Petroleum and related products. Precision of measurement methods and results - Determination of precision data in relation to methods of test

Background and Technical Evolution of Standard Development

ISO 4259-1:2017 is the core international standard for determining the precision of petroleum product test methods. Its first edition was published in 2017, replacing the previous version ISO 4259:2006. Developed by the ISO/TC 28 (Petroleum products and related products) technical committee, this standard is closely coordinated with ASTM D6300 and D3244, aiming to unify the assessment methods for the precision of petroleum product test methods globally. The standard was refined through two revisions (in 2019 and 2020) to improve outlier detection (GESD technology) and weighted regression analysis procedures, enhancing statistical robustness.

Precision is a metric for measuring the random error of a test method, including repeatability (under identical conditions) and reproducibility (between different laboratories). This standard provides a systematic framework for designing interlaboratory studies (ILS), calculating precision estimates, and evaluating method applicability, widely used in refining, quality inspection, and certification fields.

Scope and Core Concepts

This standard applies to petroleum and petroleum-related products (such as fuels and lubricants), but the methodology can be extended to other homogeneous products. Key terms are as follows:

Precision often depends on the level of results. The standard requires determining whether data transformation (such as logarithmic or square root) is necessary through weighted linear regression to ensure homogeneity of variance.

Planning and Execution of Interlaboratory Studies (ILS)

The design of ILS must follow four stages:

1. Drafting the Test Method: Includes complete operational details and placeholders for precision statements.
2. Pilot Study: Involves at least two laboratories and 12 laboratory-sample combinations to verify method feasibility and preliminarily estimate precision.
3. Formal ILS Planning: Involves at least six laboratories (eight or more recommended) to meet the requirement of 30 degrees of freedom. The number of samples must cover the method range, and leverage values are used to assess sample influence.
4. Execution: Blind sample coding, randomized test sequence, and unified reporting forms to ensure independent repeated measurements.

The determination of the number of samples is based on Table A.1, which relies on the variance component ratios from the pilot study (P = σ²₁/σ²₀, Q = σ²₂/σ²₀) and the number of laboratories to ensure reproducibility degrees of freedom ≥30.

Statistical Processing Methods and Outlier Detection

Data preprocessing is a critical step:

• GESD Pre-screening (Section 5.2): Used to identify multiple outliers simultaneously, avoiding masking effects. The GESD process is applied separately to the differences (between repeatability) and sums (paired sums) for each sample, with a significance level of 1%.
• Data Transformation (Section 5.3): If the standard deviation is related to the mean, power transformations such as y = x^(1/3) are required to stabilize the variance.
• Stepwise Outlier Testing:
  • Cochran test (1% level) is used to identify repeatability outliers (excessive differences).
  • Hawkins test (1% level) is used to identify laboratory-sample combination outliers (outlier unit means).
  • Sample rejection test (Cochran or variance ratio test) removes samples that are globally anomalous.
• Missing Value Estimation (Section 5.5): Filled using the least squares method or iterative formulas.

All tests are conducted at a 1% significance level to avoid excessive removal. If the removal ratio exceeds 10%, the results must be retained and manually evaluated.

Variance Analysis and Precision Calculation

The following components are calculated through precise Analysis of Variance (ANOVA):

Repeatability variance V_r = 2σ²₀, Reproducibility variance V_R = 2(σ²₀+σ²₁+σ²₂). The final precision estimate must be multiplied by the t value (95% two-sided) and retained to 3-4 significant figures. If the reproducibility degrees of freedom are <30, a warning about insufficient standardization is required.

Additionally, the R/r ratio (Section 7), if greater than 4, indicates that laboratory bias is dominant, requiring enhanced standardization.

Differences from Previous Version ISO 4259:2006

Implementation Recommendations

1. ILS Design Phase: Ensure at least 8 laboratories are involved, and determine the number of samples by consulting Table A.1 based on the estimated P and Q from the pilot study. Use leverage values (leq>0.5) to exclude high-influence samples.

2. Data Processing: Prioritize the use of dedicated software (such as D2PP) to execute GESD and weighted regression. Confirm whether the transformation remains applicable after outlier removal.

3. Result Reporting: The precision function must clearly state its form (e.g., r = 0.028 x^(2/3)), accompanied by an ILS summary (number of laboratories, number of samples, range, degrees of freedom).

4. Quality Control: When laboratory bias is significant (F-test significant), consider additional standardization training or method revision.

5. Specification Application: When using reproducibility for product compliance assessment, consider the dispute resolution methods provided by ISO 4259-2 (utilizing test data).

Case Study: Application of Bromine Number ILS Data

Appendix D demonstrates the complete process using the bromine number as an example: 9 laboratories tested 8 samples. After GESD pre-screening and cube root transformation of the raw data, homogeneity of variance was improved. The Hawkins test identified an outlier for Laboratory D on Sample 1. After estimating missing values, ANOVA was completed. The final repeatability function is r = 0.148 x^(2/3), and the reproducibility function is R = 0.310 x^(2/3), with an R/r ratio of 2.1, indicating moderate laboratory bias.