Interpretation, Characterisation and Sizing

Purpose

Interpretation converts measured and reconstructed evidence into a defensible statement. Detection, localisation, characterisation, sizing and acceptance are separate decisions. A response can be detected without being confidently characterised. It can be located without its displayed width representing physical flaw width. It can be sized by a qualified method without the Level 2 technician having authority to accept the component.

The discipline in this chapter is to move from known setup and data quality through path plausibility, raw signals and complementary views before assigning a label. Image-pattern recognition can support the review, but it cannot replace acoustic reasoning. The final conclusion must be no stronger than the evidence and the qualified method permit.

Learning outcomes

After completing this chapter, the learner will be able to:

apply a repeatable indication-review sequence;

confirm component coordinates and path plausibility before classification;

corroborate responses using pair A-scans, amplitude images, phase/coherence views and complementary paths;

distinguish detection, localisation, characterisation, sizing and acceptance;

apply only qualified sizing methods and state important uncertainty contributors; and

document alternative explanations, limitations and escalation needs.

Crossover map: from indication response to evidence-led conclusion

The five decisions are not interchangeable

Detection

Detection means that a response meeting the technique’s review condition is present. It does not establish cause, flaw type or size. A response may be visible in one path and absent in another because of orientation, directivity or algorithm behaviour.

Localisation

Localisation assigns a coordinate or bounded region using a declared datum and path. Coordinate confidence depends on timing, velocity, wedge/surface model, element position, encoder registration, grid and the point-spread response. A cursor can report more decimal places than the technique has demonstrated.

Characterisation

Characterisation states what the response is most consistent with: relevant discontinuity, known geometry, volumetric scattering, planar/tip behaviour or another controlled category. Use cautious language when alternatives remain. “Response consistent with upper and lower diffracting features” is stronger technically than “crack tips” when no validated evidence establishes the reflector type.

Sizing

Sizing estimates a physical dimension using an approved method. The method may use tip positions, geometric endpoints, amplitude thresholds, multiple views or another qualified rule. Display width alone is influenced by the reflector's scattering response, the position- and path-dependent system response, and subsequent processing; it is not automatically a physical dimension.

Acceptance

Acceptance compares the reportable result with a governing acceptance source. It is not created by TFM, PCI, a coherence value or the colour red. If acceptance criteria, flaw category or authority are uncertain, report the observation within scope and escalate.

Repeatable indication-review sequence

Confirm dataset identity, component, scan range, setup revision and processing version.

Confirm successful pre/post checks and data-quality evidence.

Read the response in declared component coordinates, not only screen coordinates.

Identify the reconstruction path and draw physically possible transmit/receive legs.

Inspect representative supporting elementary A-scans and observed timing.

Compare amplitude and phase/coherence views with their separate scales.

Compare complementary paths that materially change the acoustic interaction.

Test geometry, multiple, lobe, model-error and data-quality explanations.

Characterise only to the level supported by repeatable evidence.

Apply a qualified sizing method and record uncertainty/limitations.

Apply acceptance only from the controlled external source and within authority.

Preserve the image, raw data, annotations and review trail.

The order matters. If setup or data quality is invalid, later image comparison does not repair it. If the path is physically impossible, a strong pixel response should trigger troubleshooting, not confidence.

Path plausibility and supporting signals

Begin with the view definition. Confirm modes, velocities, reflection/conversion surfaces and the position of the array relative to the component. Calculate or inspect predicted arrival times for several supporting Tx–Rx pairs. A valid image feature should normally be supported across an appropriate part of the aperture, not created by one dominant or faulty channel.

The supporting A-scans need not have identical amplitude. Transmit/receive separation, element directivity, reflector orientation and path transmission vary. Look for timing and phase behaviour consistent with the model, while checking for clipping, cross-talk, ring-down and noise. An amplitude envelope is useful for signal location but does not expose RF polarity or phase alignment.

Two views reconstructed from the same pair data with similar paths may be highly correlated. Agreement between them is useful but not equivalent to independent evidence. Stronger corroboration can come from a materially different path, another scan surface, conventional PAUT/UT, a known geometry relationship or a repeat acquisition.

Reflector behaviour and response morphology

Smooth planar reflectors tend to return specular energy toward only certain transmitter/receiver combinations. Tilting the face may make an amplitude response weaken or disappear even though the discontinuity remains. Tip-like or small volumetric features can scatter over a broader angular range and produce compact responses in several views.

A crack-like feature may produce an upper diffracting response, a lower tip or corner response and intermittent face response. A weld root, counterbore or surface profile can produce similar compact or elongated features. Phase-only coherence may emphasise point-like/diffracting events and suppress a broad planar face. Amplitude TFM may show the face or corner more strongly. Neither display is a literal photograph.

Interpret morphology using:

relationship to known surfaces and weld geometry;

movement and persistence across encoded positions;

response in paths chosen for different orientations;

raw RF timing and aperture distribution;

amplitude/coherence agreement or expected disagreement; and

results from representative targets in the performance demonstration.

Multi-view convergence and contradiction

Multi-view evidence is strongest when each view has a justified path, established sensitivity and a predicted response mechanism. Create a comparison table rather than scrolling between screenshots from memory. For each view, record response coordinate, domain, scale, expected reflector behaviour, supporting A-scans and credible alternative causes.

Contradiction is information. If a response appears at incompatible coordinates in direct and skip views, check velocity, thickness, surface and path assumptions. If PCI shows two compact responses but amplitude TFM shows strong geometry at the same positions, test both crack-tip and geometry explanations. If a view expected to detect a planar face is silent, consider orientation and sensitivity before declaring the face absent.

When material evidence remains genuinely contradictory and the conclusion could affect acceptance, stop and refer the case under the authorised review process.

Sizing methods and boundaries

Tip-to-tip height

Where two responses have been demonstrated to correspond to the upper and lower tips of a planar discontinuity, through-wall height can be calculated from their declared depth coordinates.

through-wall height = |z_lower tip - z_upper tip| (declared coordinate convention)

This simple subtraction is valid only after the tip interpretation, coordinate model and sizing method are qualified. A lower corner response is not automatically the lower crack tip. Two coherence hot spots are not automatically connected by a crack face. If the component coordinate has depth increasing downward, state that convention and use the absolute difference.

Length

Encoded length may be estimated from positions where a qualified response criterion is met, accounting for scan increment, point-spread extent, skew, component curvature and endpoint repeatability. A colour blob can extend beyond the physical reflector because of the acoustic response and smoothing. Report the endpoint rule and any interpolation used.

Amplitude-based sizing

Amplitude-drop or threshold methods apply only to amplitude-domain data under an approved technique. They require appropriate reference/sensitivity controls, linear acquisition and a qualified relationship to the target class. Do not use a coherence scale, independently normalised image or phase-only display for a conventional dB-drop calculation.

Geometric sizing

Geometry-based methods use known surfaces, path intersections or paired responses. They depend heavily on accurate component geometry, thickness, velocity and probe registration. If the assumed profile differs from the actual component, precise cursor positions can still be systematically wrong.

Uncertainty and resolution

Uncertainty is not just pixel spacing. Important contributors include:

time-zero and sampling/interpolation behaviour;

material velocity and temperature variation;

wedge, surface and element geometry;

encoder scale, zero and scan increment;

point-spread width and target orientation;

path identification and mode conversion;

threshold or endpoint rule;

analyst repeatability; and

difference between calibration target and production reflector.

Use the uncertainty approach defined by the procedure or technical authority. At minimum, report resolution/limitations and avoid unjustified precision. A calculated height of 3.27 mm should not be reported to 0.01 mm when demonstrated repeatability is ±1 mm.

Field workflow: indication evidence package

Open the locked production dataset and verified recipe.

Record the response in component coordinates and assigned indication ID.

Capture fixed-scale views with path, ROI, axes and scan position visible.

Retain representative raw A-scans and matrix/aperture evidence.

Complete the path-plausibility and alternative-cause checks.

Compare qualified amplitude, coherence and complementary views.

Apply the authorised characterisation vocabulary.

Perform the specified sizing method with repeat measurements.

State uncertainty and inaccessible or ambiguous regions.

Route acceptance or technical uncertainty to the authorised decision maker.

Practical 14 — Blind multi-view interpretation

Aim

Interpret encoded datasets containing discontinuities and geometry responses without knowing the answer key, then defend the evidence before reveal.

Resources

controlled datasets with valid pre/post checks;

component drawing and datum definition;

controlled path/view descriptions;

representative-target performance evidence;

authorised sizing instructions for selected indications; and

indication/evidence worksheet.

Method

Verify file identity, configuration, coordinates and checks.

Review the required amplitude and coherence views at fixed scales.

Assign an ID to every response meeting the review condition.

Draw the assumed path and inspect supporting A-scans for each response.

Record at least one credible alternative explanation.

Compare complementary paths and note both convergence and contradiction.

Characterise each response only to the supported category.

Apply the authorised sizing method where its prerequisites are met.

Repeat measurements after a break or by a second reviewer and compare results.

Submit the evidence table, uncertainty statement and escalation items before the specimen truth is revealed.

Evidence to retain

indication table with coordinates and path/domain;

fixed-scale annotated views;

supporting A-scan extracts;

path drawings and alternative explanations;

sizing worksheets and repeated measurements;

uncertainty/limitations statement; and

pre-reveal conclusion plus post-reveal learning review.

Stop rules

Stop if the dataset or checks are invalid; the assumed path cannot be reconciled with raw timing; an unqualified sizing method is requested; coordinate convention is uncertain; contradictory evidence could change the conclusion; or acceptance criteria/authority are missing. Do not use the answer key, a colleague’s classification or a hidden vendor tool to fill an evidence gap.

Common traps

Treating an image as a photograph of the flaw.

Using blob width as physical reflector width.

Calling compact PCI responses “tips” without qualification.

Counting correlated reconstructions as independent confirmation.

Ignoring a silent view that should have been sensitive.

Using dB-drop sizing on a coherence scale.

Reporting pixel pitch as uncertainty.

Rounding away contradictory coordinates.

Allowing acceptance criteria to leak in from another project.

Characterising beyond the available evidence to avoid escalation.

Chapter summary

Interpretation is an evidence sequence. Confirm data quality, coordinates and path; inspect supporting signals; compare genuinely complementary amplitude and coherence views; challenge geometry and artefact explanations; then characterise and size only within qualified limits. Detection is not characterisation, sizing is not acceptance, and a cursor position is not automatically a physical tip.

Tip-to-tip, amplitude and geometric sizing can all be useful when their assumptions are demonstrated. Uncertainty includes model, path, waveform, encoder, point-spread and reviewer effects—not only the grid. The strongest Level 2 conclusion is sometimes a clear, well-supported escalation.