Extreme value analysis (EVA) converts routine metallographic inspection of pressure vessel steel into a statistically rigorous estimate of the largest non-metallic inclusion likely to exist anywhere in a plate, forging, or weldment. This technical article explains the sampling strategy, statistical methods (Gumbel statistics of extremes and the Generalized Pareto Distribution), applicable international standards, and fatigue-related applications of inclusion extreme value analysis for boiler and pressure vessel steels. It is written for metallurgists, welding and design engineers, quality assurance personnel, and accredited third-party laboratories responsible for material qualification and steel cleanliness assessment.
Technical Overview
Pressure vessel steels—such as Q345R and Q370R boiler plates, SA-516 carbon steel, SA-387 Cr-Mo alloy steel, SA-508 heavy forgings, and P355GH-type normalized steels—must maintain high toughness, weldability, and long-term structural integrity under cyclic pressure loading. Non-metallic inclusions such as manganese sulfides, alumina clusters, silicates, calcium aluminates, globular oxides, and titanium nitrides act as internal stress concentrators and are the dominant initiation sites for fatigue cracks, particularly in the very-high-cycle fatigue (VHCF) regime above 10 million cycles where fish-eye fractures nucleate at subsurface inclusions.
Conventional chart rating methods described in ASTM E45, ISO 4967, and GB/T 10561 classify inclusions into A (sulfides), B (aluminates), C (silicates), D (globular oxides), and DS (single large globular oxide) types with thin and thick series ratings. While valuable for lot-to-lot cleanliness comparison, chart rating describes only the inspected field and provides no quantitative information about the extreme tail of the size distribution—precisely the information needed to judge whether a large, rare, potentially damaging inclusion could exist elsewhere in a multi-ton component.
Extreme value analysis addresses this gap. By measuring the largest inclusion in each of many standard inspection fields, fitting an extreme value distribution to those maxima, and extrapolating to the volume of the actual component, EVA yields a defensible statistical estimate of the maximum inclusion size in service. Combined with the Murakami square-root-of-area (√area) model, this estimate feeds directly into lower-bound fatigue strength calculations used in design verification and fitness-for-service evaluation.
- Statistics of extreme values (SEV) based on the Gumbel distribution, applied to field-by-field maxima
- Generalized Pareto Distribution (GPD) analysis of all inclusions exceeding a defined size threshold
- Conversion of two-dimensional section measurements into estimates of the largest three-dimensional defect in the component
- Integration with hardness data to estimate VHCF fatigue limits via the √area parameter
Testing Samples
Pressure Vessel Plates
Samples are taken from finished plates of Q345R, Q370R, 16MnDR, SA-516, SA-387, or EN 10028 grades at positions defined by the purchase specification, typically quarter-thickness and mid-thickness locations at specified points across the plate width. Because inclusion size and morphology vary with solidification and rolling history, sampling should capture surface, mid-thickness, and through-thickness gradients. Full-thickness or stepwise sections are prepared so that segregation bands, where the largest inclusions concentrate, are included in the inspection plan.
Forgings
For reactor pressure vessel shells, nozzle cuttings, flanges, and tube-sheet forgings, samples are obtained from prolongations or trepanned cores at bore-surface, mid-radius, and outer-surface locations. Both axial and tangential orientations should be examined, since elongated B- and C-type inclusions are strongly anisotropic and the orientation of the polished section relative to the working direction influences measured lengths.
Pipe, Tube, and Weldment Samples
Seamless and welded pipe samples include base metal, weld metal, and heat-affected zone (HAZ) cross sections. For weld procedure qualification records and production welds, separate polished sections of weld metal, fusion line, and HAZ are prepared, because welding can introduce exogenous oxide and slag inclusions whose size distribution differs fundamentally from that of the base steel and should not be pooled with indigenous inclusions in a single statistical fit.
Specimen Preparation Requirements
Metallographic preparation follows ASTM E3 and ASTM E768 practice: sectioning without burning or smearing, mounting, and polishing to a finish of 1 μm or finer so that inclusion edges are sharp and pull-out is avoided. Over-polishing that leaches or plucks inclusions biases the extreme value result and must be controlled. Specimens are typically examined in the unetched condition, with light nital etching used only for orientation reference.
Testing Items
Accredited third-party laboratories and in-house metallurgical departments typically report the following deliverables from an inclusion extreme value study:
- Inclusion type classification (A, B, C, D, DS) and thin/thick series ratings per ASTM E45, ISO 4967, GB/T 10561, or EN 10247
- Maximum observed inclusion size in each inspection field, expressed as √area, projected length, or equivalent diameter
- Fitted Gumbel (SEV) parameters: location λ and scale α, with the coefficient of determination of the Gumbel plot
- Fitted GPD parameters: threshold u, scale σ, and shape ξ, with maximum-likelihood confidence intervals
- Predicted maximum inclusion size for declared reference volumes, for example 1,000 mm³, 10⁶ mm³, or the full plate volume, at stated confidence levels
- Number density, area fraction, and size distribution of the inclusion population as supporting cleanliness indices
- √area of the estimated largest inclusion and, where required, the corresponding estimated fatigue limit using the √area model
- Comparative assessment between heats, suppliers, refining routes, or weld lots against specification acceptance criteria
Testing Methods
Microscopic Examination and Image Analysis
Polished sections are scanned by optical microscopy at magnifications of 100× to 500×, or by automated scanning electron microscopy with energy-dispersive spectroscopy (SEM-EDS) feature analysis, following the image-analysis principles of ASTM E1245. Each inspection field of defined area S0 (commonly 0.5–1 mm² or a user-specified area) is surveyed, and the largest inclusion in the field is measured by calibrated image analysis. SEM-EDS adds chemical typing of every extreme particle, which is essential for distinguishing indigenous deoxidation products from exogenous refractory or slag inclusions.
Statistics of Extreme Values (SEV / Gumbel Method)
A minimum of 30–60 observation fields is recommended. The field maxima are ranked in ascending order, and each is assigned the reduced variate y = −ln[−ln(i/(N+1))], where i is the rank and N the number of fields. Plotting size against y on Gumbel probability paper should yield a straight line; linear regression gives the scale parameter α (inverse slope) and location parameter λ (intercept). The estimated maximum inclusion in a larger area or volume S is then obtained from the return period T = S/S0 as x = λ − α·ln[−ln(1 − 1/T)], which approaches λ + α·ln(T) for large T. Linearity of the Gumbel plot (R² typically above 0.95) is the key validity check: curvature or step changes indicate mixed inclusion populations that must be separated by chemistry or morphology before fitting.
Generalized Pareto Distribution (GPD) Method
In the GPD approach, all inclusions exceeding a threshold u—commonly set near the 90th–95th percentile of measured sizes or at a fixed dimension such as 5–10 μm—are recorded over a large scanned area. The exceedance sizes are fitted to the generalized Pareto distribution by maximum likelihood, yielding scale σ and shape ξ. The maximum size expected in volume V is estimated as x(V) = u + (σ/ξ)·[(V/V0)^ξ − 1], where V0 is the inspected volume. A positive shape parameter signals a heavy-tailed, unbounded distribution; ξ ≤ 0 indicates a physically bounded maximum. Because GPD uses many data points above the threshold rather than one maximum per field, it extrapolates more reliably to very large volumes and is preferred when sparse DS-type oxides govern the extreme tail.
Complementary Three-Dimensional Techniques
- Electrolytic dissolution of the steel matrix to isolate inclusions and measure their true three-dimensional size distribution without sectioning truncation effects
- Serial sectioning and X-ray computed tomography for direct volumetric detection of large individual particles in critical regions
- Total oxygen and nitrogen chemical analysis as bulk cleanliness indicators correlated with the extreme value results
Data Reporting and Statistical Quality Control
Reports state the number and area of fields, magnification, the measured size parameter, distribution parameters with goodness-of-fit statistics such as Kolmogorov–Smirnov or Anderson–Darling tests, and predicted maxima with confidence intervals, typically using bootstrap resampling. Any exclusion of exogenous inclusions from the fitted population must be justified by chemistry and morphology evidence and documented transparently.
Applications
Inclusion extreme value analysis supports the full life cycle of pressure vessel materials:
- Material qualification and acceptance of plates, forgings, and pipe for boilers, reactors, separators, and LPG/LNG storage vessels, where predicted maximum inclusion size is compared against specification limits
- VHCF design verification for components subject to pressure pulsation, such as pulsation dampeners and compressor-related pressure parts, using σw ≈ 1.56(HV+120)/(√area)^(1/6) MPa for interior inclusions (1.43 for surface inclusions), with √area in micrometers and HV the local Vickers hardness
- Steel cleanliness benchmarking across heats and suppliers, and quantitative evaluation of secondary refining practices such as vacuum degassing, calcium treatment, and electroslag remelting
- Root-cause investigation of fish-eye fractures, hydrogen-assisted cracking at inclusion interfaces, and other inclusion-initiated failures, feeding into fitness-for-service and remaining-life assessments
- Welding consumable and procedure development, comparing weld metal inclusion populations across flux systems and heat inputs
- Research into oxide metallurgy and inclusion engineering for next-generation high-strength vessel steels
FAQ
Why use extreme value analysis when ASTM E45 or ISO 4967 ratings are already required? Chart ratings index severity in the inspected field but cannot extrapolate to component volume; EVA quantifies the statistical tail and estimates the largest inclusion the vessel may contain, which is the quantity that governs fatigue initiation risk.
How many fields must be inspected? At least 30 fields are recommended for a stable Gumbel fit, with 50–100 preferred for critical components; GPD analysis requires larger scanned areas to capture a sufficient number of exceedances above the threshold.
Which method is better, SEV or GPD? SEV is simple, well established, and ideal for moderate extrapolation volumes; GPD uses more data and extrapolates more reliably to very large volumes where rare, sparse inclusions dominate. Best practice runs both and compares the predicted maxima.
What size parameter should be measured? The √area parameter is preferred because it links directly to fatigue strength models; elongated A-, B-, and C-type inclusions may be evaluated separately by length or projected area since their populations behave differently.
Can the result predict fatigue life? EVA supplies the maximum inclusion size, which combined with local hardness in the √area model yields a lower-bound fatigue limit estimate for VHCF design; the analysis identifies the defect but does not by itself compute cycles to failure.
Does EVA replace conventional cleanliness testing? No. It complements chart ratings, image-analysis cleanliness indices, and total oxygen measurements as part of a complete material qualification package assembled by accredited third-party laboratories or qualified in-house facilities.
How are exogenous inclusions handled? Large refractory or slag particles from a different origin break the single-population assumption of the Gumbel fit. They are identified by SEM-EDS chemistry and morphology, fitted separately, or excluded with documented justification, since one oversized exogenous particle can otherwise distort the entire extrapolation.