Materials Science · Atoms, bonding, structure and defects

Evidence linking structure to properties

Turn a micrograph and a strength difference into a testable explanation: distinguish competing mechanisms, choose matched comparisons and complementary measurements, and keep specimen replication separate from repeated image fields.

Useful background: Materials, questions and levels of evidence. Each worked case explains the quantities it uses.

Back to the chapter

After reading, you can

Explain why a structure–property association does not isolate its cause.
Construct comparisons that distinguish a grain-size explanation from a particle-state explanation.
Calculate and interpret conditional differences without turning a small teaching dataset into a universal material law.
Choose structural and property measurements for the question, and distinguish field sampling from independent specimen replication.

A picture and a property are a starting pair

Suppose one alloy image shows finer grains and its measured yield stress is higher. The observations suggest a relationship. They do not yet show that grain size alone caused the difference: composition, texture, treatment and test conditions may also have changed.

Yield stress is the stress at a stated criterion for the onset of permanent deformation. A comparison must use the same criterion, loading direction, specimen geometry, temperature and deformation rate. Otherwise the reported difference might include a changed test as well as a changed material. A grain image does not measure yield stress; a tensile result does not identify every structural cause.

A mechanism is a proposed physical route from a feature to a response. In a metal deforming mainly by dislocation motion, a grain boundary can obstruct transfer of slip between differently oriented grains. Suitable dispersed particles can also obstruct dislocations. Refining grains and changing particles are therefore two plausible explanations for a strength change. Their contributions need not be independent or the same at every temperature and scale.

State an alternative before choosing evidence. If grain size alone explains a comparison, changing particle state at the same grain-size class should not change the expected response. If particle state alone explains it, changing grain-size class at a matched particle state should not change the expected response. Those are predictions to test under stated conditions, rather than explanations selected after seeing a favorable image.

Look for a mechanism and a controlled comparison

Specify how the structural feature could influence the response, then choose measurements that distinguish alternatives. Examine representative regions and multiple specimens. Use imaging, diffraction and composition measurements for their respective strengths; a striking micrograph is not automatically a quantitative description of the whole specimen.

Start from a defined target: yield stress in a chosen loading direction for specimens from a specified alloy and preparation route. Use specimens from a common source where appropriate, record preparation, and assign independent blanks to the planned treatments. Random assignment and randomized test order help avoid systematically pairing one treatment with a different starting population or an instrument drift. They do not guarantee that all treatments change only the feature of interest.

For the grain/particle question, try to prepare four combinations: coarser grains with sparse obstacles, finer grains with sparse obstacles, coarser grains with dense obstacles, and finer grains with dense obstacles. Here sparse and dense are stipulated particle-state classes characterized by suitable measurements, not just a count of bright spots. Check the grain and particle states after treatment and document other changes, including texture, porosity, matrix composition and prior deformation.

Compare finer with coarser grains within one particle class, then repeat in the other class. Compare the particle classes within a grain class as a complementary check. A comparison is only as controlled as its preparation, measurement resolution and remaining alternatives. If the four states cannot be made without changing other relevant features, report that limitation; do not call an unavailable control completed.

Random assignment can support an inference about the assigned treatment routes. Attributing that treatment effect to grain size specifically still needs evidence connecting the route to the structural change and checking other plausible routes to the response. A mechanism and a controlled comparison strengthen one another; neither makes the other unnecessary.

Work through the alternatives on one chosen dataset

The table is an invented teaching record, not measurements of a named alloy. Each row contains three independently prepared and tested specimens under the same stipulated yield criterion and test conditions. The grain and particle classes are supplied model information. Their actual physical realization, matching tolerances and uncertainty would need evidence in a real programme.

The first and last rows alone give means of 100 and 200 MPa. Their difference is 100 MPa, but both structural classes changed. It cannot be assigned entirely to grain refinement. The added rows let us ask more specific questions instead of merely repeating that the original pair was confounded.

Within the sparse-particle class, the finer-minus-coarser mean difference is 120 − 100 = 20 MPa. Within the dense-particle class it is 200 − 140 = 60 MPa. Conversely, changing particle class gives 140 − 100 = 40 MPa for coarser grains and 200 − 120 = 80 MPa for finer grains. These are conditional sample contrasts. Each grain contrast holds particle class fixed; each particle contrast holds grain class fixed.

The two grain contrasts differ by 60 − 20 = 40 MPa. This dependence of one contrast on the other factor is an interaction in the chosen means. A model that adds one fixed grain increment and one fixed particle increment cannot reproduce all four means. This calculation describes the record; it does not identify a unique microscopic interaction mechanism.

Only three specimen results are supplied per state. They show scatter, not a known population distribution or confidence interval. The plot displays individual values and means without declaring statistical significance. Overlap of observed ranges is not a significance test, and absence of overlap would not settle every causal alternative. A larger, justified design and uncertainty analysis are needed for a real quantitative claim.

Original stipulated teaching results. Three independent specimen yield stresses per state, all in MPa. No source data, alloy calibration or qualification limit is supplied.
Grain class	Particle-state class	Specimen yield stresses (MPa)	Arithmetic mean (MPa)
Coarser	Sparse	90, 100, 110	100
Finer	Sparse	105, 120, 135	120
Coarser	Dense	125, 140, 155	140
Finer	Dense	185, 200, 215	200

Original chosen teaching data, not alloy measurements. Coarser and finer grain states are categorical positions. Sparse-particle specimen values are 90, 100, 110 and 105, 120, 135 MPa; dense-particle values are 125, 140, 155 and 185, 200, 215 MPa. Small blue circles and orange diamonds show individual specimens. Large outlined squares mark means 100, 120, 140, 200 MPa, joined only to compare means: the solid sparse line rises 20 MPa and the dashed dense line rises 60 MPa. No continuous grain-size relation, confidence interval or statistical significance is supplied.

Choose evidence that could change the conclusion

The relevant observation depends on the alternative being tested. A calibrated image or orientation map can describe grain regions and their distribution on the examined section. Electron backscatter diffraction uses local diffraction patterns to infer crystal orientation and phase against a structural interpretation. Its result depends on pattern quality, sampling spacing, indexing and the chosen grain-boundary criterion; colored regions are not independent mechanical specimens.

Characterize particles at a scale that resolves the relevant sizes and spacings. Chemical information helps distinguish features that only look similar; structural information helps test phase identity. An element being present does not alone determine its arrangement. Conversely, a single structural measurement with a detection limit does not prove that no undetected feature exists. Use complementary evidence for the alternative that matters.

If the proposed grain effect remains while the measured particle state is matched, that observation challenges a particle-only explanation. If strength instead follows particle state while the grain distributions remain matched, a grain-only explanation is challenged. If the contrasts depend on both states, investigate their joint effects and remaining features. These are bounded interpretations of specified comparisons, not universal promises that finer grains or denser particles always help.

A strong follow-up has an outcome that could count against the preferred explanation. For example, measure grain distributions and particle state without using the specimen strength ranking to choose the image fields. Then examine whether the planned matched comparisons hold. Adding many photographs selected because they look convincing supplies less discrimination than one properly planned comparison with a clear competing prediction.

Complementary questions for the same material comparison. Each method answers a defined question and retains its resolution and sampling limits.
Question	Evidence to seek	Unjustified shortcut
Did grain structure change?	Spatially sampled grain/orientation measurements with scale and boundary criterion	Count image colors as phases or assume one field describes the specimen
Did the particle state change?	Resolved particle geometry plus appropriate chemical/structural identification	Treat every bright feature as the same obstacle
Did the response change?	Independent specimen tests with a common yield criterion and stated conditions	Use image count as the number of tensile tests
Can it be assigned to one feature?	Matched contrasts, competing predictions and checks on other changes	Label every treatment effect a grain-size effect

Worked examples

Complete the confounded heat-treatment comparison

Two heat treatments change both grain size and precipitates. Their chosen mean yield stresses are 100 MPa for coarser/sparse and 200 MPa for finer/dense. Can the 100 MPa difference be assigned entirely to grain size? Use all four rows to build a more discriminating argument.

The two-row difference is 200 − 100 = 100 MPa. Both classes changed, so that comparison alone does not separate the explanations.
At sparse particle state, compare finer/sparse with coarser/sparse: 120 − 100 = 20 MPa. At dense state, compare finer/dense with coarser/dense: 200 − 140 = 60 MPa.
A particle-only account with no grain-related change in expected response would predict zero grain contrast at matched particle state. These chosen point estimates do not follow that pattern. But three specimens per state do not by themselves establish the significance or exclusive cause of a population difference.
A grain-only account is also incomplete for these means: at matched grain class, the particle-state contrasts are 40 and 80 MPa. A constant additive-increment account fails because the grain contrasts differ by 40 MPa.
Check whether the stipulated matching can actually be achieved, characterize other changes and repeat independent specimens under the same test conditions. The original 100 MPa cannot be relabeled a pure grain increment; it remains a combined-state contrast.

No. The complete comparison supports investigating both state variables and their conditional behavior; it does not isolate the original difference as a grain-size effect or establish a universal alloy law.

Test a model rather than fit only the favorable pair

A proposed additive model starts from the coarser/sparse mean of 100 MPa. It assigns a fixed grain increment from the sparse comparison and a fixed particle increment from the coarser comparison. What does it predict for finer/dense, and how does that compare with the supplied mean?

The sparse comparison supplies the chosen grain increment 120 − 100 = 20 MPa. The coarser comparison supplies the chosen particle increment 140 − 100 = 40 MPa.
The additive prediction is 100 + 20 + 40 = 160 MPa for finer/dense.
The supplied finer/dense mean is 200 MPa, leaving an observed-minus-predicted discrepancy of 40 MPa. It is the same difference of conditional contrasts calculated above.
The fixed-increment model does not reproduce these four sample means. The discrepancy is a reason to examine interaction, unmeasured state changes and sampling uncertainty; it is not direct proof that a particular dislocation mechanism supplied exactly 40 MPa.

It predicts 160 MPa, 40 MPa below the supplied 200 MPa mean. The record does not support that simple fixed-increment description as an exact account.

Count the evidence at the level of the claim

One treated specimen supplies the three equal-area masks. A report keeps only the 30-feature-cell field and says “30% feature fraction, confirmed by three independent tests.” Identify the numerical and replication errors.

The three fields contain 60 feature cells among 300 equal-area cells. Their pooled examined-area fraction is 20%, rather than the selected field’s 30%.
The selected value differs by 30 − 20 = 10 percentage points, or (30 − 20)/20 = 0.50 = 50% relative to the pooled value. Neither comparison establishes a true bulk fraction without a sampling argument.
All three fields came from the same specimen. They are spatial observations of one specimen, not three independent specimen-level mechanical tests.
Retain the field selection plan, scale, classification and specimen identity. Add independently prepared specimens for a between-specimen response claim, and keep field variation separate from specimen variation.

The retained fields give a pooled 20% examined-area fraction. Three image fields cannot be renamed three independent mechanical tests.

Try the reasoning

Why examine more than one image field? In the supplied equal-area masks, calculate the pooled fraction rather than choosing the most striking field.
Show the answer and reasoning
The pooled fraction is 60/300 = 20%. Several planned fields help reveal spatial heterogeneity and reduce reliance on a selected view.
The result describes the examined areas. Representative bulk inference and between-specimen repeatability need their own sampling evidence.
Does a positive correlation between finer grains and higher strength make a mechanism unnecessary? Give one alternative and a comparison that distinguishes it.
Show the answer and reasoning
No. Changed particle state could accompany grain refinement. Compare grain classes at matched particle state, while checking other relevant changes and test conditions.
The mechanism supplies a prediction that can be challenged. The initial association alone does not isolate which co-changing feature affects dislocation motion.
A different invented four-state record has means: coarser/sparse 150, finer/sparse 180, coarser/dense 170 and finer/dense 200 MPa. Calculate both grain contrasts. Does this record require the same conditional-contrast difference as the lesson’s main record?
Show the answer and reasoning
Both grain contrasts are 30 MPa: 180 − 150 = 30 and 200 − 170 = 30. Their difference is zero, rather than 40 MPa.
The means are consistent with a constant additive grain increment over these two states. That consistency is not proof of a unique mechanism, statistical precision or validity at untested conditions.
A claim says that a color difference in a polished image proves the yield increase came from a new crystal phase. What complementary questions should be answered before accepting that explanation?
Show the answer and reasoning
Determine what causes the image contrast, seek appropriate structural and chemical evidence for the proposed phase, measure the response under matched conditions and test competing explanations such as orientation or particle changes.
Color alone is not phase identification, and detecting a phase alone does not assign the mechanical change to it. Each measurement must address a step in the explanation.
A group images ten fields from each of two specimens per treatment. May it report twenty independent specimen yield tests per treatment if each specimen was tensile-tested once?
Show the answer and reasoning
No. It has two specimen yield results and twenty image fields per treatment, with fields nested within specimens.
Extra spatial measurements can improve structural characterization without supplying the independent specimen outcomes needed for a specimen-level strength claim.
Independent blanks were randomly assigned to two heat treatments. The strength difference repeats, but every fine-grain specimen also has a changed matrix solute content. What can random assignment support, and what causal attribution remains unresolved?
Show the answer and reasoning
It can support a comparison of the assigned treatment routes under the study conditions. It does not by itself separate grain-size effects from accompanying solute or other structural changes.
Randomization addresses systematic allocation differences; it does not force a treatment to change only one physical feature. Additional discriminating comparisons and mechanism evidence are still needed.

What to carry forward

A structure–property explanation connects a defined response to a proposed mechanism and a comparison that could challenge it.
Matched conditional contrasts expose alternatives hidden by a comparison that changes several features at once.
An interaction describes dependence of a contrast on another state; it does not alone identify the microscopic cause.
Complementary structural, chemical and mechanical evidence answer different questions.
Image fields, specimens and batches are different sampling levels; a striking picture is not a population result.

References

Teaching resources and research supporting this lesson’s explanations and examples.