A structural model, by default, assumes a part starts its life at zero stress. Geometry is meshed, a material card is assigned, a load is applied, and the solver reports how that load is carried. It is a clean, internally consistent picture. It is also, for a large share of real parts, false from the first time step.
Nearly every part with a manufacturing history carries stress that was locked in before it ever saw a service load. Welding, injection moulding, casting, forging, heat treatment, and machining all leave stress behind. That stress does not show up on a drawing, does not appear in a bill of materials, and in most workflows, does not appear in the structural model either. It simply sits inside the part, waiting to be added to whatever load comes next.
Where the stress actually comes from
Residual stress is what remains after a process that deformed a material non-uniformly, once the external cause of that deformation is gone. The mechanisms differ by process, but the underlying story is the same: different regions of a part want to change shape by different amounts, and the material's own continuity forces a compromise.
In welding, the heat-affected zone expands during heating and contracts during cooling far more than the surrounding cold material. The weld region wants to shrink; the surrounding metal will not let it, so the weld ends up in residual tension and the adjacent material in compression to balance it. In quenching and heat treatment, the surface cools and contracts before the core does, again leaving a signature stress profile through the thickness. In injection moulding, differential cooling between the skin and the core, combined with pressure-driven packing during solidification, freezes in a stress state that mirrors the flow and cooling history discussed elsewhere on this site. In machining, removing material redistributes whatever stress state already existed in the stock, which is why thin-walled parts sometimes bow or twist only after the last cut, with no load applied at all.
Why it superimposes, and why that matters
Residual stress does not sit passively in a part. Once a service load is applied, the total stress at any point is approximately the sum of the residual stress already present and the stress produced by that load:
A location that a model predicts to be safely below yield under load alone can, in reality, already be carrying a meaningful fraction of that margin before the load is even applied. The magnitudes involved are not marginal. Residual stresses in the heat-affected zone of an as-welded steel joint commonly approach 70 to 100 percent of the base material's yield strength. Shot peening, applied deliberately, typically induces compressive residual stress on the order of 300 to 600 MPa near the surface in steels, which is precisely the point: the same order of magnitude that arises unintentionally in welding can be engineered intentionally to work in the opposite direction.
This cuts both ways. Tensile residual stress at a surface is a liability: it adds directly to service-load tension at that surface, and surfaces are also where fatigue cracks and stress corrosion cracking tend to initiate. Compressive residual stress at a surface is often a genuine asset, which is precisely why processes like shot peening and cold rolling are used deliberately, to induce compressive stress at fatigue-critical surfaces and measurably extend service life. The same underlying physics that quietly works against an unplanned weld can be engineered to work in a part's favour, when it is treated as a variable rather than an afterthought.
Why it stays invisible
Residual stress does not announce itself on inspection. A part can pass dimensional checks, visual inspection, and even non-destructive testing for cracks, while still carrying a stress state that will determine whether it survives its actual duty cycle. Measuring it directly requires specialised techniques, and each comes with a depth limitation that shapes what it can actually tell an engineer. Laboratory X-ray diffraction probes only the outermost 5 to 20 micrometers of a surface, effective for near-surface gradients but blind to what lies beneath. The hole-drilling method, which measures strain relief as a shallow hole is incrementally machined, extends useful measurement to roughly 1 to 2 mm of depth. Neutron diffraction can probe several centimetres into the bulk of a part, but requires access to a neutron source and is neither fast nor routine. None of these sit anywhere near a standard incoming inspection process.
It is also, structurally, invisible to the design process for a more basic reason: nothing in a typical CAD or CAE workflow asks where it came from. Geometry comes from CAD. Material properties come from a datasheet. Loads come from a load case document. Residual stress comes from the part's own manufacturing history, which lives in a process engineer's head, a welding procedure specification, or a moulding parameter sheet, not in any file the structural analyst is likely to open.
What it costs to ignore
The failure modes this produces share a pattern with the other gaps discussed on this site: the simulation converges, reports a clean safety factor, and looks entirely defensible, right up until the part fails at a location or a load level the model gave no reason to suspect. A weld that passes a structural check based on parent-material properties can still crack at the heat-affected zone under a fraction of the predicted allowable stress, because the model never knew that zone started in tension. A precision-machined part can distort out of tolerance on the inspection table, with no load applied at all, because material removal released a residual stress state nobody accounted for during process planning.
In both cases, the equations were not wrong. The model simply started from an assumption, zero initial stress, that the physical part never satisfied.
What actually closes the gap
Closing this gap does not require modelling every part to the same level of rigor. It requires knowing which parts and which locations warrant it, and treating residual stress as a design variable rather than an unknown to be absorbed by margin.
For welded structures, that means either importing a thermal-mechanical weld simulation's stress state as an initial condition for the structural model, or applying a known stress-relief heat treatment and validating it rather than assuming it. For injection-moulded parts, it means treating the same mould-fill and cooling simulation already used for fibre orientation as a source of residual stress data, since the two are generated by the same solidification physics. For machined parts cut from stock with an unknown stress history, it means recognising that dimensional stability after material removal is itself a signal, and building process controls, such as stress-relief operations before final machining, rather than discovering the problem at final inspection.
None of this is exotic. It is the same discipline already applied to mesh convergence and material card accuracy: treating an input to the model as something that needs to be interrogated, not assumed. A structural model that starts from zero stress is answering a real question. It is just not always the question the actual part is going to face.