Statistical analysis of plastic tensile data converts raw stress-strain measurements into reliable, defensible material properties for product design, quality control, and certification. This technical guide explains how tensile results obtained under ISO 527 and ASTM D638 are evaluated using descriptive statistics, outlier screening, normality testing, Weibull analysis, and statistically substantiated allowable values. It covers specimen preparation, key measured parameters, analytical workflows, industrial applications, and expert answers to frequently asked questions.
Technical Overview
Tensile testing is the most widely performed mechanical test on plastics, generating fundamental properties such as tensile strength, yield stress, elongation, and modulus. However, polymers are inherently variable materials: molecular weight distribution, processing conditions, filler dispersion, molecular orientation, and environmental conditioning all influence measured results. A single tensile value therefore carries limited technical meaning without a statistical frame of reference.
Statistical analysis of plastic tensile data addresses this variability by quantifying central tendency, dispersion, and distribution shape across a sample population. The primary international frameworks are the ISO 527 series (ISO 527-1, ISO 527-2, ISO 527-3), ASTM D638, and the corresponding regional standards GB/T 1040 and JIS K 7161/7162. When performed by accredited third-party laboratories, the resulting datasets support incoming material acceptance, batch-to-batch comparison, process capability evaluation, and the derivation of statistically substantiated design allowables used in structural and regulatory contexts.
A rigorous statistical workflow typically moves through four stages: validity screening of individual test curves, descriptive statistics of replicate specimens, distribution identification and outlier management, and inferential analysis such as confidence intervals, hypothesis testing, or Weibull modeling. Each stage is described in the sections below.
Testing Samples
Injection-Molded Dumbbell Specimens
The reference geometry for rigid plastics is the injection-molded dumbbell or dog-bone coupon, typically ISO 527-2 Type 1A (4 mm thickness) or Type 1B, and ASTM D638 Type I. Specimens should be molded under documented and controlled conditions, since melt temperature, injection speed, and holding pressure directly alter measured strength and elongation. Multi-cavity molds require cavity identification so that cavity-to-cavity effects can be separated from true material variability.
Machined Specimens from Sheet, Plate, and Finished Parts
When material must be characterized in its as-manufactured state, specimens are machined from sheet, plate, pipe, or finished components in accordance with ISO 527-2 Type 1B or ASTM D638 Types I through V. Machining direction relative to extrusion or rolling direction must be recorded, because molecular orientation can create anisotropic tensile populations that must be analyzed as distinct statistical groups rather than a single pooled dataset.
Film and Thin-Sheet Specimens
Flexible films below approximately 1 mm are tested as rectangular strips under ISO 527-3 or ASTM D882. Film data frequently shows higher scatter than molded specimens due to thickness measurement sensitivity and grip effects, so larger replicate counts and stricter outlier criteria are often applied during statistical processing.
Additively Manufactured Specimens
Parts produced by fused filament fabrication, stereolithography, or other polymer additive processes are commonly evaluated using ASTM D638 Type V or ISO 527-2 Type 1B coupons. Build orientation, layer thickness, and infill strategy introduce systematic effects, so statistical designs typically treat each orientation as a separate population with its own mean, standard deviation, and distribution analysis.
All specimen families require conditioning prior to testing, normally 23 °C and 50% relative humidity per ISO 291 or ASTM D618, because moisture content and thermal history shift tensile properties measurably. Conditioning state must be fixed before any statistical comparison between datasets is attempted.
Testing Items
The statistical evaluation covers every parameter extracted from the tensile stress-strain curve, together with the derived statistical descriptors:
- Tensile strength at yield and tensile strength at break (MPa)
- Tensile stress at break and stress at specified strain, for example the nominal strain at 50% (MPa)
- Tensile stress at yield, the first stress at which strain increases without a stress increase (MPa)
- Tensile strain at yield and tensile strain at break (%)
- Percent elongation at yield and at break
- Tensile modulus, determined between 0.05% and 0.25% strain per ISO 527-1, or chord modulus between specified strain points per ASTM D638
- Secant modulus, for example the 1% secant modulus, where nonlinear initial behavior exists
- Poisson ratio, where transverse strain measurement is performed
- Statistical outputs for each parameter: sample mean, standard deviation, coefficient of variation, minimum and maximum, range, confidence intervals, and distribution-fit statistics
Testing Methods
Validity Screening of Individual Test Results
Before any calculation, each specimen result is screened for technical validity. Failures within the gauge length are normally accepted; failures at grips or outside the gauge section are flagged and treated according to a predefined exclusion rule. Stress-strain curves are inspected for extensometer slip, premature initiation artifacts, and compliance errors. Exclusions must be documented, justified, and applied before statistical processing, never after the statistics have been reviewed, in order to preserve the integrity of the analysis.
Descriptive Statistics
For each parameter, the sample mean, standard deviation, and coefficient of variation (CV, the ratio of standard deviation to mean) are computed from at least the five replicate specimens required by ISO 527 and ASTM D638, with larger samples preferred. As general engineering guidance, well-molded ductile thermoplastics often show tensile-strength CV values below 5%, brittle or filled materials may reach 8-12%, and higher values indicate processing inconsistency, conditioning problems, or test artifacts that warrant investigation.
Outlier Detection and Management
Suspect results are evaluated objectively using statistical outlier criteria such as the Grubbs test or the procedures of ASTM E178, rather than by visual judgment. An outlier is removed only when both a statistical justification and an identifiable physical cause, such as a void, machining defect, or grip slip, exist. When a cause cannot be established, the value is retained, and robust statistics or distribution-based methods are used so that a single specimen does not distort the reported properties.
Normality Testing and Distribution Fitting
The choice of downstream statistics depends on distribution shape. Normality is checked using the Shapiro-Wilk or Anderson-Darling tests, supported by probability plots. Tensile strength of ductile polymers is often approximately normal, while brittle fracture strength and elongation at break are frequently skewed and are better represented by Weibull or log-normal distributions. Selecting the correct distribution prevents underestimation of lower-tail strength, which is precisely the region of interest for design.
Weibull Analysis of Strength Data
For brittle plastics, fiber-reinforced compounds, and additively manufactured materials, two-parameter Weibull analysis is applied to fracture strength. The Weibull modulus characterizes scatter: a low Weibull modulus indicates wide strength dispersion and defect-dominated failure, while a high modulus indicates homogeneous material behavior. Characteristic strength at 63.2% failure probability, together with confidence bounds on the Weibull parameters, provides a defensible basis for reliability statements and material ranking.
Hypothesis Testing and ANOVA
Comparative questions, such as whether two resin lots or two formulations differ significantly, are answered with t-tests, Welch tests for unequal variances, or one-way analysis of variance for multiple groups, followed by post-hoc comparisons where differences are found. This converts subjective observations of batch variation into statistically supported conclusions at a declared significance level, typically 5%.
Confidence Intervals and Statistically Substantiated Allowables
Reporting a confidence interval alongside the mean expresses the precision of the estimate and prevents over-interpretation of small replicate sets. For design purposes, lower tolerance bounds, in which a specified proportion of the population is expected to fall above the value at a stated confidence level, are used as statistically substantiated allowables. Such values are far more conservative and far more defensible than minimum or average values from a handful of coupons.
Modulus Determination by Regression
Tensile modulus is obtained by linear regression over the prescribed strain window, with the coefficient of determination checked to confirm linearity of the fitted segment. Secant or chord methods are substituted where the initial curve is nonlinear, and the method used must accompany the reported value because ISO and ASTM modulus definitions are not numerically interchangeable.
Measurement Uncertainty
Finally, statistical analysis of tensile data is placed in the context of measurement uncertainty. Type A uncertainty is derived from the replicate scatter itself, while Type B contributions cover force calibration, extensometer accuracy, dimensional measurement, and grip alignment. Accredited third-party laboratories reporting under ISO/IEC 17025 combine these components so that reported statistics can be interpreted against the actual capability of the measurement process.
Applications
Statistical analysis of plastic tensile data is applied across the full product lifecycle:
- Incoming inspection and supplier qualification, where lot acceptance limits are set from statistically derived specification windows rather than single-value comparisons
- Production quality control, where tensile statistics feed statistical process control charts and capability indices to detect drift before nonconforming product is produced
- Material development, where ANOVA and confidence intervals identify genuine formulation, filler, or processing improvements amid natural material scatter
- Design engineering, where statistically substantiated allowable strengths support wall-thickness and safety-factor decisions for structural plastic components
- Automotive, aerospace, medical device, packaging, and electronics industries, where certification dossiers require demonstrated statistical control of mechanical properties
- Failure analysis and root-cause investigation, where tensile distributions from field parts are statistically compared against control material to confirm degradation, embrittlement, or processing faults
- Regulatory and customer submissions, where accredited third-party laboratory reports with documented statistical treatment strengthen technical credibility and reduce audit findings
FAQ
How many specimens are required for meaningful statistics?
ISO 527 and ASTM D638 specify a minimum of five specimens per condition. For statistical purposes, larger samples, typically eight to fifteen or more, are strongly recommended, because confidence intervals narrow with sample size and outlier decisions become more reliable. Design allowable calculations generally demand substantially larger datasets than routine QC testing.
Which standards govern the testing and the analysis?
Test execution follows ISO 527-1, ISO 527-2, and ISO 527-3, ASTM D638, ASTM D882 for films, GB/T 1040, and JIS K 7161/7162. Statistical methodology draws on ASTM E178 for outliers, ISO 5725 concepts for repeatability and reproducibility, and ISO/IEC 17025 for laboratory reporting and uncertainty.
Why report the coefficient of variation instead of just the average?
Two materials can share the same mean strength while differing enormously in consistency. The coefficient of variation expresses scatter relative to the mean, enabling fair comparison across materials and flagging processing or testing problems. Averages without dispersion data can be seriously misleading for design and acceptance decisions.
When is Weibull analysis preferred over normal statistics?
Weibull analysis is preferred when failure is defect-controlled and strength data is skewed, which is typical of brittle plastics, glass- or carbon-fiber-reinforced compounds, films, and additively manufactured parts. Normal-based statistics remain appropriate for approximately symmetric data, as confirmed by normality testing.
What are the main sources of scatter in plastic tensile data?
Principal sources include molding or processing variation, moisture conditioning deviations, specimen dimensional error, molecular and filler orientation, internal voids, grip alignment, and crosshead speed control. A well-designed statistical analysis distinguishes these contributions instead of masking them in a single spread value.
Can test results from ISO 527 and ASTM D638 be compared directly?
Not directly. Although broadly similar, the standards differ in specimen geometry, modulus strain windows, and data treatment, so results should each be analyzed within its own statistical framework. Cross-standard comparison should only be made with a documented conversion rationale or parallel testing on both geometries.