Last verified 19 August 2026.
The four steps
Translation
Turn the design requirement into four things: the function the component performs, the constraints it must satisfy, the objective to be maximised or minimised, and the free variables you are allowed to change.
The Granta documentation puts it as “translate design requirements into a set of conditions for selecting a material.” It is the step most often skipped and the one that determines whether everything after it is meaningful. A constraint that reads “good corrosion resistance” cannot be screened against, so in practice it gets waved through — and a requirement that was never screened is a requirement that was never applied.
A translated constraint has a property, an operator, a value, a unit and a condition. An illustrative one, invented for the shape rather than for the number: PREN ≥ 32, seawater, 40 °C. Not an adjective.
Screening
“Constraints act as gates. If a material meets the demands of the constraints, it passes through to the next selection step.”
Screening is a filter, not a score. A material either clears the gate or it does not, and nothing is gained by weighting the gates against each other. Materials that fail are eliminated, and — this is the part people drop — the elimination and its reason are worth recording. Six months later the only question anyone asks is why you did not use the obvious alternative.
Ranking
“Surviving candidates are ordered by their ability to meet a criterion of excellence, such as minimizing weight or cost.”
This is where material indices earn their keep. An index is a combination of properties that determines performance for a given function, independent of the specific geometry — Ashby writes the performance equation as a product of functional, geometric and material terms, and the material term is the index. For a light stiff tie, it is E/ρ. For a light stiff beam in bending, E^(1/2)/ρ. For a light strong beam, σ^(2/3)/ρ.
The practical consequence is that you can rank materials before you have finished the geometry, which is exactly when the choice is still cheap to change.
Documentation
“Research shortlisted candidates thoroughly, considering case histories of failures, manufacturing limitations, and even global issues that might affect its availability.”
The step everybody agrees with and few complete. It is where the qualitative requirements live — weldability with your procedures, availability in your thickness, approvals history, supply chain — and they are real constraints even though they are not numeric.
Source: Ansys Granta EduPack, Materials Selection white paper, quotations verbatim.
Property charts
A material property chart is “a diagram with one (or a combination of) material attribute plotted on each axis.” Logarithmic scales, because engineering properties span many orders of magnitude, and material families occupy distinct regions — which is what makes the charts readable at a glance.
Two things make them more than a picture. A material index of the form M = Eᵃ/ρ plots as a straight line on log–log axes, so ranking becomes a matter of sliding a line of fixed slope across the chart. And the empty regions are as informative as the populated ones: they show you where no material exists, which is where a composite or a hybrid might.
The assumption that no longer holds
The method assumes the data comes from one curated database, human-checked, internally consistent. When Ashby wrote it, that was true — the method and the dataset were designed together, and the only realistic question about a number was whether you had read the right row.
A shortlist assembled today mixes specification minima from standards, typical values from mill datasheets, DFT results from open computational databases, machine-learned predictions, and numbers a language model produced without a source. They are rendered identically, to the same number of significant figures, in the same confident tone.
The method has no step for this, because it did not need one. Screening and ranking operate on values without asking what any of them rests on, and a ranking computed from a specification minimum and a model prediction looks exactly like one computed from two certified test results.
This is not a flaw in Ashby’s method. It is a change in the data supply chain that arrived thirty years after the method was written.
What we do about it
TRACE is our answer: keep the four steps, add one before screening, and give the fourth a defined structure.
| Ashby | TRACE |
|---|---|
| Translation | T — Translate, unchanged |
| (implicit) | R — Retrieve: assemble candidates, name every source, record the date |
| — | A — Assess: grade every value by how far it can be defended |
| Screening + Ranking | C — Compare |
| Documentation | E — Export: a record a third party can audit |
Assess is the only genuinely new step. It exists because there is now something to assess, and because a derived quantity — an index, a margin, a rank — can only be as defensible as its weakest input. Computing E^(1/2)/ρ from a certified modulus and a predicted density gives you a predicted index, whatever the arithmetic looks like.
If you take the method and skip the framework, take one habit from it anyway: write down where each number came from while you are deciding, not afterwards.
What this page does not cover
- Deriving material indices. The performance-equation derivation is in Ashby’s own text and is not reproduced here.
- Multi-objective selection, trade-off surfaces and penalty functions.
- Process selection, which Ashby treats with a parallel method.
- Hybrid and composite design, the use of the empty regions on a chart.
- Whether Granta is worth buying. It is a mature environment with a curated dataset. We are faster before you know what you are looking for and honest about what we do not have; that is a different claim.