ASTM D7235 Guide for Establishing a Linear Correlation Relationship
ASTM D7235 Standard Guide for Establishing a Linear Correlation Relationship Between Analyzer and Primary Test Method Results Using Relevant ASTM Standard Practices
5. Calibration
5.1 When an analyzer is initially installed, and after major maintenance has been preformed, diagnostic tests should be conducted to demonstrate that the analyzer meets manufacturer's specifications and historical performance standards. These diagnostic tests may require that the analyzer be adjusted so as to provide predetermined output levels for certain reference materials. Such adjustment may be done in hardware, software or both. This should not be confused with the development of correlation to a PTM, which, is described below.
6. Correlation to Primary Test Method
6.1 General Approach:
6.1.1 Define the sample set to be used for assessment.
6.1.1.1 The material type and property range for which the analyzer system results are to be assessed versus the primary test method is defined.
6.1.1.2 The recommended sample set design criteria for this assessment are:
(1) A minimum of six replicates at each major product/property level combination,
(2) The range of major product/property levels exceed at least two times the published reproducibility of the PTM, and
(3) A minimum of 30 total samples.
6.1.1.3 Replication at a specific level/product combination can be different batches of production material that are nominally similar in property level (within 1.2 times the ASTM reproducibility of the PTM) and composition.
6.1.2 Obtain site precision information for the analyzer system and PTM for the material type and range defined in accordance with the procedures outlined in Practice D6299. If analyzer system precision cannot be obtained using the methodology in Practice D6299, manufacturer's published repeatability precision or site precision from other similar systems may be used as a surrogate. Users are cautioned to ensure consistency in statistical definitions between the manufacturer's published repeatability and the site precision statistic as defined in Practice D6299 for the PTM.
SAMPLING METHOD
6.1.3 Line Sampling (Preferred) - Samples meeting the requirements of the sample set design criteria above are taken in accordance with Practice D3764, Line Sample Procedure, at the crosscheck sample points of the analyzer system (see Fig. 1), after the sample conditioning subsystem, at a frequency of no more than once per day. Avoid taking this sample at the same time of day to ensure any time-of-day related effect is captured in the dataset.
6.1.4 Automated Composite Sampling - For installations that have automated composite sampling systems meeting the requirement of Practice D4177 (or equivalent), and, the application is intended to provide a predicted PTM result for a batch of homogenous production material using an Flow-Proportioned Average Property Value (FPAPV) calculated from the analyzer system as per Practice D6624, the correlation equation can be established using samples taken from the composite sample accumulator at the end of each batch of production meeting the requirements of the sample set design above. While this approach is theoretically equivalent to the line sampling approach, it is vulnerable to failure of assumptions that the requirements of Practices D6244 and D4177 are met.
ASSESSMENT OF TEST RESULTS
6.1.5 For Line Sampling method, assess the degree of agreement between results generated by the analyzer unit, captured at the time of sampling and the corresponding PTM.
6.1.6 For Automated Composite Sampling method, assess the degree of agreement between the FPAPV value generated in accordance with Practice D6624 using a non-bias corrected analyzer output for each composite sample versus the corresponding PTM result.
6.1.7 Using site precision information, and statistical principles/calculations of Practice D6708 (see Annex A1 for an example of assessment), answer the following questions:
6.1.7.1 Refer to Table 1 for the assessment outcomes associated with the answers to questions in 6.1.7.
6.1.7.2 If the D6708-based assessment outcome is successful, the scaling/bias-correction from D6708 assessment are mathematically applied to the analyzer unit results to produce a predicted-PTM result by the analyzer unit.
6.1.8 Obtain the dataset comprising (predicted PTM result - PTM result). This dataset is commonly known as the prediction residuals.
6.1.8.1 Plot the residuals in chronological order (a run chart as per Practice D6299).
6.1.8.2 Examine the plot for any cyclic or unusual patterns.
6.1.8.3 Calculate the root-mean-square standard deviation of the residuals (see Annex for example).
6.1.9 Interpret the usefulness of the degree-of-agreement between the scaling/bias-corrected analyzer unit and PTM test results versus the intended analyzer application, taking into account any sample specific biases observed, the magnitude of the standard deviation of the residuals when compared to the PTM site precision standard deviation, and how the scaling/bias-corrected analyzer results is used. It is recommended that this interpretation be carried out by a team comprising subject matter expertise in analytical technology, the manufacturing process that the analyzer system is intended to monitor, and statistical modeling.
6.1.10 If the degree of agreement after scaling/bias-correction is deemed useful, apply the correction to the analyzer unit test results mathematically in the routine use of the analyzer system.
6.1.11 For total analyzer systems that are used to generate FPAPV (per Practice D6624), the predicted-PTM result can be used to generate a bias-corrected FPAPV that is intended to predict the batch quality if the PTM had been used to analyze a representative sample of the homogenized batch of material.
6.1.12 The scaling/bias-correction equation is continually validated using the predicted-PTM result and actual PTM results in accordance with the line sample procedure of Practice D3764 and control charting technique in Practice D6299.
6.1.13 For FPAPV applications, the complete FPAPV system should be continually validated using predicted-PTM derived FPAPV and PTM results obtained on samples that the FPAPV is intended to represent, in accordance with control charting technique in Practice D6299.
6.1.14 Work procedures need to be developed to address out-of-control situations for the control charts described above. The specific tasks tend to be application specific, and will depend on the criticality of the application both in a real time sense, and business risk. In general, the following process of elimination should be carried out, in the order presented, to rule out the plausible root causes:
6.1.14.1 Is it due to sample difference between the PTM and the analyzer unit?
6.1.14.2 Is it due to the analyzer unit out of statistical control?
6.1.14.3 Is it due to the primary test method out of statistical control?
6.1.15 If the plausible root causes in 6.1.14 are ruled out, then it is highly likely that the out-of-control situation is due to the sample matrix being significantly different from those used to develop the scaling/bias-correction equation. Repeat the Practice D6708 assessment using results from more current data set based on the control chart performance.
6.2 Single-Level Bias Assessment - Single Level Initial Validations are typically performed using the Line Sample Procedure described in Practice D3764.
7. Analyzer System Application Configuration
7.1 Fig. 1 illustrates two common analyzer system application configurations applicable to this guide. Figs. 2 and 3 are reproduced from Practice D3764 to provide additional guidance.
