Reading the Rainbow: Nine Failure Modes of a Stress Plot

In a design review, the contour plot is rarely the evidence. It is the argument. The numbers behind it sit in a results file that nobody in the room will open, while the image is projected four feet wide and consumed in three seconds. That asymmetry is not a failure of reviewers; it is how attention works. It is also where margin quietly disappears.

Post-processors are built to make results legible, and legibility is not fidelity. Every default in the visualization pipeline was set by someone, at some point, for some other class of model. What follows are nine ways those defaults overstate margin — and what to do about each before the plot leaves your desk.

1. The colormap performs arithmetic on your eyes

Rainbow (jet) maps spend large swaths of hue on narrow value ranges and compress other ranges into indistinguishable bands. The result is a picture that appears to show gradient where the field is nearly flat, and appears flat where the field is changing. Perceptually uniform maps such as viridis, cividis, or batlow encode magnitude monotonically and survive grayscale printing and most forms of color vision deficiency. If your organization still standardizes on jet, the standard is costing you accuracy.

2. The legend decides what is visible

Auto-scaling to the full range is honest but often unreadable; scaling to the 99th percentile is readable and dishonest. When the legend is clipped, the true peak migrates off the top of the bar and becomes invisible — the region that carries the margin question now renders as the same saturated color as everything near it. State the legend bounds explicitly, in the units of the model, and annotate any value that falls outside them.

3. Deformation scale is magnification, not measurement

The default exaggeration factor in most post-processors is chosen because a true-scale plot looks static. That is a rendering convenience. Reviewers reading a 50x scaled shape will estimate curvature and bending angle from the picture, and those estimates will be wrong by the exaggeration factor. Put the scale factor in the caption, every time, and include at least one true-scale view in the deck.

4. Averaged nodal contours hide the element that governs

Averaging nodal values across element boundaries produces smooth, plausible fields. It also erases discontinuities at material interfaces, contact surfaces, and mesh transitions — precisely the locations where the design question usually lives. Plot unaveraged element values alongside the averaged field, and treat any visible seam in the unaveraged plot as a question the model has asked you.

5. One hot node is not a stress

A single element lit up in red is the most screenshotted object in engineering, and the least meaningful. Point loads, sharp re-entrant corners, and single-element constraints produce singular fields whose peak value diverges as the mesh is refined. Before that element appears on a slide, run one refinement and plot the peak against element size. If the peak climbs without bound, you are looking at a discretization artifact, not a physical result, and it should be handled with submodeling or a documented exclusion — not quietly ignored.

6. The exported frame is not the envelope

In a transient or nonlinear run, the single time step that looks worst is not necessarily the one that governs. An envelope plot — the maximum value reached at each location over the full event — answers a different and often more useful question than any single frame. The same discipline applies to unconverged snapshots: a residual that never came down produces a picture that is stable in appearance and meaningless in fact.

7. Units, orientation, and the crop that removes doubt

A legend without units is an opinion. Mixed US customary and SI assemblies are common in supplier-supplied models, and a peak reported as 34,000 has different consequences in psi and in MPa. Equally common: an image cropped tight around the region of interest, stripped of the axis triad and the section cut, so that no reader can tell which face they are looking at. Frame wide enough to orient the viewer, then zoom.

8. The probe location goes unstated

When a plot is captioned with a maximum value, the caption should also state where that value occurred and which quantity it represents — von Mises, maximum principal, maximum shear. Different invariants rank locations differently in ductile and brittle regimes, and a review that debates the number without agreeing on the quantity is a review that will need a second meeting.

9. The picture arrives without its case

A contour plot separated from its load case, mesh density, material card revision, and solver version is a picture of a model that cannot be reproduced. The plot becomes an assertion that the reviewer must either accept or dismiss. Attach the case identifier, and the discussion moves from whether the image is convincing to whether the model is appropriate.

The caption standard that survives review

A plot that will be defended needs six elements inside the frame or the caption: the quantity and invariant plotted, the units, the legend bounds with the true extrema, the deformation scale factor, the load case and mesh identifier, and the probe value with its location. Six elements, thirty seconds of work per figure, and the conversation changes character entirely.

The larger point is organizational. Post-processor defaults are a form of inherited policy. Somebody should own them the way somebody owns the material library — reviewed annually, documented, and adjusted when the class of model changes. Until then, every engineer who exports an image is making a visualization decision on behalf of the whole review, whether or not they realize it.

A contour plot cannot overstate margin on its own. It can only do so through defaults that nobody examined. Examine them.