Introduction

While geological reference materials (RMs) have existed for many years, their use has become widespread only in the last decade. This increase in popularity stems from the recognition of the critical role RMs play in monitoring the quality of assay data generated in analytical laboratories. To the geologist, reference materials have application in grass roots exploration, resource definition, minesite exploration and grade control. To the chemist they facilitate the calibration of analytical equipment, evaluation and validation of analytical methods and routine in-house Quality Assurance/Quality Control.

Many terms are used to describe RMs or Certified Reference Materials (CRMs) and this can be confusing to the uninitiated. The term ‘standard’ strictly refers to an (ISO) approved and documented procedure in industry but its use as a synonym for RMs is widespread and entrenched throughout the mining and chemical industry. Other terms in common usage include Standard Reference Materials (SRMs), In-house or Internal Reference Materials (IRMs) and Matrix-Matched or Mine Matched Certified Reference Materials (MMCRMs). See our Glossary of Terms for further detail.

Essential attributes

  • Proven Homogeneity

    The CRM in question must have a proven level of homogeneity such that the observed variance in repeat assays can be attributed almost exclusively to measurement error. In other words, any sampling error resulting from inhomogeneity of the reference material should be small enough in comparison to measurement error that it’s negligible.

  • Statistically Robust Characterisation

    The CRM should be well characterised by round robin evaluation at a minimum of 10 recognised mineral testing laboratories and certified in accordance with International Standards Organisation (ISO) recommendations. This evaluation program should include analysis of variance (ANOVA) treatment to establish uniformity of the measured property throughout the entire batch.

  • Reputation

    A reference material is no better than the user’s perception of it. Therefore, it is critical, that the user has total confidence in its quality. If this is not the case and analytical problems are suspected, the task of assigning the source of error to the suspect laboratory is fraught with uncertainty. It is imperative, therefore, that the CRM producer’s credentials and reputation are unassailable and that the certification documentation is sufficiently comprehensive.

Using CRMs

CRMs are most commonly used in the mining industry to monitor bias in chemical analyses of geological samples. Critical concentrations in mining operations are cutoff and head grades and CRMs are generally selected to approximate these grades. CRMs are usually inserted at a frequency of 1 in 20 to 1 in 30 into the sample stream and the results produced by the laboratory are then compared against the certified values. CRM blanks are devoid of the metal(s) of interest and are used to monitor contamination within the laboratory.

Control Limits

No analytical method is 100% accurate and therefore a certain amount of error is tolerated. This margin of error is variously referred to as a window of acceptability, control limit or performance gate. Generally, results lying within two (or sometimes three) standard deviations either side of the certified value are deemed acceptable, although precise application of control limits should be at the discretion of the QC manager concerned.

There are various methods used to determine the standard deviation. These methods are empirically derived and based on an analysis of errors contributing to the spread of results obtained in the round robin certification program. These are laboratory measurement errors and sampling errors. Measurement errors include between-laboratory bias, between-batch bias (reproducibility errors) and within-batch precision (repeatability errors). Sampling errors relate to the level of homogeneity of the CRM and should be negligible in comparison with measurement errors.

Confidence Interval

ISO requires that Certificates of Analysis include a measurement of uncertainty of the certified value. This is generally expressed as a 95% Confidence Interval and should not be confused with Control Limits. Put simply, Control Limits provide an expectation of acceptable laboratory performance while Confidence Intervals provide an estimate of the reliability of the certified value.

Tolerance Interval

This parameter is a measure of homogeneity of the CRM. We have pioneered a method of reduced analytical subsampling for evaluating the homogeneity of gold in CRMs. This involves the analysis of gold by high precision neutron activation analysis (NAA) on analytical subsample weights of 0.5g to 1.5g (compared to 25g to 50g for the fire assay method). By employing a sufficiently reduced subsample weight in a series of determinations by the same method, analytical error becomes negligible when compared with subsampling error. The corresponding standard deviation at a 25g to 50g subsample weight can then be determined from the observed standard deviation of the 0.5g to 1.5g data using the known relationship between the two parameters (Ingamells, C. O. and Switzer, P. (1973), Talanta 20, 547-568). The absolute homogeneity of gold is then determined from tables of factors for two-sided tolerance limits for normal distributions. All OREAS and custom gold CRMs undergo this stringent testing and without exception exhibit a very high level of repeatability consistent with excellent homogeneity.

Glossary of Terms

The following terms are used here and in literature elsewhere to discuss reference materials.

Best Practice – A set of working methods that have been found, through experience and research, to be the best available to use in a particular business or industry. The methods should be described formally and in detail. Mineral assay laboratory competence and documentation is accredited and tested through the National Accreditation Authority (as per ILAC system ISO/IEC 17025). Laboratories in general strive for accreditation for the analytical methods of key commercial importance. For example a laboratory might be 17025 accredited for its iron ore analysis methods but not for platinum analytical methods or may be accredited for Au analysis by Pb collection fire assay with ICP finish at grades between 5 and 10 ppm but not for grades between 0.2 and 1.0 ppm.

Over and above the ISO accreditation system and relevant to mineral laboratories are the Best Practice Laboratory QA/QC systems developed for pathology laboratories through the work of Dr James O. Westgard. “Westgard Rules”. These are multi-rule QC rules that use Control Samples (CRM's) to help analyse whether an analytical run is in-control or out-of-control and should be used by both the laboratory and the customer to monitor that the laboratory’s “Best Practice” procedures are actually working.

Ideally, in the mining and exploration industries, a cost effective "Best Practice" sampling and assay program should at least be recording the following statistics for its internal Quality Management Program:

  1. The laboratory bias (for the particular method used). This should be very low for samples close to the mine cut-off grade and for very high grade samples. It can be higher for less revenue sensitive analyses, such as for tailings or for geochem programs.
  2. The relative standard deviation (RSD) or coefficient of variation (CV) for the control sample results. It is important to consider that the single laboratory’s Control Sample RSD should be lower than the RSD from inter-laboratory testing; but the laboratory's Control Limits must still be within the Control Limits derived from the inter-laboratory testing.
  3. The error detection rate (the critical systematic error). How many of the laboratory errors identified by Control Samples turn out to be real errors upon careful investigation; and not false alarms.
  4. The false error detection rate. How many of the laboratory errors identified turn out to be false alarms.

Bias – How far the measured value lies from the true (or reference) value it is estimating. Bias can also be known as systematic error (an estimate of the level of trueness).