What is FEA and When Do
You Actually Need It?

Finite element analysis occupies an unusual position in engineering practice. It is simultaneously over-used — applied to problems that a five-minute hand calculation could solve more reliably — and under-used on genuinely complex problems where engineers default to conservative simplifications rather than understanding the actual behaviour of their structure. It is also widely misunderstood, both by engineers who run it and by clients and managers who review the output.

A colourful stress plot looks authoritative. It is not automatically correct. This article explains what FEA is, how it works, what types of analysis are available, when it is genuinely the right tool, when it is not, and how to avoid drawing wrong conclusions from results.

What FEA Actually Is

The finite element method is a numerical technique for solving differential equations over complex geometries — most commonly the equations governing stress, strain, heat transfer, fluid flow, or electromagnetic fields in engineering structures. The term "finite element analysis" in mechanical engineering almost exclusively refers to structural stress analysis, though the same method underlies thermal, acoustic and electromagnetic simulations.

The fundamental idea is straightforward: divide a complex geometry into a large number of small, simple sub-regions called elements. Within each element, the behaviour (stress, displacement, temperature) is approximated by a simple mathematical function. Connecting these elements at their shared nodes and enforcing equilibrium produces a very large system of simultaneous equations. The solver — the computer — then solves this system to find the displacement at every node, from which stress and strain throughout the model are derived.

The accuracy of the result depends on the size and type of elements (the mesh), the material model, the accuracy of the applied loads and boundary conditions, and the type of analysis performed. None of these inputs are automatically correct — they all require engineering judgement to define appropriately. A computer cannot know whether your loads are representative, whether your boundary conditions are physically realistic, or whether your mesh is fine enough in regions of high stress gradient. That is the analyst's responsibility.

Types of FEA Analysis

Not all FEA is the same. The type of analysis appropriate for a given problem depends on what is being assessed — and using the wrong analysis type will give results that are either unconservative or so conservative as to be unhelpful.

Linear Static Analysis

The most common type. Assumes the structure is linearly elastic (stress proportional to strain, material behaviour governed by Hooke's law), that displacements are small (geometry does not change significantly under load), and that the load is static (no dynamic effects). The result is a stress and displacement field for a single load case.

Linear static analysis is appropriate for most general structural design checks — beam and frame structures, brackets, machinery bases, pressure vessel nozzles under simple loading. It is fast, well-understood and directly comparable to hand calculations based on elastic theory.

Non-Linear Analysis

Non-linearity comes in three forms, and distinguishing between them matters:

• Material non-linearity: The material yields — stress is no longer proportional to strain. Required when the design intent is to allow local plasticity (plastic design of structures, elastic-plastic assessment of pressure vessels under ASME VIII Div. 2), or when assessing ultimate load capacity rather than elastic stress limits.
• Geometric non-linearity: Displacements are large enough that the deformed geometry significantly affects the load path. Required for thin shells, flexible structures, post-buckling behaviour, and some sealing problems where contact geometry changes under load.
• Contact non-linearity: Two parts interact through a contact interface that opens, closes or slides. Required for bolted joints, interference fits, sealing faces, and any assembly where load transfer through contact is load-dependent.

Non-linear analyses are significantly more computationally expensive and more sensitive to solver settings and mesh quality than linear analyses. They require more careful interpretation and more experience to run correctly.

Modal and Frequency Response Analysis

Modal analysis finds the natural frequencies and mode shapes of a structure — the frequencies at which it will resonate if excited. Frequency response analysis then predicts the amplitude of vibration under a sinusoidal exciting force across a range of frequencies. These analyses are required when vibration is a concern — rotating machinery, pipework subject to flow-induced vibration, structures near vibratory sources, or equipment subject to dynamic loads during transport.

Transient Dynamic Analysis

Predicts the structural response to a time-varying load — an impact, a pressure surge, an explosion, a seismic event. More computationally intensive than frequency response and requires careful definition of the time history of the applied load, which itself is often uncertain.

Thermal and Thermo-Mechanical Analysis

Thermal analysis predicts temperature distribution through a structure for a given heat input and boundary conditions. Thermo-mechanical analysis uses those temperatures as input to a structural analysis, producing thermal stresses. Required for heat exchangers, pressure vessels with significant thermal gradients, fired equipment, and any structure where differential thermal expansion generates meaningful stress.

Fatigue Analysis

Predicts the service life of a structure under cyclic loading. Either uses the linear elastic stress results from static FEA combined with S-N (stress vs. cycles) data for the material, or uses non-linear elastic-plastic analysis to calculate local strain ranges for strain-life (ε-N) fatigue assessment. Required for pressure vessels in cyclic service (ASME VIII Div. 2, EN 13445-3 Annex B), rotating components, and any structure subject to meaningful cyclic load variation.

When Hand Calculations Are Sufficient — and Preferable

FEA is frequently used where it adds cost and complexity but no engineering value over a hand calculation. The cases where a hand calculation is not just sufficient but actually preferable:

Simple Geometries with Known Stress Distributions

Beams in bending, axially loaded columns, circular pressure vessels under internal pressure, shafts in torsion — all have closed-form analytical solutions that are exact within the assumptions of the theory. A Eurocode or ASME code check for a standard section gives a more defensible answer than an FEA model of the same geometry, because the code check is directly tied to a validated design framework and the basis of the result is transparent.

Early-Stage Concept Design

In the early stages of design, geometry changes frequently. Building and rebuilding FEA models to track an evolving concept is an inefficient use of analysis resource. Hand calculations — even rough order of magnitude estimates — are faster, more flexible, and develop engineering intuition about the structure's behaviour in a way that FEA does not.

Checking FEA Results

Every FEA result should be checked against a hand calculation, even a simplified one. If the hand calculation gives a bending stress of 80 MPa and the FEA gives 350 MPa in the same region, one of them is wrong and the hand calculation is often right. The ability to sanity-check FEA results with hand methods is fundamental to competent FEA practice.

When the Uncertainty in Loads Exceeds the Uncertainty in the Calculation Method

If the operating load is known to ±30%, the difference between a hand calculation and an FEA result is irrelevant — both are dominated by the load uncertainty. Investment in more sophisticated analysis is only justified when the analysis uncertainty is the limiting factor.

When FEA Is Genuinely the Right Tool

FEA earns its place when the problem genuinely cannot be solved by hand methods to adequate accuracy:

Complex Geometry

Pressure vessel nozzle intersections, cast components with complex blended geometry, machined components with multiple stress-raising features, welded joints with complex load paths — these geometries produce stress distributions that analytical solutions do not capture. FEA is the appropriate tool to understand the actual stress concentration and its distribution through the section thickness.

Multiple Simultaneous Load Cases

When a structure is subject to pressure, thermal gradient, deadweight, wind load and seismic acceleration simultaneously, the superposition of hand calculations for each load case becomes cumbersome and the interaction between load cases is difficult to verify. FEA handles multiple simultaneous loads directly within a single model.

Post-Yield Assessment — Fitness for Service

When a structure has a defect, has been overloaded, or is being assessed against a fitness for service standard (API 579, BS 7910), the elastic stress distribution alone is insufficient. Elastic-plastic FEA, used alongside fracture mechanics assessment, provides a more realistic picture of remaining load capacity than elastic analysis alone.

Pressure Vessel Design by Analysis (DBA)

ASME VIII Division 2 and EN 13445-3 both permit pressure vessel design by analysis as an alternative to design by formula. This allows non-standard geometries to be qualified by FEA rather than by the standard thickness formulae. DBA in this context is not an alternative to code compliance — it is a code-defined analysis route with its own requirements for load combinations, stress categorisation, and acceptance criteria.

Vibration Assessment

Natural frequency calculation for complex multi-component structures, pipework support natural frequency assessment, or structural response to rotating machinery unbalance forces — these are cases where modal FEA provides information that is not readily obtainable by hand.

Optimisation

When the goal is to minimise mass or material cost while meeting stress and deflection constraints, parametric FEA models allow rapid exploration of the design space. This is a legitimate and powerful application of FEA, but requires a validated model as the starting point.

Reading FEA Results — The Colour Plot Problem

The visual output of FEA — contour plots of stress, displacement, or strain mapped onto the deformed geometry in a spectrum of colours — is simultaneously the most useful and the most misleading aspect of the method. Several specific issues affect interpretation:

Stress Singularities

In a linear elastic FEA model, stress at a perfectly sharp re-entrant corner is theoretically infinite. In practice, the FEA will produce a very high stress at that location that increases as the mesh is refined — this is a mathematical artefact of the linear elastic model applied to an idealised sharp corner, not a real physical stress. Real components do not have perfectly sharp corners, and real materials yield locally at high stress concentrations.

The key skill in FEA interpretation is distinguishing between a real high-stress region that drives a design change and a mesh-dependent singularity at a boundary condition or geometric discontinuity that should be disregarded or assessed differently. A singularity at a fixed constraint (a bolted joint modelled with an encastre boundary condition, for example) is expected and should not be used as the basis for a design decision.

Mesh Sensitivity

The stress in a region of high stress gradient — around a fillet, a notch, a hole — depends on the mesh refinement in that region. A coarse mesh will underestimate the peak stress; a fine mesh will capture it more accurately. The correct approach is to perform a mesh convergence study: progressively refine the mesh in the region of interest and confirm that the result is converging to a stable value. If the peak stress is still changing significantly as the mesh is refined, the result is not reliable.

Results in regions remote from stress concentrations are typically much less sensitive to mesh refinement. Displacement and overall structural stiffness converge faster than peak stress values.

Colour Scale Manipulation

The default colour scale in most FEA post-processors spans the full range from minimum to maximum stress in the model. If a single node has an artificially high stress (due to a point load or a singularity), the colour scale compresses all the physically meaningful stresses into a narrow band of similar colour, making the results look uniform when they are not. Adjusting the colour scale to exclude known singularities and show the range of interest is a standard part of post-processing — not data manipulation, but necessary to communicate results meaningfully.

Stress Categorisation

In pressure vessel assessment (ASME VIII Div. 2, EN 13445-3), stresses are not simply checked against a single limit. They are categorised as primary membrane, primary bending, secondary, or peak, and different limits apply to each category. Running FEA and checking the von Mises stress against the yield strength of the material without performing stress categorisation is not a code-compliant design by analysis — and may be either unconservative (if peak stresses are checked against the primary stress limits) or needlessly conservative (if secondary stresses are treated as primary).

What FEA Cannot Tell You

Understanding the limits of the method is as important as understanding its capabilities:

• FEA does not account for manufacturing defects. The model represents the ideal geometry. Weld flaws, porosity, surface roughness and residual stress from welding are not included in a standard stress analysis unless specifically modelled.
• FEA does not validate the loads. If the applied loads are wrong, the results are wrong. FEA provides no mechanism to verify that the input loads are physically representative.
• FEA does not account for creep, relaxation or long-term behaviour unless specifically set up as a time-dependent analysis — which requires additional material data and significantly more complexity.
• FEA does not replace code compliance. A structure that passes an FEA stress check has not been shown to comply with a design code unless the FEA has been performed within the framework of that code's design by analysis requirements.
• A linear elastic FEA result with stress above yield does not mean the structure has failed. Local yielding is permitted in ductile structures — the question is whether yielding is localised (acceptable) or spreads to form a plastic mechanism (failure). This requires non-linear analysis or yield line methods to assess correctly.

Validating an FEA Model

Any FEA model used for engineering decisions should be validated before the results are relied upon. Validation means demonstrating that the model gives results that are consistent with known behaviour. Approaches include:

• Hand calculation check: For a simplified geometry or loading condition, compare FEA results with the analytical solution. Displacement of a cantilever beam, stress in a pressurised cylinder, natural frequency of a simple supported beam — all have closed-form solutions. Agreement within a few percent gives confidence the model is set up correctly.
• Mesh convergence: Demonstrate that the results of interest have converged with mesh refinement and are not dependent on the mesh density.
• Symmetry check: If the geometry and loading are symmetric, the results should be symmetric. Asymmetric results from a symmetric model indicate an error in the model.
• Energy check: Most FEA solvers report total strain energy in the model. Comparing this between load cases and between mesh refinements provides a sanity check on whether the model is behaving consistently.
• Physical reasonableness: Does the deformed shape make intuitive sense? Does the structure deflect in the direction of the applied load? Are the regions of high stress where you would expect them — at the root of a cantilever, around a hole, at a section change? If not, something is wrong.

Competence and Accountability

FEA is a tool that can be operated without the competence to interpret the results correctly — and the outputs look the same either way. A colour plot produced by a competent analyst who has validated their model, performed mesh convergence, correctly categorised stresses and sanity-checked against hand calculations looks identical to a colour plot produced by someone who ran the software for the first time and accepted the default settings.

This creates a professional responsibility question that the engineering community has not fully resolved. In the UK, there is no formal licensing requirement to perform or certify FEA. Under CDM and the professional obligations of chartered engineers, the person signing off an FEA-based design decision carries personal responsibility for that decision. Before relying on FEA results — your own or someone else's — it is worth asking: has the model been validated? Has mesh convergence been demonstrated? Are the loads and boundary conditions physically representative? Has the result been checked against a hand calculation? If these questions cannot be answered, the result should not be relied upon.

Summary

FEA is a powerful and legitimate engineering tool when applied by a competent analyst to a problem that genuinely requires it. It is not a shortcut, it is not a replacement for engineering judgement, and a colourful result is not evidence that the result is correct.

Hand calculations should be the default for problems that can be solved analytically. FEA earns its place on complex geometry, multiple simultaneous loading, post-yield assessment, design by analysis under pressure codes, and vibration problems. In every case, the result should be validated, checked for mesh sensitivity, sanity-checked against simplified models, and interpreted by someone who understands the assumptions embedded in the analysis type.

The most useful thing a good FEA analyst brings is not the ability to run the software — it is the ability to set up the problem correctly, recognise when the results are not to be trusted, and communicate the limitations of the analysis alongside the results.