

Correlation hierarchy — read from the top down:

Level 1  LOAD
         Is the applied test load correct, and is the same load
         represented in the FE model (magnitude, direction, point)?
             ↓
Level 2  REACTION
         Do measured support reactions match applied loads?
         Do FE reactions match the applied FE load?
             ↓
Level 3  GLOBAL STIFFNESS
         Does the measured deflection shape and magnitude
         match the FE prediction? Is the stiffness right?
             ↓
Level 4  REGIONAL STRAIN
         Do strains along the primary load path match in
         pattern and magnitude across multiple gauges?
             ↓
Level 5  LOCAL DETAIL
         Do peak strains at features, joints and concentrations
         match? This is the finest level of correlation.

GLOBAL CORRELATION SHOULD GENERALLY BE ESTABLISHED BEFORE
LOCAL CORRELATION IS INTERPRETED.

The Strain-Gauge Correlation Workflow

The following diagram summarises the complete workflow from test configuration definition to documentation of residual uncertainty. The twenty steps that follow expand each box in detail. The workflow is sequential — each step builds on the previous, and skipping a step leaves a gap that may not be visible in the final comparison but that undermines its validity. The engineer should treat this workflow as a checklist for every correlation exercise, not as a menu from which to pick convenient steps.

Strain-gauge / FEA correlation workflow:

  1. Define physical test condition
         ↓
  2. Confirm linear-elastic response
         ↓
  3. Match the load (incl. multi-axis superposition)
         ↓
  4. Match gauge location (FE extraction point)
         ↓
  5. Match gauge direction (strain transformation)
         ↓
  6. Represent physical gauge length (spatial averaging)
         ↓
  7. Use correct surface (shell top/bottom vs mid-plane)
         ↓
  8. Account for bending (opposite-face decomposition)
         ↓
  9. Check zero, preload and datum
         ↓
 10. Temperature effects (thermal output)
         ↓
 11. Compare load vs strain (not just endpoint)
         ↓
 12. Use normalised strain response (με/kN)
         ↓
 13. Compare multiple gauges as pattern (scatter plot)
         ↓
 14. Use strain ratios where useful
         ↓
 15. Check sign
         ↓
 16. Check repeatability
         ↓
 17. Correlate reactions and displacements too
         ↓
 18. Interpret differences physically
         ↓
 19. Update model carefully
         ↓
 20. Document the correlation

Phase 1 — Steps 1 to 3: Establish the Common Problem

The first phase ensures that the test and the FE model are solving the same problem. If the load, the boundary condition or the response regime differs between test and model, no amount of careful strain extraction can rescue the correlation — the two are different problems, and any agreement is fortuitous. This phase is the foundation; errors here propagate through everything that follows.

Step 1 — Define the Physical Test Condition

Before any comparison, the engineer must define exactly what physical condition is being correlated. What is the test article — the specific hardware, with its specific as-built dimensions and material properties? What is the load case — the magnitude, direction, point of application and number of loads? What is the boundary condition — the supports, the fixture, the constraints? What is the environment — the temperature, the humidity? What is the load state being compared — the maximum load, an intermediate level, the zero/preload state? The FE model must be set up to represent this exact condition. If the FE model was built for a design load case that differs from the test condition, it must be re-run with the test load, test boundary and test geometry before correlation. Comparing the test against an FE model of a different condition is the most fundamental error — it compares two different problems.

CONSTRAINT: The FE model must represent the actual tested condition — the same article geometry, the same load, the same boundary, the same environment. Re-run the model for the test condition before correlating, or the comparison is meaningless.

Step 2 — Confirm Linear-Elastic Response

This workflow assumes a linear static FE model. For the correlation to be valid, both the FE model and the test structure must be behaving linearly and elastically at the load level being compared. The FE model is linear by construction — it uses linear material properties and small-displacement kinematics. The test structure may not be. If the test article yields, if joints slip, if contact opens, if the deflection is large enough to change the load path, the test response is nonlinear and the linear FE model is not valid for comparison. The linearity is checked by the load ladder (Step 11 and the load-scaling article): if the strain scales proportionally with load from zero to the maximum, the response is linear; if it bends away, it is not. If the structure is nonlinear at the test load, the correlation must either be performed at a lower load where linearity holds, or a nonlinear FE model must be built. A linear FE model correlated against a nonlinear test produces discrepancies that are physically meaningless — they measure the nonlinearity, not the model error.

Step 3 — Match the Load

The FE load must match the test load — same magnitude, same direction, same point of application. If the test applies 50 kN downward at a fitting and the FE model applies 50 kN at the same fitting but in a slightly different direction or through a different load-introduction area, the strain field near the fitting will differ, and the correlation will measure the load difference, not the model quality. For multi-axis or multi-point loads, each load component must be matched. Where the test applies several loads simultaneously, the FE model must apply the same combination — or, if the structure is linear (Step 2), the FE results for individual load cases can be superposed to represent the combined test load. The superposition must use the correct load factors: the FE result for each load case is multiplied by the ratio of the test load to the FE reference load for that case, and the scaled results are summed. This is valid only for linear response; for nonlinear response, the combined load case must be run directly.

Load scaling for a single load case:

ε_pred  =  ε_ref  ×  (F_test / F_ref)

where:
ε_pred  =  FE strain predicted at the test load
ε_ref   =  FE strain at the reference load used in the analysis
F_test  =  actual test load
F_ref   =  reference load in the FE analysis

For multi-axis superposition (linear response only):

ε_pred  =  Σ  ε_ref,i  ×  (F_test,i / F_ref,i)

where the sum is over all independent load cases i.

Phase 2 — Steps 4 to 8: Extract Equivalent FE Strain

The second phase is the heart of the correlation: extracting from the FE model a strain value that represents exactly what the physical strain gauge measures. This is where most correlations go wrong, by comparing the gauge reading against an FE value that does not represent the same quantity. Each step in this phase closes a specific gap between the FE strain field and the physical measurement.

Step 4 — Match Gauge Location

The FE strain must be extracted at the physical location of the gauge centre, not at the nearest node or the nearest element centroid by default. The gauge has a specific position on the structure surface, defined by 3D coordinates. The FE strain at that position is found by identifying the element that contains the gauge location and interpolating the strain field to that point. Most post-processors support probe or query tools that interpolate to a specified coordinate. If the gauge is near an element boundary, the interpolation should use the elements on the correct side — the side corresponding to the physical surface. Using the nearest node strain, or the nearest element centroid strain, introduces a positional error that depends on the mesh density and the strain gradient. In a fine mesh with a low gradient, the error is small; in a coarse mesh or a high gradient, it can be significant. The golden rule is: extract at the gauge centre, by interpolation, every time.

VERIFICATION: Confirm that the FE extraction point is on the correct element face, on the correct side of the mesh, corresponding to the physical surface where the gauge is bonded. An extraction on the wrong face of a shell gives the opposite bending strain.

Step 5 — Match Gauge Direction (Strain Transformation)

The FE strain at the gauge location is a tensor — it has components in the element or global coordinate system. The gauge measures the extensional strain in a single direction: the gauge grid direction. The FE tensor must be transformed into the gauge direction using the strain transformation equation. The gauge direction is defined by the angle θ recorded during installation, measured from a reference axis that must be identifiable in the FE coordinate system. If the reference axis on the structure does not correspond to an axis in the FE model, the transformation angle is wrong, and the comparison is invalid. The engineer must establish the correspondence between the test reference axis and the FE coordinate system before transforming. For a rosette, each of the three gauge directions is transformed separately, giving three FE values to compare against the three measured values.

Strain transformation into gauge direction:

εθ  =  (εx + εy) / 2  +  (εx − εy) / 2 · cos(2θ)  +  (γxy / 2) · sin(2θ)

where:
εθ   =  FE strain in the gauge direction (to compare with gauge reading)
εx   =  FE normal strain in FE x-direction at gauge location
εy   =  FE normal strain in FE y-direction at gauge location
γxy  =  FE engineering shear strain at gauge location
θ    =  angle from FE x-axis to gauge direction
        (must be consistent with the test reference axis)

Step 6 — Represent Physical Gauge Length (Spatial Averaging)

A strain gauge measures the average strain over its grid length, not the point strain at its centre. In a uniform strain field, the average equals the point value, and this step is unnecessary. In a strain gradient, the average over the gauge length differs from the point value, and the FE strain must be averaged over the gauge length to represent the measurement. The averaging is performed by integrating the FE strain in the gauge direction along the gauge length, or approximately by sampling the FE strain at several points along the gauge length and taking the mean. The need for averaging depends on the strain gradient and the gauge length: a 5 mm gauge in a region where the strain varies by 5% over 5 mm needs averaging; a 5 mm gauge in a uniform field does not. The engineer should examine the FE strain contour along the gauge length and judge whether the variation is significant. If it is, the averaged value should be used; if it is not, the point value is adequate. This step is frequently omitted, and in high-gradient regions it is a common source of systematic discrepancy.

Gauge-length averaging:

ε_gauge  ≈  (1 / L_g)  ·  ∫  εθ(s)  ds      (integrated over gauge length)

or approximately by sampling:

ε_gauge  ≈  (1/n)  ·  Σ  εθ(s_i)            (average of n points along gauge)

where:
ε_gauge  =  FE strain representing the gauge measurement
L_g      =  physical gauge length (grid length)
εθ(s)    =  FE strain in gauge direction at position s along the gauge
s_i      =  sample points along the gauge length

Use when the strain varies significantly over L_g.

Gauge-length averaging in a strain gradient:

  Strain
   ↑
   |      ·  point value at centre
   |    / · \
   |  /  ·   \      FE strain varies along gauge length
   |/    ·     \
   |------·-------→  distance along gauge
   0      ↑       L_g
          |
     ε_gauge = average over L_g (shaded region)

  If strain is uniform:  ε_gauge ≈ point value (no averaging needed)
  If strain has gradient: ε_gauge < peak (averaging matters)

Step 7 — Use the Correct Surface

For a shell element model, the strain varies through the thickness: the top surface, the bottom surface and the mid-plane have different strains when bending is present. A strain gauge is bonded to a physical surface — the top or the bottom of the real structure. The FE strain must be extracted from the corresponding surface of the shell element. Most post-processors output shell strain at the top surface, the bottom surface, or the mid-plane, and some output the membrane and bending components separately. The engineer must confirm which surface the post-processor is reporting and ensure it matches the physical gauge surface. Extracting the mid-plane strain when the gauge is on the top surface gives the membrane strain without the bending contribution — wrong by the entire bending component. For solid element models, the gauge is on the surface of the solid, and the strain at the surface element face should be used; the interior strain is not the surface strain. This step is simple but critical, and it is a common error when the post-processor defaults to mid-plane or to the "top" surface that may not correspond to the physical top.

MISTAKE: Extracting shell mid-plane (membrane) strain for a gauge on the physical top surface. The mid-plane strain omits the bending component, which may be the dominant part of the measurement. Always extract from the surface corresponding to the physical gauge location.

Step 8 — Account for Bending (Opposite-Face Decomposition)

When a gauge is on one face of a beam-like or plate-like member, the measured strain is the sum of membrane and bending strain. The FE prediction at that surface is also the sum of membrane and bending. If the correlation shows a discrepancy, it is not clear whether the membrane prediction, the bending prediction, or both are wrong. If gauges are placed on both faces at the same location and direction, the test readings can be decomposed into membrane and bending components, and the FE prediction can be decomposed the same way. The membrane and bending components can then be correlated separately. This is far more diagnostic than correlating the combined surface strain: if the membrane matches but the bending does not, the model's bending stiffness or moment arm is suspect; if the bending matches but the membrane does not, the load magnitude or axial stiffness is suspect. Wherever the structure and access permit, opposite-face gauge pairs should be used at critical sections.

Opposite-face decomposition:

Test:    ε_top,test    =  ε_mem,test  +  ε_bend,test
         ε_bot,test    =  ε_mem,test  −  ε_bend,test

         ε_mem,test   =  (ε_top,test + ε_bot,test) / 2
         ε_bend,test  =  (ε_top,test − ε_bot,test) / 2

FE:      ε_mem,FE      =  shell membrane strain at that location
         ε_bend,FE     =  shell bending strain at that surface

Correlate ε_mem,test vs ε_mem,FE  and  ε_bend,test vs ε_bend,FE  separately.

Opposite-face gauge pair on a beam in bending + tension:

       Top face:    ε_top  =  ε_mem + ε_bend   (tension + tension)
       ───────────────────────
              ↕ thickness
       ───────────────────────
       Bottom face: ε_bot  =  ε_mem − ε_bend   (tension − compression)

  Membrane:  ε_mem  = (ε_top + ε_bot) / 2
  Bending:   ε_bend = (ε_top − ε_bot) / 2

  Correlate membrane and bending separately to diagnose
  whether the discrepancy is in axial stiffness or bending.

Phase 3 — Steps 9 to 12: Ensure the Measurement and Reference Are Clean

The third phase addresses the practical conditions that can contaminate the comparison: the zero state, the temperature, the load path and the normalisation. These are not extraction steps — they are sanity checks that ensure the numbers being compared are physically meaningful and consistently referenced.

Step 9 — Check Zero, Preload and Datum

Both the test strain and the FE strain must be referenced to the same zero state. In the test, the zero is taken at some defined condition — typically the article installed in the fixture with no applied load, after preloading and settling. In the FE model, the zero is the undeformed state. If the test article is preloaded by the fixture, by gravity or by assembly loads before the test load is applied, and if the FE model does not include those preloads, the test strain includes a pre-strain that the FE strain does not, and the comparison is offset. The engineer must either ensure that the test zero is taken after all preloads have settled, or include the preloads in the FE model so that the FE strain is also referenced to the pre-loaded state. The datum must be consistent: both test and FE must report the strain change due to the applied test load, relative to the same starting condition. A mismatch in datum produces a constant offset across all gauges that can mimic a systematic error.

CONSIDERATION: The test zero and the FE zero must represent the same physical state. If the test article is preloaded by fixture, gravity or assembly, either zero the test after preload settles, or include the preload in the FE model.

Step 10 — Temperature Effects

If the test temperature differs from the temperature at which the gauges were zeroed, the gauge readings include apparent thermal strain — the thermal output of the gauge (see the strain gauge fundamentals article). This apparent strain is not predicted by the mechanical FE model and will appear as a discrepancy. The thermal output must be subtracted from the gauge reading using the manufacturer's thermal output curve for the gauge type and substrate, or cancelled by a dummy-gauge bridge configuration. If the structure itself is at a different temperature than the FE model assumes, the material properties (modulus, thermal expansion) may also differ, and the FE model should use the properties at the test temperature. In most laboratory correlation tests, temperature is controlled and this step is minor; in environmental or outdoor tests, it can be the dominant source of discrepancy. The engineer should always record the test temperature and check whether thermal output is significant relative to the mechanical strain being measured.

• Record the temperature at which the gauges were zeroed and at which the test was run
• Subtract gauge thermal output using the manufacturer's curve if temperature changed
• Use a dummy gauge on unstrained same-material substrate at the same temperature if possible
• Use FE material properties at the test temperature if the temperature differs from nominal
• Check that the apparent thermal strain is small relative to the mechanical strain of interest

Step 11 — Compare Load Versus Strain, Not Just the Endpoint

A correlation that compares only the strain at the maximum load misses the most powerful diagnostic: the load–strain relationship. In a linear elastic structure, strain is proportional to load, and the plot of strain against load should be a straight line through the origin. The FE model assumes this linearity. If the test load–strain plot is linear and passes through the origin, the structure is behaving as the model assumes, and the endpoint comparison is valid. If the plot is curved, the structure is nonlinear — yielding, joint slip, contact — and the linear model is not valid for comparison. If the plot is linear but does not pass through the zero taken at the start, there is a preload or zero offset. If the plot is linear but the slope differs from the FE prediction, the stiffness is wrong. Comparing the full load–strain curve, not just the endpoint, exposes all of these. The test should always be run as a load ladder with multiple levels.

Load–strain linearity check:

  Strain
   ↑          · FE prediction (straight line through origin)
   |        /
   |      /    · test (linear, matches FE — good)
   |    /    /
   |  /    /
   |/    /
   |   / · test (nonlinear — structure yielding or slipping)
   |  /
   | /· test (linear, offset — preload or zero error)
   |/___________→ Load
   0

  Linear through origin  → valid for linear FE comparison
  Nonlinear               → linear FE not valid; investigate nonlinearity
  Linear but offset       → check preload, zero datum, thermal output

Step 12 — Use Normalised Strain Response (microstrain per kN)

Comparing absolute strain at a single load level conflates two things: the structural response (strain per unit load) and the load magnitude. If the test load and the FE load differ slightly — and they often do, because of load-cell calibration or load-introduction differences — the absolute strains differ even if the structural response is identical. Normalising the strain by the load removes this conflation. The normalised strain — microstrain per kN — is a property of the structure and the gauge location, independent of the exact load magnitude. The test normalised strain is the slope of the load–strain line (from Step 11). The FE normalised strain is the FE strain divided by the FE load. Comparing the normalised values isolates the structural response from the load magnitude and is the most robust single-number comparison. It also allows comparison across different load levels and different test runs.

Normalised strain response:

ε_norm,test  =  ε_test / F_test        (test, from slope of load–strain line)
ε_norm,FE   =  ε_FE  / F_FE          (FE, strain divided by FE load)

Compare ε_norm,test  vs  ε_norm,FE  (both in με/kN).

Ratio  =  ε_norm,test / ε_norm,FE

A ratio of 1.0 means the structural response matches perfectly,
independent of any difference in load magnitude.

Phase 4 — Steps 13 to 16: Read the Correlation as a Pattern

The fourth phase moves from the single-gauge comparison to the pattern across all gauges. A single gauge comparison is a data point; the pattern across many gauges is the evidence. This phase is about presenting and reading that pattern correctly.

Step 13 — Compare Multiple Gauges as a Pattern (Scatter Plot)

The single most powerful presentation of strain correlation is the measured-versus-predicted scatter plot. Each gauge is one point: the x-coordinate is the FE predicted strain, the y-coordinate is the measured test strain. If the model were perfect, all points would lie on the 45° line (y = x). The scatter about this line, the systematic offset, the slope of the best-fit line and the outliers all carry diagnostic information. A uniform offset suggests a load or modulus error. A slope different from 1.0 suggests a stiffness error that scales with strain magnitude. A single outlier suggests a local gauge or modelling problem. A clustered pattern of outliers suggests a regional model deficiency. This plot should be the primary tool for presenting and discussing the correlation — it shows the whole pattern at a glance and focuses attention on the gauges and regions that need investigation.

Measured vs predicted strain scatter plot:

  Measured strain (test)
   ↑
   |            ·  y = x line (perfect correlation)
   |          /
   |        /  ·
   |      /  ·    ·
   |    / ·         ·
   |  / ·     ·         ·
   |/ ·          ·          ·
   |——————————————→ Predicted strain (FE)
   0

  Points on the line      → model matches test
  All points above the line → test consistently higher (load, modulus, stiffness)
  Slope ≠ 1.0             → stiffness error scaling with strain
  Single outlier          → local gauge or modelling issue
  Clustered outliers      → regional model deficiency
  Random scatter          → measurement noise or local detail

Step 14 — Use Strain Ratios Where Useful

In addition to the absolute and normalised comparisons, strain ratios between gauges can be diagnostic. The ratio of strain at two gauges on the same load path is independent of load magnitude and depends only on the structural load distribution. If the FE model predicts a ratio of 2.0 between two gauges and the test shows a ratio of 1.5, the load distribution in the model is wrong, even if the absolute magnitudes could be matched by adjusting the load. Ratios are particularly useful for gauges that are on the same member but at different positions, or on symmetric positions that should carry equal load. A symmetric pair that should read identically but does not flags an asymmetry — in the load, the boundary, or the structure — that the model may or may not capture. Ratios strip out the load magnitude and focus on the structural behaviour.

• Compare strain ratios between gauges on the same load path — independent of load magnitude
• A ratio mismatch means the load distribution is wrong, not just the load magnitude
• Use symmetric gauge pairs — equal readings expected; inequality flags asymmetry
• Compare the measured ratio to the FE predicted ratio directly
• Ratios are most diagnostic when the gauges are far enough apart that the ratio is not trivially 1.0

Step 15 — Check Sign

The sign of the measured strain must match the sign of the predicted strain. A sign error is the most obvious and the most informative discrepancy: it means the gauge is measuring compression where the model predicts tension, or vice versa. A sign error at a single gauge usually means a gauge wiring error, a gauge direction recording error, or a local load reversal that the model does not capture. A sign error at many gauges usually means the load is applied in the wrong direction in the test or the model, or the coordinate systems are inconsistent. Sign errors should be checked first, before any magnitude comparison, because a sign error invalidates the magnitude comparison — a gauge reading −500 με compared to a prediction of +500 με is not a "1000 microstrain error", it is a sign error that must be resolved before the magnitudes are discussed.

VERIFICATION: Check the sign of every gauge against the prediction before comparing magnitudes. A sign error means the comparison is meaningless until the direction is reconciled — resolve wiring, orientation recording, load direction and coordinate system first.

Step 16 — Check Repeatability

A strain gauge that does not repeat — that gives a different reading on the second application of the same load — is not producing a reliable measurement. Repeatability is checked by unloading and reloading to the same level and comparing the readings. If the gauge returns to the same value, it is repeatable. If it drifts, if it shows hysteresis (a different path on unloading than on loading), or if it does not return to zero on unloading, the measurement is compromised. Drift may indicate bonding problems, temperature change, or zero instability. Hysteresis may indicate joint slip, material micro-yield, or gauge installation issues. Non-repeatable gauges should be flagged and excluded from the correlation, or used only with a stated uncertainty that reflects the non-repeatability. A correlation based on non-repeatable data is a correlation of noise.

VERIFICATION: Every gauge should be checked for repeatability by unload–reload. Non-repeating gauges indicate bonding, temperature, joint or zero problems and should be flagged before the correlation is interpreted.

Phase 5 — Steps 17 to 20: Contextualise, Interpret and Document

The final phase places the strain correlation in its full context — reactions and deflection, physical interpretation, model updating and documentation. Strain correlation in isolation is incomplete; combined with the global checks and a physical interpretation of discrepancies, it becomes a complete engineering argument.

Step 17 — Correlate Reactions and Displacements Too

Strain correlation is the local level of the hierarchy. The global levels — reactions and displacements — must also be correlated, for two reasons. First, they are part of the evidence: if the reactions and deflection match, the global model is supported, and the strain correlation is testing the local detail. If the reactions or deflection do not match, the global model is wrong, and local strain correlation is premature. Second, they are diagnostic: a strain discrepancy combined with a deflection discrepancy points to a stiffness error; a strain discrepancy with correct deflection points to a local modelling issue. The reaction check is the simplest — the sum of measured reactions should equal the applied load, and the FE reactions should match. The deflection check compares measured displacements at LVDT or photogrammetry points against the FE displacement at the same locations. Both should be presented alongside the strain correlation in the final report.

Step 18 — Interpret Differences Physically

When a discrepancy is found, the engineer must interpret it physically, not just numerically. "The gauge reads 15% high" is a numerical statement; "the gauge reads 15% high, the deflection is also 15% high, and the gauge is on the primary load path, which suggests the model stiffness is 15% low — possibly due to an over-flexible joint that the model idealises as rigid" is a physical interpretation. The physical interpretation considers the pattern across gauges, the global checks, the direction of the discrepancy and the known uncertainties, and proposes a cause that is consistent with all the evidence. The cause may be in the test (instrumentation, fixture), in the model (load, boundary, property, mesh), or in the real structure (as-built geometry, material scatter). The engineer should test the proposed cause: if the joint is over-rigid in the model, reducing the joint stiffness should improve both the deflection and the strain correlation. If it does, the interpretation is supported. If it does not, the interpretation is wrong and must be revised. This cycle — observe, interpret, test, revise — is the engineering heart of correlation.

MISTAKE: Reporting a correlation as "gauge X is 15% off" without a physical interpretation. A discrepancy without an interpretation is an observation, not a conclusion. Propose a cause, test it against the full pattern, and revise if it does not explain all the evidence.

Step 19 — Update the Model Carefully

If the correlation identifies a model deficiency, the model may be updated to improve the correlation. But model updating is dangerous: it is easy to change a parameter to make one gauge match, while degrading the correlation at other gauges or at the global level. This is curve-fitting, not correlation. Every proposed model change must satisfy three tests. First, it must be physically supportable — there must be a physical reason for the change, not just a numerical one. Increasing a joint stiffness because the deflection is too large is supportable if the joint was idealised and the real joint is stiffer; increasing it just to make a gauge match is not. Second, it must improve the overall pattern, not just one point — the change should improve or maintain the correlation at all gauges and the global checks, not trade one agreement for another. Third, it must be within the plausible range of the parameter — a modulus change of 50% to make a gauge match is not plausible and indicates an error elsewhere. Model updating that fails these tests is not improving the model — it is hiding the discrepancy.

CONSTRAINT: Every model change must be physically supportable, must improve the overall pattern (not just one gauge), and must be within the plausible range of the parameter. A change that improves one gauge at the cost of the pattern is curve-fitting, not correlation.

Step 20 — Document the Correlation

The final step is documentation. A correlation exercise that is not documented might as well not have been done — the evidence is lost, the reasoning is lost, and the conclusion cannot be reviewed or revisited. The documentation should include: the test condition and article definition (Step 1); the FE model description and the test condition run (Steps 1–3); the gauge instrumentation plan with locations, directions, lengths and surfaces (Steps 4–8); the zero, preload and temperature conditions (Steps 9–10); the load–strain plots and linearity assessment (Step 11); the normalised strain comparison table (Step 12); the measured-versus-predicted scatter plot (Step 13); the reaction and deflection correlation (Step 17); the physical interpretation of discrepancies (Step 18); any model updates and their justification (Step 19); and the residual uncertainty and the envelope of validated use. The documentation should allow another engineer to reproduce the comparison, understand the reasoning and reach the same conclusion. This is the evidence chain that supports the credibility of the model.

• Document the test condition, article and FE model setup
• Document the gauge instrumentation — locations, directions, lengths, surfaces
• Document the load–strain plots and linearity assessment
• Document the normalised strain comparison and scatter plot
• Document the reaction and deflection correlation
• Document the physical interpretation of every significant discrepancy
• Document any model updates and their physical justification
• State the residual uncertainty and the envelope of validated use

The Practical Correlation Checklist

The following checklist condenses the workflow into the questions that should be answered for every gauge in a strain correlation. It is the single-page summary that an engineer should have to hand when performing the correlation, and that a reviewer should ask to see when assessing it. If any answer is "no" or "not sure", the corresponding step has not been adequately addressed, and the correlation for that gauge is incomplete.

• Load — Is the FE load the same as the test load?
• Linearity — Is the test response approximately linear and elastic?
• Location — Does the FE extraction correspond to the actual gauge centre?
• Direction — Has the FE strain been transformed into the gauge-grid direction?
• Surface — Is the correct physical surface being used?
• Gauge Length — Does spatial averaging matter at this location?
• Zero State — Are test and FE strains referenced to the same assembled/preloaded condition?
• Temperature — Could apparent thermal strain contaminate the measurement?
• Fixture — Does the FE boundary condition represent test-fixture compliance adequately?
• Reaction — Do measured/applied loads and FE reactions balance?
• Global Stiffness — Does measured deflection support the FE stiffness?
• Repeatability — Does the gauge follow the same path on repeated loading?
• Pattern — Do all gauges show a coherent spatial correlation pattern?
• Model Update — Can every proposed model change be supported physically?

A Practical Correlation Table

The following table structure is recommended for recording the correlation for each gauge. It forces the engineer to state every piece of information needed for the comparison, and it presents the result in a form that can be reviewed at a glance. Every column should be filled for every gauge; a blank column is a step that has not been completed.

Common Reasons Correlation Fails

When the correlation does not match, the cause is usually one of a small number of recurring problems. The table below summarises the most common causes, the pattern they produce, and the diagnostic that identifies them. This is not an exhaustive list — the article on discrepancies covers this in more detail — but it is the first-look table for the engineer staring at a scatter plot that does not lie on the 45° line.

Bringing It All Together

A linear static FE to strain gauge correlation is not a single calculation — it is a twenty-step workflow that spans test definition, FE extraction, measurement hygiene, pattern interpretation and documentation. The engineer who works through the steps methodically, who applies the checklist to every gauge, and who interprets discrepancies physically rather than numerically, will produce a correlation that is a defensible engineering argument. The engineer who shortcuts the steps, who compares untransformed nodal strain against a gauge reading, or who reports discrepancies without interpretation, will produce a comparison that is not worth the paper it is printed on. The difference is not talent — it is discipline. The discipline is summarised in the two statements that frame this article.

THE MOST USEFUL STRAIN CORRELATION COMPARES THE SAME PHYSICAL QUANTITY, AT THE SAME LOCATION, IN THE SAME DIRECTION, OVER THE SAME LOAD STATE.

Key Takeaways

• The workflow is twenty steps in five phases — test definition, FE extraction, measurement hygiene, pattern reading, contextualisation
• Global correlation (load, reaction, deflection) must be established before local strain correlation is interpreted
• FE strain must be extracted at the gauge centre, transformed to the gauge direction, averaged over the gauge length, and taken from the correct surface
• The measured-versus-predicted scatter plot is the primary diagnostic tool — read the pattern, not just individual points
• Every model update must be physically supportable and must improve the overall pattern — curve-fitting is not correlation
• Document the full correlation — test condition, extraction, comparison, interpretation, updates, residual uncertainty

Key takeaways

• The most useful strain correlation compares the same physical quantity, at the same location, in the same direction, over the same load state — every shortcut from this principle contaminates the comparison.
• Global correlation (load, reaction, deflection) should generally be established before local strain correlation is interpreted — if the global stiffness is wrong, local strain agreement is coincidence.
• FE strain must be extracted, transformed into the gauge direction, averaged over the gauge length, taken from the correct surface, and scaled to the test load before comparison — each step matters.
• A 20-step workflow takes the engineer from defining the test condition through to documenting residual uncertainty — skipping steps is the most common cause of meaningless correlation.
• Every proposed model update must be supported physically — a change that improves one gauge at the cost of the pattern is curve-fitting, not correlation.