Measurements, idealisations and checking structural predictions

Predict the same quantity that is observed

Specify the point, direction, reference and loading condition of a prediction. A modelled midpoint displacement relative to fixed supports cannot be compared directly with crosshead motion that includes fixture movement. A strain gauge samples a local direction over a small region; it does not measure the entire stress field or automatically provide stress without an appropriate constitutive relation.

Keep the material law, geometry, boundary conditions and load history in the model record. In the small linear beam approximation, deflection can depend strongly on span and section dimensions. An apparently minor dimensional error may produce a noticeable response error. Measurement uncertainty belongs to the observation process; uncertain model inputs and model-form limitations also matter.

Plot disagreement before assigning its cause

A residual is an observed value minus its corresponding prediction. Plotting residuals against force can reveal a trend that one average hides. A roughly constant residual suggests different questions from a residual that grows with load. Neither pattern proves a cause: zero offset, fixture compliance, load error and a poor material model can sometimes create similar signatures.

The figure uses exact teaching predictions and invented observations with supplied expanded uncertainty U=0.01 mm, k=2. These intervals are not guaranteed bounds on the true displacement. Model-input uncertainty is not supplied or plotted. Therefore the figure can expose disagreement with the stated line but cannot provide a complete probabilistic validation verdict.

Use new evidence to test the revised explanation

Parameter fitting selects model values to match a record, for example adjusting an effective compliance. It is useful, but agreement with the same fitted points is expected and provides limited independent evidence. Preserve the original prediction, explain the fitted quantities and test the revised model against separate conditions within the intended scope.

A numerical model adds further checks: units, boundary conditions, equilibrium and whether refining its discretisation changes the result. More decimal places do not repair a wrong support condition. Agreement at a few points also cannot establish failure load or an unseen local peak. Choose evidence that challenges the important assumption rather than only reproducing the quantity used in fitting.

Interpret a load-dependent residual

The supplied prediction is δ=0.020F mm for F in N. At 5, 10 and 15 N, it gives 0.10, 0.20 and 0.30 mm. Invented observations are 0.12, 0.23 and 0.35 mm. Residuals are 0.02, 0.03 and 0.05 mm, all positive and increasing overall. The observations lie above the exact teaching line even when their supplied±0.01 mm intervals are drawn.

At 10 N alone, the apparent force/displacement stiffness is 10/0.23≈43.5 N/mm, whereas the model gives 50 N/mm. That numerical difference is not automatically a material modulus difference. A flexible support would also increase movement. A useful next comparison could distinguish motion at the support from motion within the beam, using a defined reference and independent measurement evidence.

Try the reasoning

1. What is the residual at 15 N?

   Show the answer and reasoning

   0.35−0.30=0.05 mm, using observed minus predicted.

2. Why is fitting compliance to these points not independent validation?

   Show the answer and reasoning

   The same points helped choose the parameter. New observations are needed to challenge the revised prediction.

3. Can a local strain gauge establish every peak stress in a component?

   Show the answer and reasoning

   No. It measures a local directional response; geometry, material interpretation and unmeasured locations remain relevant.

What to carry forward

Preserve disagreement as evidence, identify competing explanations, and test the revised model beyond its fitted record.