Designing a Test for Analysis Correlation

What It Means to Design a Test for Correlation

A test designed for analysis correlation is not the same as a test designed purely for qualification or for load envelope determination. A qualification test asks whether the structure survives the required loads — it is a pass/fail exercise. A correlation test asks whether the model predicts what the structure actually does — it is a measurement exercise. The difference shapes every decision: where to put gauges, how to introduce load, how to support the article, what to measure and how many load levels to run. A correlation test is designed backwards from the model: identify the assumptions the model depends on, then instrument and load the structure to test exactly those assumptions.

Why It Matters

If the test is not designed to challenge the model, the correlation exercise may produce data that is uninformative or, worse, misleadingly reassuring. Gauges placed only on flat accessible surfaces will measure low, uninteresting strains and will tell you nothing about whether the model captures the critical load path. A load introduced through a fitting that the model idealises differently from the test will produce a correlation that measures the fixture difference, not the structure. Designing the test for correlation is the difference between generating evidence and generating numbers.

A correlation test that does not challenge the model's important assumptions produces data that cannot distinguish a good model from a bad one. Instrument the assumptions, not the convenience.

Start from the Model

The design of a correlation test begins with the model, not with the hardware. The engineer reviews the finite element results and identifies the regions and assumptions that matter: where the primary load path carries the load, where the model assumes a particular joint stiffness, where a stress concentration has been idealised, where the boundary condition is uncertain, where the model predicts high strain that drives a design margin. These are the regions the test must interrogate. The test plan is then built around placing instrumentation at these locations and applying loads that exercise these load paths.

• Identify the primary load path from the FE results — where is the load actually carried?
• Identify the assumptions the model depends on — joint stiffness, boundary fixity, load introduction
• Identify the regions that drive design margins — where the model predicts critical strain or stress
• Identify the uncertain regions — where the model idealises a detail whose real behaviour is unknown
• Instrument all of these — the test exists to challenge them

Gauge Placement Strategy

Gauge placement is the single most important decision in a correlation test, and it is where most tests go wrong by placing gauges where they are easy rather than where they are informative. A gauge placed on a flat skin panel away from any load path measures background strain that almost any reasonable model will predict — it confirms nothing. A gauge placed at the root of a fitting, on a shear web, at a stress concentration or at a joint measures the strain that the model must get right if it is to be trusted. The guiding principle is: place gauges where the model is most likely to be wrong in ways that matter.

Load Introduction and Boundary Conditions

The way load is introduced into the test article and the way the article is supported are part of the correlation, not just test infrastructure. If the model assumes a pinned support and the test uses a bolted fixture with finite stiffness, the correlation will be contaminated by the fixture compliance — every strain will reflect the real boundary, not the idealised one. The engineer has two choices: either design the test fixture to match the model boundary condition, or model the test fixture in the FE analysis so that the comparison is like-for-like. The second is usually more honest, because the real fixture is never ideal, and the model should represent what was actually tested when correlating.

MISTAKE: Correlating an FE model with idealised boundary conditions against a test article held in a compliant fixture. The discrepancy measures the fixture, not the structure. Model the fixture, or design the fixture to match the model.

Instrumentation Coverage Across the Hierarchy

A well-designed correlation test instruments the full correlation hierarchy, not just strain. Load cells measure the applied load — the Level 1 check. Reaction cells or support load cells measure the reacted load — the Level 2 check. Displacement transducers (LVDTs, string pots, photogrammetry targets) measure global deflection — the Level 3 check. Strain gauges measure regional and local strain — Levels 4 and 5. When a strain discrepancy appears, the engineer can walk up the hierarchy: was the load right? Was the reaction right? Was the deflection right? If all three are right and the strain is wrong, the problem is local. If the deflection is also wrong, the problem is global. Without the full hierarchy, this diagnosis is impossible.

Instrumentation mapped to correlation hierarchy:

  Load cell          →  Level 1: Load
  Reaction cell      →  Level 2: Reaction
  LVDT / photogrammetry →  Level 3: Global Stiffness
  Regional gauges    →  Level 4: Regional Strain
  Detail gauges      →  Level 5: Local Detail

Each level enables diagnosis of discrepancies at the level below.

Multi-Level Loading

A correlation test should not run a single load level and stop. It should run a ladder of load levels — typically 20%, 40%, 60%, 80% and 100% of the maximum test load — and record strain at each. This serves two purposes. First, it checks linearity: in a linear elastic structure, strain should scale proportionally with load, and the FE model assumes linearity. If the test strain bends away from the straight line, the structure is behaving nonlinearly — yielding, slipping in joints, contact opening — and the linear FE model is no longer valid for comparison. Second, it provides a load-normalised strain (microstrain per kN) that is independent of the exact load magnitude, allowing comparison even if the test and FE loads differ slightly.

• Run a load ladder — at least three to five load levels up to the maximum
• Check linearity — strain should scale proportionally with load in a linear elastic structure
• Detect nonlinear behaviour — joint slip, yielding, contact opening all show as non-proportional response
• Compute load-normalised strain (με/kN) — allows comparison independent of exact load magnitude
• Check repeatability — unload and reload to confirm the gauge follows the same path

Gauge Quantity Versus Gauge Quality

There is a persistent temptation to instrument heavily — hundreds of gauges — on the assumption that more data is always better. It is not. A hundred gauges smeared across accessible surfaces produce a large dataset that is mostly uninformative and that overwhelms the correlation analysis. Ten to twenty gauges placed precisely at the locations that test the model's important assumptions produce a small, interpretable dataset that directly interrogates the model's credibility. The engineer should resist the pressure to instrument for quantity and should instrument for information. Every gauge should have a stated purpose: what model assumption does this gauge challenge?

CONSIDERATION: Every strain gauge in a correlation test should have a written rationale — which model assumption does it challenge? A gauge without a rationale is data without information.

Documentation of the Test Configuration

A correlation test is only as useful as its documentation. The FE engineer who will perform the correlation needs to know exactly where each gauge was placed (coordinates and surface), what direction each gauge was oriented (angle relative to a defined reference), what gauge length and type were used, how the load was introduced, what the fixture stiffness was, and what the load sequence was. If this information is not recorded precisely, the FE extraction cannot match the test, and the correlation is compromised from the start. A test that produces good data but poor documentation is a test that cannot be properly correlated.

Pre-Test Analysis: The Correlation Predictions

A powerful but frequently omitted step in test design is the pre-test analysis prediction. Before the test is run, the FE model is used to predict the expected strain at every gauge location, at every load level. These predictions are documented and provided to the test team. During the test, the live strain readings are compared against the predictions in real time. This serves three purposes. First, it catches gross errors immediately — a gauge reading the wrong sign or a strain an order of magnitude away from the prediction flags a problem while the article is still on the rig, not weeks later during post-test analysis. Second, it identifies gauges that are not producing useful data — a gauge that reads near zero where a significant strain was predicted may be miswired, debonded or on the wrong surface, and this can be investigated during the test. Third, it builds the engineer's confidence in the model before the formal correlation: if the live readings broadly follow the predictions, the model is likely sound, and the post-test correlation will confirm the details. The pre-test prediction is a discipline that catches problems early and makes the test more productive.

VERIFICATION: Produce a pre-test strain prediction for every gauge at every load level. Compare live readings against predictions during the test. A gross mismatch during the test is an opportunity to investigate while the article is still on the rig.

Key Takeaways

• Design the test backwards from the model — instrument the assumptions, not the convenience
• Place gauges where the model is most likely to be wrong in ways that matter — load paths, joints, concentrations, boundaries
• Instrument the full hierarchy — load, reaction, deflection, regional strain, local detail
• Run a load ladder — linearity and load-normalised strain are powerful correlation tools
• Document the test configuration precisely — a test that cannot be matched cannot be correlated

Key takeaways

• A correlation test should be designed to challenge the model — instrumentation placed where the model makes its most important and least certain assumptions.
• Gauge quantity does not equal gauge quality — ten well-placed gauges that interrogate the load path are worth a hundred gauges smeared across easy-to-reach surfaces.
• The test boundary condition and load introduction must be represented in the model, or the correlation will measure the difference between two different problems.
• Instrumentation should cover the full correlation hierarchy — reactions, deflection, regional strain and local detail — so that discrepancies can be diagnosed by level.