Understanding Strain Correlation Discrepancies

Discrepancies Are Diagnostics

When the measured strain does not match the predicted strain, the natural reaction is to think the model is wrong. Sometimes it is. But often the discrepancy is pointing to a problem in the test, the instrumentation, the fixture or the real structure — and the model is right, or right enough. The engineer's job when a discrepancy appears is not to fix the model immediately but to diagnose the cause. The discrepancy is a diagnostic signal: its pattern — which gauges, how much, in which direction, with what global evidence — points to a specific cause. This article is about reading that pattern and tracing the discrepancy to its source. It is the troubleshooting guide for the engineer looking at a correlation that does not match and needing to decide what to do about it.

Why It Matters

Jumping to the conclusion that the model is wrong leads to two kinds of error. First, the engineer may update the model to match a discrepancy that is actually a test error — degrading the model to match bad data. Second, the engineer may miss a real model deficiency because the discrepancy was attributed to a test problem that does not exist. The disciplined approach — separate the sources, read the pattern, test the interpretation — prevents both errors. It ensures that model updates are made for the right reasons and that test problems are caught and corrected rather than absorbed into the model. This discipline is what separates a correlation exercise that improves the model from one that merely hides the discrepancies.

A discrepancy is a diagnostic signal, not a verdict. Diagnose the cause before changing the model — updating a model to match a test error degrades the model and hides the real problem.

The Three Sources of Discrepancy

Every discrepancy originates from one of three sources, and the first diagnostic step is to determine which. Test and instrumentation discrepancies arise from measurement error — gauge bonding, wiring, thermal output, bridge configuration, zero drift, placement error. Model discrepancies arise from the FE model — wrong load, idealised boundary, incorrect material property, coarse mesh, omitted feature, incorrect joint stiffness. Real-structure discrepancies arise from the test article being different from the drawing — as-built thickness, material property scatter, residual stress, preloaded joints, geometry tolerance. These sources produce different patterns, and the pattern is the first clue to the source.

Pattern 1 — Consistent Ratio Across All Gauges

When all gauges show the same ratio of test to prediction — for example, every gauge reads 15% higher than the FE — the discrepancy is systematic, not local. A systematic ratio points to a global cause that scales all strains equally. The most common causes are: the load is different from what was assumed (test load higher, or FE load lower, than intended); the material modulus is wrong (FE modulus too high makes predicted strain too low); or a global stiffness error that affects all load-path strains. The diagnostic is to check the load first — is the test load cell reading correct, is the FE load the same? Then check the modulus — is the FE modulus the correct value for the test material? Then check the deflection — if the deflection is also 15% high, the stiffness is globally low, which is consistent with all strains being high.

• Check the load — is the test load the same as the FE load? Load cell calibration?
• Check the modulus — is the FE modulus the correct value for the test material and temperature?
• Check the deflection — is the global deflection also off by the same ratio?
• If load and modulus are correct and deflection is also high → global stiffness error (joint, boundary)
• A consistent ratio is the easiest discrepancy to diagnose — it is a single global cause

Pattern 2 — Regional or Clustered Discrepancy

When the discrepancy is concentrated in one region — a cluster of gauges in one area are off, while gauges elsewhere match — the cause is local to that region. The pattern points to a local modelling deficiency: a joint that is modelled too stiff or too soft, a load introduction that is idealised, a boundary condition that does not represent the fixture, a feature that is omitted or coarsely meshed. The diagnostic is to identify what is different about the region: what model assumption is being tested by the discrepant gauges? If the discrepant gauges are all near a joint, the joint model is suspect. If they are near the load introduction, the load distribution is suspect. If they are near a stress concentration, the local mesh or feature fidelity is suspect. The regional pattern narrows the cause to a specific modelling assumption.

A clustered discrepancy points to a local modelling deficiency. Identify what the discrepant gauges have in common — a joint, a boundary, a load introduction, a feature — and that is the suspect assumption.

Pattern 3 — Single Gauge Outlier

A single gauge that disagrees with the prediction while all its neighbours agree is usually a gauge or extraction problem, not a model problem. The diagnostic is to check the gauge first: is the sign correct (wiring)? Is the direction recorded correctly? Is the gauge well bonded (repeatability)? Is the zero stable? Is the gauge in a high-gradient region where placement or extraction error is large? Is the FE extraction at the correct location, surface and direction for this gauge? Only after all of these are checked and confirmed should the discrepancy be attributed to the model. A single-gauge outlier in a region where the model is otherwise well correlated is far more likely to be a measurement or extraction issue than a local model deficiency that affects only that one gauge.

• Check the sign — a sign error means wiring or direction recording error, not model error
• Check the direction — is the gauge angle correctly recorded and used in the FE transformation?
• Check the bonding and repeatability — does the gauge repeat on reload?
• Check the extraction — is the FE strain extracted at the right location, surface and direction?
• Check the gradient — is the gauge in a high-gradient region where placement error matters?
• Only after all checks pass should a single-gauge outlier be attributed to the model

Pattern 4 — Sign Errors

A sign error — the gauge reads compression where the model predicts tension, or vice versa — is the most informative discrepancy. It means the gauge and the model disagree on the fundamental direction of the strain, not just the magnitude. A sign error at a single gauge usually means a wiring error (the gauge is wired backwards), a direction recording error (the gauge angle is wrong), 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 investigated first, before any magnitude comparison, because a sign error invalidates the magnitude comparison — a gauge reading −500 με against a prediction of +500 με is not a 1000 με error, it is a direction error that must be resolved before the magnitudes are discussed.

MISTAKE: Treating a sign error as a magnitude error. A gauge reading −500 με against a prediction of +500 με is not a 1000 με discrepancy — it is a direction error. Resolve the sign before comparing magnitudes.

Pattern 5 — Nonlinear or Non-Proportional Response

If the load–strain plot is not a straight line through the origin, the structure is behaving nonlinearly, and the discrepancy between test and linear FE is not a model error — it is a regime error. The linear FE model is not valid for a nonlinear structure. The diagnostic is to identify the nonlinearity: downward curvature suggests yielding (reduce the load to the elastic range, or build a nonlinear material model); upward curvature suggests joint slip or contact closure (reduce the load or model the contact/joint nonlinearity); hysteresis suggests joint slip or friction (investigate the joint behaviour). The action depends on the intended use: if the model is needed only for elastic loads, correlate at a lower load where linearity holds; if the model is needed for the nonlinear regime, a nonlinear FE analysis is required. A linear model correlated against a nonlinear test will always show discrepancy, and updating the linear model to match it is the wrong response.

Pattern 6 — Constant Offset Across All Gauges

A constant offset — all gauges read the same amount higher or lower than the prediction, regardless of magnitude — is different from a consistent ratio. A ratio scales with the strain magnitude; an offset does not. A constant offset points to a datum or preload mismatch: the test zero and the FE zero do not represent the same state. If the test article is preloaded by fixture, gravity or assembly and the FE model does not include the preload, every gauge carries a constant pre-strain that appears as an offset. The diagnostic is to check the load–strain plot: if it is linear but does not pass through the origin, the offset is a datum error. The fix is to reconcile the zero state — either re-zero the test after preload settles, or include the preload in the FE model.

A constant offset across all gauges, with a linear load–strain plot that does not pass through the origin, is a datum or preload mismatch — not a model error. Reconcile the zero state between test and FE.

Using Global Checks to Diagnose

The global checks — reactions and deflection — are powerful diagnostics for strain discrepancies. If the strain is high and the deflection is also high, the global stiffness is low — a single cause explains both. If the strain is high but the deflection is correct, the local load distribution is wrong, not the global stiffness. If the reactions do not balance the applied load, the load introduction or measurement is wrong, and no strain comparison is valid until it is fixed. The global checks narrow the cause by providing independent evidence: a strain discrepancy with supporting global evidence points to a global cause; a strain discrepancy without global evidence points to a local cause. Always read the strain discrepancies alongside the reaction and deflection correlation.

Using global checks to diagnose strain discrepancies:

  Strain high + Deflection high  →  Global stiffness low (modulus, joint, boundary)
  Strain high + Deflection OK    →  Local load distribution wrong (not global stiffness)
  Strain high + Reactions wrong  →  Load introduction or measurement error (fix first)
  Strain high + All global OK    →  Local modelling issue (feature, mesh, extraction)
  Strain OK  + Deflection high   →  Strain happens to match; stiffness still wrong — investigate

  Always read strain correlation alongside reactions and deflection.

The Diagnostic Process

The disciplined response to a discrepancy is a process, not a leap. First, check the signs — resolve any sign errors before anything else. Second, check the global evidence — reactions and deflection. Third, classify the pattern — systematic ratio, regional cluster, single outlier, offset, nonlinearity. Fourth, propose a cause consistent with the pattern and the global evidence. Fifth, test the cause — if it is a load error, check the load; if it is a joint stiffness, vary the joint stiffness and see if the correlation improves; if it is a gauge problem, re-bond or re-wire. Sixth, if the cause is confirmed, take the appropriate action — fix the test, update the model, or measure the as-built article. Seventh, re-correlate and confirm the discrepancy is resolved or explained. This process prevents the two errors of diagnosis: blaming the model for a test problem, and blaming the test for a model problem.

• Check signs first — resolve sign errors before magnitude comparison
• Check global evidence — reactions and deflection
• Classify the pattern — systematic, regional, single, offset, nonlinear
• Propose a cause consistent with the pattern and the global evidence
• Test the cause — vary the suspect parameter or check the suspect measurement
• Take action — fix the test, update the model, or measure the as-built article
• Re-correlate and confirm the discrepancy is resolved or physically explained

When to Update the Model

A model update is justified only when the discrepancy has been traced to a model deficiency, the proposed change is physically supportable, and the change improves the overall correlation — not just one gauge. If the discrepancy is traced to a test problem, fix the test. If it is traced to a real-structure difference, update the model to represent the as-built article. If it is traced to a model assumption, update the assumption — but only with physical justification and only if the change improves the pattern. A model update that makes one gauge match while degrading others is curve-fitting, not correlation. The model update is the last step in the diagnostic process, not the first — it is taken only after the cause is confirmed and the change is justified.

Update the model only after the discrepancy is traced to a model deficiency, the change is physically supportable, and the change improves the overall pattern. Updating to match one gauge while degrading others is curve-fitting.

Key Takeaways

• A discrepancy is a diagnostic — its pattern points to a specific cause in the test, the model or the real structure
• Separate the three sources before concluding the model is wrong — test, model, real structure
• A consistent ratio is a systematic error — check load, modulus, global stiffness
• A clustered discrepancy is a local modelling deficiency — identify the common assumption
• A single outlier is usually a gauge or extraction problem — check the gauge before the model
• Use global checks (reactions, deflection) to distinguish global from local causes
• Update the model only after the cause is confirmed and the change is physically supportable and improves the pattern

Key takeaways

• A discrepancy is a diagnostic, not a failure — its pattern points to a specific cause in the test, the model or the real structure.
• Separate the three sources: test/instrumentation, model, and real structure — before concluding the model is wrong.
• A consistent ratio across all gauges is a systematic error — look to load, modulus or global stiffness.
• A clustered or regional discrepancy points to a local modelling deficiency — joint, boundary, load introduction or feature fidelity.
• A single outlier is usually a gauge or local extraction problem — check the gauge before blaming the model.