Imagine a stepped machine shaft with a finite shoulder fillet. An FE model has solved, and one region around the fillet stands out immediately.
The familiar shortcut is tempting:
Find the maximum stress. Compare it with an allowable value. Make the conclusion.
Before doing that, connect the result back to four things: the physical loading, the stress quantity being displayed, the engineering question and the credibility of the local numerical result.
A stress contour is an output. It still needs engineering interpretation.
ASSUMPTION FOR THIS ARTICLE
We examine one static structural load case containing bending and torque. We are not calculating the cyclic stress history created by shaft rotation or performing a fatigue-life assessment. The shaft is treated as a homogeneous, isotropic material with small-deformation, linear-elastic behaviour and a finite shoulder-fillet radius. Plastic redistribution, contact and fracture assessment are outside this baseline.
Before the Contour, What Stress Should We Expect?
Assume the upstream reactions and section resultants have already been established. In the shoulder region, the shaft carries bending and torque.
Start there; not with the results menu.
Bending creates longitudinal normal stress in the shaft. One side can be more tensile while the opposite side can be more compressive, depending on the bending sense. Torque creates torsional shear stress.
First-order mechanics provides the nominal reference: bending contributes longitudinal normal stress, torque contributes shear, and the finite fillet creates a local 3D stress state.
So before opening the FE results, we already know the local material is not experiencing one mysterious scalar “stress.” Different stress contributions have identifiable physical origins.
Once we open a von Mises contour, information from that local stress state has been compressed into one scalar quantity. That scalar may be useful, but it should not erase our understanding of where the state came from.
Away from the shoulder, first-order bending and torsion mechanics gives us a nominal baseline. Near the fillet, the geometry disturbs that simpler stress field.
Here, the first-order mechanics gives us the baseline; the FE model resolves how the local fillet modifies that response.
One Stress State, Different Ways of Looking at It
At a point in a three-dimensional solid, stress is not inherently one number. The local stress state can be described through normal and shear components relative to a chosen coordinate system.
That coordinate system matters. A directional stress component has meaning only when we know what the reported direction represents physically.
For our shaft, a normal-stress component aligned with the shaft axis can help us understand how the bending-related normal stress changes around the shoulder.
Principal stresses give another view of the same state. In three dimensions there are three principal stresses, and in the corresponding principal directions the shear components vanish.
The maximum principal stress is the largest algebraic principal normal stress; the minimum principal stress is the smallest. Depending on the loading, the principal values may include both tension and compression, or they may all have the same sign.
Von Mises stress does something different. It reduces information from the multiaxial stress state to a single non-negative equivalent value. For many homogeneous isotropic ductile metals, it is commonly used as an equivalent stress measure in an initial-yield assessment.
That does not make von Mises “the stress in the component,” and principal stress is not a competing “better stress.”
Directional components preserve directional information. Principal stresses reveal the extreme normal stresses and their principal directions. Von Mises provides an equivalent scalar for a particular class of yielding assessment.
The solver may offer all of them. The engineering question determines which information we need.
When I review a stress result, this is where I stop asking which stress is “best” and ask what I am actually trying to assess.
At one material point, directional components, principal stresses and von Mises are different representations derived from the same local stress state.
Which Result Matters to the Engineering Question?
Suppose our question is deliberately narrow:
For this isotropic ductile-metal model, is the shoulder region approaching initial yielding?
Von Mises equivalent stress is relevant to that question.
But if we instead want to understand how strongly tensile the local state becomes, or how the bending-related axial normal stress changes through the fillet, a principal stress or directional component may reveal information that von Mises intentionally does not preserve.
So the useful question is not:
Should I always use von Mises or principal stress?
It is:
What behaviour am I trying to assess, and what information from the stress state does that assessment require?
That also tells us where this article stops.
A static stress contour is not a fatigue-life assessment; fatigue requires the cyclic stress or strain history and an appropriate fatigue method. Likewise, if an existing crack is the concern, the problem may require fracture-mechanics quantities rather than an ordinary pointwise maximum stress.
One scalar stress plot cannot answer every structural-integrity question.
Why Is the Highest Value at the Fillet?
Away from the diameter change, the nominal bending-and-torsion picture gives us a useful reference.
At the finite-radius fillet, geometry changes how the internal loading is transmitted through the material. The stress field becomes more localised, and the peak can rise above the nominal section stress.
That is a stress concentration.
Its existence does not mean the FE model is wrong. A real hole, finite-radius notch or finite fillet can create a physically meaningful local elastic stress concentration.
If that region matters to the engineering decision, resolving it may be exactly why we need a more detailed model.
But a real local concentration does not mean the largest number in a smooth contour should automatically become the engineering conclusion. We still need to ask how that value was obtained numerically.
Does the Local Maximum Deserve Trust?
What is the contour actually showing?
A smooth FE contour can look as though the solver calculated one continuous stress value at every coloured point on the screen. The numerical process can be more involved.
For many common solid-element formulations, stresses are evaluated at integration points inside the elements. Post-processing may then extrapolate those results toward nodes and average contributions from neighbouring elements to create a smooth contour.
So the displayed value at a visible location may not be identical to an underlying integration-point value.
Averaging is not automatically wrong; it can make a stress field easier to visualise. But visual smoothness is not proof that the local field has been resolved adequately.
Likewise, an unaveraged contour can appear less smooth because neighbouring element contributions remain separate. Large differences between them can be a reason to inspect the local stress gradient and mesh resolution more closely.
They are a diagnostic signal, not a universal pass/fail criterion. There is no single percentage difference that proves a mesh is acceptable for every structure and stress question.
A smoother contour is not evidence that the local stress has converged.
What happens when the mesh is refined?
Now refine the region around the finite shoulder fillet.
For this finite-radius, linear-elastic baseline, with an appropriate numerical representation, the local stress quantity we are tracking should move toward a stable value as the field is resolved more adequately.
The important phrase is: the result we are tracking.
Mesh convergence is always convergence of a defined quantity. We should not simply say, “The mesh converged.” We need to know what was checked.
Now consider a different model. Replace the real finite fillet radius with a perfectly sharp re-entrant shoulder.
That is a different idealisation; not the same physical fillet with a finer mesh.
In a linear-elastic model, some sharp re-entrant corners and other idealisations can create a stress singularity. Point loads, abruptly terminated distributed loads and certain sharp-edge contact conditions can create similar problems.
At a mathematical stress singularity, continued local mesh refinement does not recover a finite point stress. The reported peak can continue increasing as the elements become smaller.
The correct conclusion is not immediately, “The whole FE model is wrong,” and it is not, “The real component has infinite stress.”
What we can say is more precise:
The point stress at that singular location cannot be treated as a finite mesh-converged result in the present idealisation.
Another result in the same model may still be useful. Stress sufficiently away from the singular location, deformation, reaction, a section quantity or another response may still be adequate for the engineering question.
This is why three ideas should remain separate.
Solver convergence tells us, where applicable, whether the numerical solution procedure has satisfied its convergence requirements for the model.
Mesh convergence asks whether the defined result we care about settles toward a sufficiently stable value as the discretisation is refined.
Physical adequacy asks whether the model represents the behaviour required to answer the engineering question.
Passing one does not automatically prove the others.
From Contour Value to Reviewable Conclusion
When a stress result looks important, trace it rather than simply report its maximum.
Use the Stress Result Trace below to move from a contour value to a reviewable engineering conclusion.
For our shaft, that means connecting bending and torque to the local stress state, recognising the finite fillet as a concentration, selecting von Mises for the present initial-yield question, and checking the result before using it.
The model may still need refinement. The loading or material assumptions may need better evidence. The conclusion may change after technical review.
That is part of engineering analysis.
A junior analyst is not expected to resolve every structural-integrity question independently. What helps a senior reviewer is making the reasoning visible: what created the stress, why the result was selected, what the local peak represents, how it was challenged and what limitations remain.
Within Struxinova, guided engineering situations practise this connection between physical loading, modelling, stress interpretation and checking. Guided practice develops the reasoning discipline; it does not, by itself, prove independent professional competence.
A high stress contour can tell us where to look more closely.
But before the maximum becomes an engineering conclusion, we should be able to explain:
What stress is this? What created it? Why does it matter? And why does this local value deserve to be used?
PRACTITIONER QUESTION
For engineers who review structural or CAE work: when a junior analyst reports a maximum stress, what is the first thing you want them to explain before you trust the conclusion?
About the author
Avinash S is the CEO and Partner at InnoventEdutec, leading the Struxinova and Mathinova learning initiatives. He has more than 16 years of experience spanning engineering skill development, application engineering, technical-content development, project leadership and learning-product strategy. His work includes university- and industry-aligned learning programmes, academic and OEM engineering projects, engineering simulation programmes and technical training. Through Struxinova, he focuses on scientific thinking, engineering judgement, applied structural-mechanics fundamentals and physics-based simulation validation.