analytical services and qa-qc

24
Analytical Services and QA/QC Lynda Bloom, Analytical Solutions Ltd. Prepared for the Society of Exploration Geologists, April, 2002 1.0 Introduction The principal objective of exploration geochemical surveys is to locate mineral deposits at the lowest possible cost. It is necessary to have reliable survey results so that areas with no apparent geochemical response can be abandoned with confidence. Misinterpretation, sampling inconsistencies or poor quality analytical data can lead to expenditures on areas with “false anomalies”, which is a waste of time and resources. The focus of this paper is the control and improvement of analytical data quality for exploration geochemical surveys. Many types of geochemical surveys (hydrogeochemistry, biogeochemistry, selective extractions and rare earth lithogeochemistry) take advantage of modern technology, such as ICP-MS, and require measurement of elements at the sub-ppb level. These measurements may test the limits of the technology, and contamination is a serious concern. In order to interpret these analytical data, it is necessary to have an understanding of the errors associated with sample handling, preparation and laboratory procedures. Once it is recognized that all data have an associated error, quality assurance can be implemented to measure these errors. Figure 1 is a map of bromine values in lake sediments from an Ontario Geological Survey of the Herman Lake map sheet, covering an area of 40 km by 20 km. The east half of the sheet was sampled at a different time than the west half and the samples were analyzed approximately one year apart. The baseline bromine values are different in the east and west halves of the survey area due to changes in instrumentation, although the same analytical method was requested for both parts of the survey. This example demonstrates the importance of understanding the analytical processes, error measurement and monitoring results. Figure 1: Herman L ake Sheet Bromine Values

Upload: jose-garcia

Post on 16-Sep-2015

8 views

Category:

Documents


4 download

DESCRIPTION

QA

TRANSCRIPT

  • Analytical Services and QA/QC

    Lynda Bloom, Analytical Solutions Ltd. Prepared for the Society of Exploration Geologists, April, 2002

    1.0 Introduction The principal objective of exploration geochemical surveys is to locate mineral deposits at the lowest possible cost. It is necessary to have reliable survey results so that areas with no apparent geochemical response can be abandoned with confidence. Misinterpretation, sampling inconsistencies or poor quality analytical data can lead to expenditures on areas with false anomalies, which is a waste of time and resources. The focus of this paper is the control and improvement of analytical data quality for exploration geochemical surveys. Many types of geochemical surveys (hydrogeochemistry, biogeochemistry, selective extractions and rare earth lithogeochemistry) take advantage of modern technology, such as ICP-MS, and require measurement of elements at the sub-ppb level. These measurements may test the limits of the technology, and contamination is a serious concern. In order to interpret these analytical data, it is necessary to have an understanding of the errors associated with sample handling, preparation and laboratory procedures. Once it is recognized that all data have an associated error, quality assurance can be implemented to measure these errors.

    Figure 1 is a map of bromine values in lake sediments from an Ontario Geological Survey of the Herman Lake map sheet, covering an area of 40 km by 20 km. The east half of the sheet was sampled at a different time than the west half and the samples were analyzed approximately one year apart. The baseline bromine values are different in the east and west

    halves of the survey area due to changes in instrumentation, although the same analytical method was requested for both parts of the survey. This example demonstrates the importance of understanding the analytical processes, error measurement and monitoring results.

    Figure 1: Herman Lake Sheet Bromine Values

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    2

    2.0 Definitions Quality assurance has a broad definition outside the mining industry and has been defined as: All those planned or systematic actions necessary to provide adequate confidence that a product or service will satisfy given needs (Kirschling, 1991). Quality control is one aspect of quality assurance. The difference between the two concepts is described by Vaughn (1990), as Assurance in the quality context is the relief of concern about the quality of a product. Sampling plans and auditsthe quality control devicesare designed to supply part of this assurance. The terms commonly used to discuss geochemical data are defined below. Precision: the reproducibility of a result. The results can be said to be of low precision when multiple analyses of the same sample or duplicate analyses of single samples show a wide variation in results. Accuracy: the relationship between the expected result (particularly of standards) and the result actually achieved from the analysis. Bias: the amount by which the analysis varies from the correct result. The amount of bias can only really be determined by a large number of repeat analyses of known standards over a period of time. A demonstration of the difference between accuracy and precision is provided in Table 1.

    Table 1 Accuracy vs. Precision

    Precise but Inaccurate

    Accurate but Imprecise

    Accurate and Precise

    Value 1 Value 2 Value 3 Value 4 Value 5 Value 6

    130 135 130 125 125 135

    150 95 80 105 95 105

    95 100 100 105 95 105

    Average

    130

    100

    100

    Range

    5

    50

    5

    Expected Value

    100

    100

    100

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    3

    Detection Limit: refers to the limit to which an analytical signal can be measured and distinguished from the background noise with a specific level of confidence in the observation. The term is probably the most widely abused and least useful of any in laboratory literature. Detection limits, used by instrument manufacturers, are often recorded by analyzing single element samples in pure, simple matrices. These figures do not relate to analysis of samples and are not useful guidelines for geologists. There are a number of ways of determining the detection limit of a particular analytical process, some of which are illustrated below, but they should all indicate one thing, that is, the lowest value that can reliably be determined. The most common definition of detection limit is the lowest concentration of an element that can be detected with a 95% probability. This is calculated by taking a large number of instrumental readings at or near the blank level, calculating the standard deviation and determining the value corresponding to 2 standard deviations plus the blank value. For atomic absorption spectrometry, the detection limit is essentially a fixed value for any particular element and once the instrument has been optimized, should not change. However, variations may occur over time with power fluctuations, temperature changes or electrical problems. For some techniques, such as x-ray fluorescence spectrometry or neutron activation analysis, detection limits for an element can generally be improved by increasing the counting time. As a rule, squaring the counting time will halve the detection limit. For example, if the counting increases from 5 minutes to 25 minutes, the detection limit should improve from 10 ppm to 5 ppm. Improved detection limits and therefore longer counting times result in higher per sample costs. Detection limits may not be improved in cases where there are significant interferences from other elements. The detection limit of a particular process has a significant effect upon the precision (and accuracy) of analytical results at levels that approach the detection limit. It is generally assumed that the precision of values between the detection and 10 times the detection limit is 50%. It is important to choose an analytical method with a detection limit that is lower than the expected geochemical background. Preferably, the detection limit should be at most 1/10th the geochemical background of the area. Sensitivity: is often used to refer to the detection limit of the technique but there is only a secondary relationship between the two terms. The sensitivity of a technique relates to the slope of the calibration graph; i.e., it is the slope of the relationship between instrumental signal and analytical concentration.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    4

    As a rule of thumb, the higher the sensitivity, the lower the detection limit, and vice versa. However, some instruments allow the operator to increase the sensitivity by varying amplification factors. While this increases the slope of the calibration curve, it does not improve detection limits as it also increases noise. Drift: occurs when the instrumental response to a given concentration of analyte or to background conditions varies progressively with time. This change need not be linear or even progressive, just observable. It is often associated with changes in ambient temperature or warming up of the electronic components of the instrument during operation. Progressive blocking of nebulizers or atomization sources are a physical cause of drift. If this is not recognized, the results will be biased. Internal standards and normalization standards can be used to minimize the effects of drift. Noise: refers to short-term instability of instrumentation and consequently, of the signal. When variation in atomization and nebulization conditions, sources, or the components themselves gives rise to signal variability, this is referred to as noise. The noise of the instrument is a major factor in lowering detection limits, as the noisier the instrument, the higher the detection limit. This parameter can be used to monitor degradation of instruments or components. Analytical range: of a technique or method is the concentration range over which valid data can be collected within pre-determined statistical parameters. Analytical range is affected by such instrumental factors as integration time, pre-concentration factors, and solution matrix (changes of matrix, especially, will vary analytical ranges by orders of magnitude). Analytical range and signal integration time are inter-related and longer integration times usually extend the analytical range to lower levels. Sensitivity decreases with increasing concentration for most spectrophotometric methods. It is necessary to dilute sample solutions or re-calibrate equipment when concentrations are greater than the upper limit of detection. Dilution of samples in production laboratories is generally imprecise and data are less likely to be reproducible towards the upper limit of detection. It is recommended that high grade samples be resubmitted for analysis by assay techniques that are suitable for the concentration range. Certified Reference Materials (CRM): are homogeneous materials that have been analyzed by a large number of different analysts, usually internationally and using a wide variety of analytical techniques, to provide a representative concentration value for specific analytes. The expected, acceptable or working values are almost always quoted as total element concentrations. There are very few CRMs for the partial

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    5

    extractions or digestions commonly used in the mining industry, such as the ubiquitous aqua regia digestion. Reference (and certified reference) materials should have: n the same matrix as the samples to be analyzed n the same levels of trace elements n the same speciation (valency and binding) as in the sample matrix or similar

    mineralogy. They should also be homogeneous (i.e., the difference between representative sample measurements must be smaller than the overall uncertainty limits of the measurements). Note that homogeneity for one analyte does not imply that the material is also satisfactory for a wide range of elements. CRMs should be physically and chemically stable for an indefinite period of time, which is particularly problematic for sulphide-rich samples that could oxidize. CRMs that are not stable should have an expiration date. The various applications of CRMs can be summarized as: n calibration of equipment n achievement of traceability of calibration n improvement of measurement quality n verification of accuracy of results. They can also be used in statistical quality control procedures, although this option is expensive because large quantities of material are required. Control samples must have some of the characteristics of a reference material.

    Secondary standards: are in-house standards usually used for quality control purposes. The accurate quantification of analytes is of relatively minor importance as long as the same result is obtained on a day-to-day basis. This type of material is considerably cheaper than CRMs and can therefore be used far more often. It is also more than likely that the secondary standard is more representative of the material being analyzed by the laboratory than the CRM and therefore is more useful. Calibration standards: are appropriate standards, usually made from spectrographically pure chemicals, prepared in such a way as to be used to directly calibrate instrumentation being used for analysis. Interference: is the effect of constituents in the sample directly upon analytes and/or upon a measured parameter, causing a bias in results when the latter are compared to equivalent results from samples not containing those constituents. Interference effects can lead to either an increase or a decrease in the measured parameter. However, in some extreme cases they can eliminate the acquisition of any valid data.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    6

    Reagent Blank: refers to the concentration of the elements of interest in the sample solution that has been taken through the same analytical procedure as the samples being analyzed, without the sample being added. This measures the potential contamination contributed by reagents. The analytical blank is the above plus the signal component attributable to the instrument noise. Errors in analytical data: All analytical data is subject to errors or to bias. Errors can be divided into the following categories: n Random Errors : This type of error is endemic to analytical chemistry and is part

    of the functioning of every instrument and technique ever developed. It arises from unstable power supplies, non-reproducible atomization and other fluctuations. While it is not possible to remove random error, it is possible to minimize its magnitude and measure its degree.

    n Systematic Errors : Such errors can arise from solution matrix effects specific to

    a particular technique, instrument specific inter-element effects or relative bias associated with variations between sample type and available appropriate standards. Personal bias can also be a source of systematic error, especially when there is a preconceived idea as to the required concentrations in CRMs and in-house standards run with sample batches.

    One of the few checks to determine systematic error is the use of standards or controls; however, it is advisable not to acquaint the analyst with either the location of these samples in the batch or the expected concentration of analytes in them. In this way, bias can be removed. This approach has a fundamental advantage, in that wrongly certified values soon become known.

    n Gross Error: These errors result in completely incorrect results being obtained.

    They are the result of such things as mislabelling of samples, incorrect preparation procedures, vessels contamination, incorrect instrument set up, bad calculations, etc.

    Such errors are usually random, can be identified, and are corrected quickly with the use of quality control procedures such as the submission of blanks, duplicates and controls with samples.

    3.0 Six Sigma Approach or DMAIC Michael Thompson, one of the co-designers of the Thompson-Howarth precision plot, said, "All analytical measurements are wrong; it's just a matter of how large the errors are, and whether they are acceptable." In this instance, "errors" refers to the inaccuracies and imprecision of the data. These are not "mistakes" but result from the naturally occurring limitations of selecting small representative samples from large volumes of material and from the sensitivity of analytical methods.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    7

    There are many sources of error to be taken into account when assessing a geochemical or assay database. These may include sample inhomogeneity, contamination, data accuracy and analytical precision. Each project will have different challenges so that the program to measure these errors will vary in its design. Six Sigma is an organizational quality system that can be applied to geochemical data. The Six Sigma approach is a system that was originally implemented at General Electric and has been adopted by large corporations to save millions of dollars. Application of the Six Sigma approach system in this context is suitable to maximize the effectiveness of quality control programs and ensure that the appropriate measures are adopted. The key steps of a Six Sigma improvement project are referred to as DMAIC

    Define Measure Analyze Improve Control

    3.1 Define Although it is important to measure all sources of errors, there are financial and practical constraints. The first step of the process is to define what has to measured and how often, which is usually done in conjunction with understanding the consequences of introducing errors. Different types of projects have different requirements. For example, it is important to measure the accuracy of Ti and Zr analyses, for lithogeochemical surveys, since subtle variations in these data will be used to identify rock types and degree of alteration. A fluctuation in the accuracy of the data between batches is unacceptable. A rigorous system of reference materials is required. Similarly, the accuracy of low-level fire assay determination of gold in soils may be important. Figure 2 is an example of a soil survey in Africa in a lateritic terrain over an area of approximately 6 km by 6 km. Gold values range from detection limit (5 ppb) to several grams per tonne. The approximately 5,000 soils were sampled at two different times, with Phase 2 sampling, in the southeast corner of the project area, completed almost a year later than Phase 1. The background Au values in Phase 2 are elevated relative to the background values in Phase 1. The selection of anomalous areas is biased because of the shift in background values. It is most likely that the problem arose due to a shift in the accuracy of the fire assay determinations for Au. A system of reference materials, blanks and field duplicates would have identified the shift in background Au values and the laboratory could have been requested to repeat the determinations.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    8

    Selective extractions (such as MMI, Enzyme Leach, cold hydroxylamine, sodium pyrophosphate and others) are generally expected to have high signal-to-noise ratios. The ratio of threshold to anomalous values may be a factor of ten or higher. In this case, the precision of each determination is not as important as for other types of surveys. It is necessary to monitor the accuracy of the determinations, especially whether the extractions were performed consistently and if the chemistry of the sample affected metal extractability. A system of reference materials is required but crosscheck analyses may not be useful. Due to the low elemental concentrations, sources of contamination need to be controlled and monitored. In some surveys, it is important to monitor gross errors more closely than systematic errors. Gross errors include switching samples during analysis, samples numbered incorrectly in the field, and other randomly introduced human errors. Regional stream

    Figure 2: Distribution of Au in soils over a area of 6 km by 6 km.

    Phase 1

    Phase 2

    Break between Phases 1 & 2

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    9

    sediment surveys may be designed with the collection of one sample per 10 square kilometres or more. If two consecutive samples are mixed up and only one is anomalous, then follow-up work is concentrated in the wrong area. Not only is the cost of the follow-up work lost but the prospective ground is never tested. Gross errors in regional surveys is a serious problem and requires a quality control program to address it. Many of the most rigorous quality control programs are designed for advanced drill programs. In an unusual case, Cu values were reported by two different laboratories on the same pulp for almost 10,000 samples using acid digestion with instrumental finishes. Most of the Cu values reported within 10% except for a group of 115 samples with differences of over 50% between values (Figure 3).

    0

    0.5

    1

    1.5

    2

    2.5

    0 0.2 0.4 0.6 0.8 1 1.2

    Cu (%) Lab 1

    Cu

    (%

    ) L

    ab 2

    +10%

    -10%

    The differences are attributable to gross errors in almost every case. In most cases, it appears that samples were switched when they were weighed, at either of the two laboratories. One of the two laboratories is highly computerized so it is unlikely that errors were made in data transfer or data transcription. The other laboratory is less computerized and some errors are related to data entry. The error rate is in the order of 0.1% for this comparison of Cu values on the same pulp. There are relatively few processes involved in these determinations, relative to fire assay for example. Samples were weighed, acids added, the test tube racks were presented to the atomic absorption spectrometer or ICP, and data were collected and transferred to reports. For methods that require greater sample handling, such as fire assay, higher rates of gross errors are expected.

    Figure 3: Comparison of Cu determinations by Labs 1 and 2. The majority of the 10,000 data points fall between the 10% lines and are not shown on this graph.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    10

    Gross errors that could be introduced in sample drying, crushing and pulverizing were not measured when comparing the Cu determinations on the same pulp. There are many steps in these processes where gross errors can be introduced. Where the errors are due to two consecutively numbered samples being switched, there is essentially no impact on ore reserve calculations since the samples are most likely from the same mineralized zone. However, if the samples were from a regional survey the consequences might be more severe. This example demonstrates that a quality control program may need to measure reproducibility (i.e., precision) and the number of defects or gross errors in the data set. One other design criteria is compliance with regulations. Some jurisdictions and corporations have requirements to report quality control measures. These are not technical considerations but need to be incorporated into the quality control program. At the Define stage of the process, the key items to be monitored are determined and the risks associated with errors are recognized. The next step is to identify what needs to be Measured. 3.2 Measure To determine what needs to be measured, it is necessary to have a thorough understanding of the processes. At each stage of sample preparation, weighing, digestion, etc., there is the potential for errors. In the Define stage, it was determined which criteria were important. Knowing, for example, that it is critical to monitor accuracy, the flow sheet in Figure 4 can be examined for processes where inaccuracy could be introduced. Once these processes are identified, it is possible to take measurements that will identify when the process is out of control. Flow charts of all processes, from sampling to data acquisition, can be developed for each project and evaluated to identify sources of contamination, inaccuracy and sample inhomogeneity. The quality control program is custom designed for a project. The level of confidence in the laboratory, anticipated grades, distribution of the mineralization and other factors determine the approach selected.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    11

    Analytical Solutions Ltd.

    Flow chart: insertion of blanks, standards and dups

    3" dia. core/0.5 m sample/~1.5 kgSplit core in half 1 in 25 cases sampleArchive remaining core

    second half of the coreand treat as first half

    Dry at 110 degrees C for 24 hoursInsert coarse blanks (1/50) Prepare coarse blank

    Crush to 80% -2mmClean crusherSplit 500 gm

    Archive - 2mm reject Insert standards Purchase/prepare std.

    In 1 in 25 cases, Pulverize to -88 micronssubmit 500 g split Split 100 gof -2 mm material

    400 g 100 g

    In 1 in 25 cases, Randomly selectrenumber and submit Submit for chemical analysis 1/25 for submissionfor duplicate analyses to secondary lab

    LegendReview QC

    Monitor Contamination Merge analytical data with sampleIntroduction of Sampling Error location and description in theMonitor Sampling Error databaseMonitor Accuracy

    Drill Core Sampling, Sample Preparation and Analysis

    Figure 4: Flow sheet for drill core sampling, sample preparation and analysis

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    12

    Some basic concepts such as the insertion of blanks and field duplicates are described below for a typical regional geochemical survey. These descriptions assume that samples are submitted to the same laboratory for preparation and analysis. Some mining companies are using one laboratory for sample preparation, inserting quality control samples and then submitting all the pulps to another laboratory for analysis. This approach has the advantage that blanks, standards and duplicates can be submitted blind. When this approach is used some of the issues discussed below are not relevant. Blanks: It is recommended that a non-mineralized material be inserted on a routine basis. This material is not crushed or pulverized. Higher-than-expected analytical results would indicate that the material was contaminated during sample preparation, during laboratory analysis or perhaps replaced with a mineralized sample. Incorrect results would then mean that a batch of samples would have to be re-analyzed or changes made to standard procedures. The blank material should closely resemble the material being submitted for analysis. For a stream sediment survey, a bulk sample would be collected from an area where there is no known mineralization. Similarly, a bulk sample would be collected from a pit in a barren area to acquire material that resembles the sampling horizon in a soil survey. Blanks should be submitted so that they are not distinctive from the other samples in the shipment. It is necessary to pre-assign sample numbers for blanks so that these numbers are not used inadvertently in the field. Duplicate samples: Duplicate samples are used to (a) monitor sample batches for potential sample mix-ups and (b) monitor the data variability as a function of both laboratory error and sample homogeneity. Collection of duplicate samples from the same site is a useful means of monitoring both site homogeneity and analytical precision. The degree of reproducibility will determine how sensitive the data are to site variation and therefore improve the interpretation of anomalies. Very poor reproducibility of results could indicate that sample mix-ups have occurred in the laboratory. Field duplicates are generated by collecting a sample twice from the same site using the same procedure each time. The duplicate samples should not be labelled with consecutive numbers. The sample numbers should be separated by at least 20 sample numbers so that the two samples will be analyzed in different laboratory batches. The duplicate sample can be labelled with a random number from the sample tag book; the duplicate number is recorded at the same location as the original sample. Alternatively, field crews can be issued with a separate sample tag book with a different series of numbers, which is specifically used for duplicate samples.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    13

    After approximately 25 duplicate pairs have been generated, the data should be reviewed to determine if the data precision is acceptable. It may be necessary to increase the size of the sample collected or alter laboratory procedures if sample inhomogeneity is suspected to be a problem. If the laboratory is suspected of numerous errors, it may be necessary to select a different laboratory for the project. Duplicate samples should also be generated at each stage where a sample is split. Sample preparation duplicates are generated by pulverizing two splits of the crushed sample. If there is a two-stage crushing procedure, preparation duplicates should be generated for each step of the procedure. Laboratories routinely analyse pulps in duplicate and this data can be requested from the laboratory. It is also recommended that the second half of split drill core be routinely sampled and assayed to determine the variability in two halves of the same core. However, many companies are reluctant to utilize both halves of the drill core and this is usually a management decision. Control samples: For most regional geochemical surveys, blank samples and field duplicates are included routinely. In some cases, it is also preferable to insert control samples on a routine basis. Drilling campaigns will also include the use of control samples as the accuracy of the assays must be documented for ore reserve calculations. Control samples are a preferred method for monitoring the consistency of a laboratory. Usually a homogeneous, fine grained pulp is submitted routinely to a laboratory for analysis. Standard reference materials are available from various government institutions (CANMET in Canada, National Bureau of Standards in the U.S.) but are generally too expensive to be used on a routine basis. More commonly, 5- to-10 kg pulps are prepared from material at the project site, covering the range of expected values. No more than seven control samples are necessary but it is important to prepare an adequate volume of material. It is useful to have a series of control samples to cover the range of anticipated values and also so that it is more difficult for the laboratory to anticipate the correct value. However, if too many different control samples are introduced it is difficult to accumulate the necessary statistics on data variability and it is more likely that mistakes will occur when recording which control samples were inserted. It is common to submit material that is considered suitable for control samples to five or six laboratories to determine an accurate value for each material. Multiple aliquots of the control samples should be submitted along with purchased certified reference materials. There are also a number of commercially available control standards that are available at costs in the order of $50-100 per kilogram or 5% of the cost of certified reference materials. There are a wide variety of materials at different grades and styles of mineralization. These standards are useful for short drill programs where project-derived controls are not available. They can also be effective where there are concerns that a

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    14

    laboratory is familiar with the standards being submitted and is optimizing the results reported. Randomization: Some companies have implemented a practice of renumbering all samples using a random numbering system (Plant, 1973). This approach will identify laboratory drift and bias more readily than submission of samples that were collected along sample lines or throughout a drill hole, and numbered consecutively. Some geologists are reluctant to use this technique as errors may occur in reassigning the correct sample numbers and locations. Crosschecks or umpire assays: Many companies have adopted a practice of resubmitting 5-10% of all sample pulps for analysis to a second laboratory. This approach identifies variations in analytical procedures between laboratories, possible sample mix-ups, and whether substantial biases have been introduced during the course of the project. These are routine checking programs and are far superior to the more common practice of submitting a selection of samples at the end of a project to an alternative laboratory for analysis. Unfortunately, if a problem is identified at the end of a project, the decisions based on the assays such as drill hole locations, anomaly follow-up, etc., may already have been made and the budget spent. Selection bias can be introduced if check samples are not selected randomly from the entire analytical range (Long, 1999). Standards should always be included with the submission of check samples so that if a bias is identified between the results of two laboratories, it can be determined which laboratory produced the correct assays. There is a difference between submitting pulps for check assays and submitting rejects. Rejects refers to a second split of the crusher product (usually 90% passing 2 mm). When rejects are submitted it is difficult to discern subtle analytical biases as the sampling errors are likely greater than the potential analytical errors. Submitting rejects is a worthwhile test of the splitting procedures at a laboratory but not a good test of analytical accuracy. Laboratory Communication: An important feature of project management is to maintain close communications with the laboratory being used for analyses. Questions should be asked concerning the type of sample preparation equipment, cleaning methods and standard operating procedures. It is possible to request documentation of the analytical procedures from a laboratory and incorporate the information in an appendix of a report. Documentation of the procedures is particularly important for a long-term project where procedural changes from year to year may significantly bias the database of information. It is always recommended to include clear and precise analytical instructions with every sample batch submitted. Where possible, a laboratory tour should be arranged. It is becoming more common to sign laboratory contracts that specify analytical methods, price, turnaround time and quality control expectations. A contract can clarify when a laboratory will repeat assays free-of-charge if there are quality control failures.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    15

    In summary, a variety of quality control measures can be used to monitor data quality. Some of the approaches used to monitor data quality include: n insertion of control samples n insertion of international reference materials n submission of field duplicates n submission of sample preparation duplicates (approximately 90% passing 2 mm) n randomization of sample numbers before submission to a laboratory n comparison of multi-element trends for elements determined by different

    laboratory procedures n comparison of results for the same element determined by different methods n routine insertion of an unprepared, barren sample (blank) n routine insertion of a pulverized barren sample (blank) n analysis of 5-10% of sample pulps at an umpire laboratory n analysis of 5-10% of sample rejects at an umpire laboratory. 3.3 Analyze Overall quality of the data can be improved by quickly identifying and remedying the problems. Results for control samples and blanks should be reviewed as soon as every laboratory certificate is received. Control charts are used to monitor the data and decide immediately whether the results are acceptable. 3.3.1 Control Charts Each time the laboratory reports a value for one of the control samples, including the blank, the value is plotted on a control chart. These charts can be computer plotted but it is just as useful to plot the results by hand on a piece of graph paper. The graph paper is prepared with the mean, mean plus two standard deviations and the mean minus two standard deviations drawn as lines across the chart. The mean plus two standard deviations is the Upper Control Limit and the mean minus two standard deviations is the Lower Control Limit. The mean and standard deviation are derived from the multiple analyses performed at several laboratories to establish acceptable values for the control samples. It is important to understand the statistics associated with the estimation of the accepted value in order to evaluate the results. The mean of the element concentrations for a CRM are derived from the multiple analyses performed at numerous laboratories to establish acceptable (or expected) values for control samples. The determination of the mean is complicated when there are apparent outliers or sets of data from specific laboratories that appear to be biased. The International Organization of Standardization (ISO) recommends that outliers should not be excluded on purely statistical evidence until they have been thoroughly investigated and , where possible, reasons for the discrepancies identified. A variety of statistical tests for outliers exist (Verma, 1997) and are applied by the supplier of the CRM.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    16

    The establishment of an appropriate range of acceptable values is more difficult. Gold standard MA-1b, produced by the Canadian Certified Reference Materials Project, a division of Natural Resources Canada, is used to demonstrate this point. The Certificate of Analysis for MA-1b reads as follows

    REFERENCE GOLD ORE MA-1b

    This is the label on the purchased bottles and many purchasers assume that 95 out of 100 times a laboratorys results for MA-1b should therefore fall between

    16.7 to 17.3 g/g Au. However, it is stipulated in the literature that accompanies the bottle, The uncertainty estimates the expected range of reproducibility of this mean within 95% probability were the measurement program to be repeated many times. In fact, the 95% Confidence Limit quoted denotes that if the certification program were to be conducted 100 times, the overall mean in 95 cases would be expected to fall within the prescribed limits. The certification program for MA-1b included 175 acceptable analytical determinations by 28 laboratories. When certified reference materials or standards are inserted with samples there is only one determination and the 95% Confidence Limit quoted is not applicable for measuring the acceptability of the reported value. Along with the MA-1b documentation is an additional table of statistics that is reproduced below.

    Distribution of results by method

    g/g Method No. of Sets

    No. Results Mean CI SLc Src CV,%

    FA/G FA/AAS INAA FA/INAA FA/ICP Overall

    20 8 3 1 1 33

    113 44 20 4 5 186

    16.96 17.26 17.21 16.23 17.36 17.05

    0.30 0.85 1.85 0.26

    0.61 0.99 0.66 0.70

    0.37 0.42 0.65 0.36 0.29 0.42

    1.93 2.30 3.88 2.25 1.66 2.22

    Recommended Value 95% Confidence Interval

    Au 17.0 g/g 0.3 g/g

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    17

    SLc is the between-set standard deviation and Src is the within-set standard deviation. The bottles of MA-1b are labelled with the overall mean, 17.0 and the 95% Confidence Limit, 0.30 g/g. The more meaningful statistics for explorationists, when a single determination by a single laboratory is being evaluated, are the overall between- and within-set standard deviations. The within-set standard deviation is a reflection of the homogeneity of the material in the bottle received by the participating laboratory in combination with that laboratorys ability to reproduce the analytical method routinely. The between-set standard deviation is likely the most useful statistic as this takes into account slight biases between laboratories, the differences between the sub-samples received by the laboratories, in addition to the factors described for the within-set standard deviation. The mean two standard deviations approximates the expected range of values for 95% of the cases. Using SLc , the between-set standard deviation, of 0.70 g/g and the calculated mean value of 17.0 g/g , the results for MA-1b are expected to fall within 15.6 to 18.4 g/g Au. Based on the 95% Confidence Limit as indicated on the label of the bottle, the range of results for MA-1b are 16.7 to 17.3 g/g Au which is a considerably narrower range of acceptable values. The range established by using SLc , the between-set standard deviation, is equivalent to 8.2% of the certified value of 17.0 g/t Au. If the user must demonstrate that the method gives an accuracy to an uncertainty better than 8%, she should select a different certified gold ore which has less between-set variability. On the other hand, if the users method gives no better than ~ 8%, MA-1b is a suitable reference material (Steger, 1998). This is a very complicated subject and using SLc , the between-set standard deviation, to estimate the allowable range of values is only a first approximation. The International Standards Organization has several committees that have been working on these questions for 25 years which are covered by ISO Guides 30 to 35. There are several statistical approaches documented to evaluating the acceptance of single and replicate assays, which seem to involve equations (as shown below) where different sets of conditions need to be met.

    | XC - XL | 2 sLm2 + sRm2 Further details on the evaluation of results for certified reference materials are described in a condensed version of ISO Guide 33 available from the Canadian Certified Reference Materials Project at CANMET, Ottawa, Canada, [email protected] . This does not take into account the fact that 5% of the cases will be outside this acceptable range, based on the definition of the mean plus or minus two standard deviations.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    18

    Many suppliers of reference materials report the error as the 95% Confidence Limit or some other statistic to demonstrate that the materials are homogeneous and the round robin was conducted properly. These values should not be used to decide on which assays should be rejected for a quality control program. It may also be necessary to consider whether the analytical methods being used are equivalent to those used for the establishment of acceptable values. When a number of different reference materials are used for the same project, it may be difficult to interpret trends based on control charts for individual standards. The results for any number of standards can be compiled on one graph by plotting the reported value as a percentage of the expected value (Figure 5). Chemists more often plot Z-scores against time, where the Z-score is the reported value less the expected value then divided by the standard deviation. Division by the standard deviation provides additional insight as to whether the differences are significant.

    Analytical Solutions Ltd.

    Compilation of Results for 22 Standards Ranging in Gold Grade from 0.7 to 11 g/t

    75

    80

    85

    90

    95

    100

    105

    110

    115

    120

    125

    Time Sequence by Batch and Sample Number

    Au

    Ass

    ay a

    s %

    of

    Exp

    ecte

    d V

    alu

    e fo

    r S

    tan

    dar

    ds

    Batch 752Batch 761

    Moving Average

    N=307

    When unacceptable values are found, it is appropriate to contact the laboratory and request additional analyses. It is not important to get the correct value for the control sample but this information is used to ensure that the samples analyzed in the same batch have been reported properly. As a guideline, if a control sample is inserted with every 20 samples then the 10 samples before the out-of-control sample and 10 samples after it should be requested for re-analysis. If there are other reasons to suspect the results in a

    Figure 5: Compilation of Results for Multiple Controls for the Same Project

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    19

    laboratory batch, for example results for the duplicates, then additional tests may be required. Control charts should be prepared for each control sample or blank and for each element. In the case where a 30+ element ICP scan has been requested, the preparation of these graphs becomes an onerous task. It is acceptable to plot a selection of elements but it is important that at least several elements are monitored for each analytical method. 3.3.2 Plotting Duplicates It is also necessary to review the data for various types of duplicates. The preparation of simple X-Y plots of the two results using a spreadsheet program is the easiest way to do this. The same scale is selected for the X- and Y- axes; where there is a broad range of values it may be preferable to use a logarithmic scale for the axes. All values should plot close to the X = Y line and precision envelopes can be drawn so that points that fall outside these envelopes are automatically recorded as unacceptable (Figure 6).

    1

    10

    100

    1000

    10000

    100000

    1 10 100 1000 10000 100000

    Cu 1 (ppm)

    N=488

    X=Y+10%-10%

    Alternatively, Howarth-Thompson plots (Thompson, 1992) are used when there is a broad range of values to evaluate and in order to calculate precision. These plots have the advantage of graphically displaying the difference between values so that both the absolute difference between values as well as the percentage difference can be easily determined. The graph is constructed by (i) calculating the average of the two results

    Figure 6: Precision Envelopes

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    20

    (ii) calculating the difference between the two results (iii) converting the differences into absolute values (iii) plotting the average against the difference on log-log axes of an X-Y plot. Variations of this graph can be used to compare data where it is important to monitor whether there is a positive or negative bias between two sets of results, for example, the results from two laboratories. In this case, the difference between two results is plotted against the average of two results and arithmetic axes are used (Figure 7). The same data set is displayed in Figures 6 and 7 but it is easier to visualize the differences in Figure 7.

    - 2 0 0 0

    - 1 5 0 0

    - 1 0 0 0

    - 5 0 0

    0

    5 0 0

    1000

    1500

    2 0 0 0

    0 5 0 0 0 10000 15000 2 0 0 0 0 2 5 0 0 0 3 0 0 0 0

    Mean Cu (ppm) Lab Dups

    N=480

    X=Y

    -5%

    -10%

    +5%

    +10%

    A comparison of results for duplicate samples or crosschecks will identify laboratory problems and also provide an estimate of sample homogeneity. In certain cases it may be proven that the laboratory is providing reliable data but that field sampling programs and/or sample preparation procedures need to be modified in order to improve the precision of the data. 3.4 Improve A quality control program is designed to measure the variation in precision, accuracy, sample representivity and other parameters, as required. The quality control data provide the numerical basis to improve processes and plan for improvements.

    Figure 7: Mean vs. the Difference Plot

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    21

    Necessary action could include a request for re-assaying, cancellation of a laboratory contract or changes to sampling or analytical procedures. For example, it may be necessary to alter sample collection, crushing, splitting and grinding procedures based on the results of quality control data in order to improve sample representivity. The additional cost of using more laborious splitting procedures or pulverizing a larger sample aliquot could be justified based on scientific data. It may be recognized that lower detection limits and therefore different analytical methods and instrumentation are required. The interpretation of geochemical data requires an understanding of the precision and accuracy of the data. In the simplest case, if an anomalous value is determined to be greater than 1000 ppm and the precision of the data is 50%, then any value greater than 500 ppm is technically anomalous. If the precision of the data is 10%, then only values above 900 ppm are anomalous. The need to strictly apply precision limits is alleviated somewhat by evaluating trends in the data and integration with other data sets. However, the use of ratios or other multi-element calculated scores further complicates the issue as errors may be cumulative. The quality control requirements developed at the Define stage of the process should have incorporated the concerns that could arise during interpretation. However, if problems are recognized with the interpretation and integration with other data sets, additional improvements to the sampling and analytical processes may be necessary. 3.5 Control For ongoing control of the data, it may be necessary to refine measurements and investigate alternative methods of displaying the data. A critical concern is that the quality control data are reviewed regularly. It is important to automate as much of the process as possible or use a third-party to evaluate the data and act as an intermediary with the laboratory. Company management needs to take a lead role. The cost of quality control procedures including the additional assays has to be included in the budget. Any project review should start with a brief examination of the quality control measures. A project review should not proceed if it cannot be demonstrated that the data are valid. 4.0 Conclusions The Six Sigma DMAIC approach is an attractive model for the implementation of a quality control program and the evaluation of the data. The approach develops a series of measurements that are used to control data quality and to make improvements to the processes. Improvements to sampling, preparation, analysis and quality control can be implemented on the basis of numerical data and evaluated for their cost-effectiveness.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    22

    SELECTED BIBLIOGRAPHY Amor, S., Bloom, L. and Ward, P., 1998, Practical application of exploration geochemistry. Proceedings of a short course presented by the Prospectors and Developers Association of Canada, Toronto. Bloom. L., 1999, Third party vetting of geochemical programs or return on quality, in Quality Control in Mineral Exploration: A short course presented during the 19th International Geochemical Exploration Symposium, April, 11, 1999. Bloom, L., 1998, The Role of Economic Geologists in Evaluating Assay Data Quality. Proceedings of a short course presented by the GAC and PDAC, November, 1998. Bloom, L., 1993, Man-made parameters in elemental analysis: SME Pre-print 93-79: p.5. Burn, R.G., 1981, Data reliability in ore reserve assessments: Mining Magazine, October, p. 289-299. Clifton, H.E., Hunter, R.E., Swanson, F.J. and Phillips, R.L., 1969, Sample size and meaningful gold analysis: United States Geological Survey, Professional Paper 625-C. Fletcher, W.K., 1981, Quality control in the laboratory, in Govett, G.J.S., ed., Analytical Methods in Geochemical Prospecting: Handbook of Exploration Geochemistry, v.1, p. 25-46, Elsevier. Fletcher, W.K., 1987, Analysis of soil samples, in Fletcher et al., eds., Exploration Geochemistry: Design and Interpretation of Soil Surveys, (1987): Soc. Econ. Geol., p. 79-96. Garrett, R.G., 1969, The determination of sampling and analytical errors in exploration geochemistry: Econ. Geol., v. 64, p. 568-571. Govindaraju, K., 1994, Compilation of working values and sample description of 383 geostandards: Geostandards Newsletter, v.18, p. 1-158. Gy, P., 1976, The sampling of particulate materials - a general theory: Symposium on sampling practices in the mineral industries, The Australian Inst. of Mining and Metallurgy, Victoria, Australia, p. 17-33. Hall, G.E.M., 1996, Twenty-five years in geoanalysis, 1970-1996: Jour. Geochem. Expl., v. 57, Nos. 1-3, 1-8. Hall, G.E.M. and Bonham-Carter, G., 1988, Review of methods to determine gold, platinum and palladium in production-oriented laboratories, with application of a statistical procedure to test for bias: Jour. Geochem. Expl., v. 30, p. 255-286.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    23

    SELECTED BIBLIOGRAPHY (Continued) Hall, G.E.M., Vaive, J.E., Coope, J.A. and Weiland, E.F., 1989, Bias in the analysis of geological materials for gold using current methods: Jour. Geochem. Expl., v. 34, p. 157-171. Hill, W.E., 1974, The use of analytical standards to control assaying projects: Vancouver IGES, p. 651-657. Howarth, R.S. and Thompson, M., 1976, Duplicate analysis in geochemical practice, Part II: Analyst 101 (1206), p. 699-709. Kane, J.S., 1992, Reference samples for use in analytical geochemistry: Their availability, preparation and appropriate use: Jour. Geochem. Expl., v. 44, p. 37-63. Kirschling, G., 1991, Quality Assurance and Tolerances: Springer Verlag, 335 p. Kretz, R., 1985, Calculation and illustration of uncertainty in geochemical analysis: Jour. Geol. Educ., v. 33, p. 40-44. Levinson, A.A., Bradshaw, P.M.D. and Thomson, I., 1987, Discrepancies in analytical determinations of gold, Arizona, U.S.A., in Levinson, A.A., Bradshaw, P.M.D. and Thomson, I., eds., Practical Problems in Exploration Geochemistry: Applied Publishing, p. 148-149. Long, S., 1999, Impact of selection bias, in Quality Control in Mineral Exploration: a short course presented during the 19th International Geochemical Exploration Symposium, April, 1999. Mining Standards Task Force, Toronto Stock Exchange, 1999. Mineral Exploration Best Practices Guidelines, October 1999. Plant, J.A., 1973, Random numbering system for geochemical samples: IMM Trans. 82, B64-B65. Plant, J.A., Jeffrey, K., Gill, E. and Fage, C., 1975, The systematic determination of accuracy and precision in geochemical exploration data: Jour. Geochem. Expl., v. 4(4), p. 467-486. Potts, P.J., Tindle, A.G. and Webb, P.C., 1992, Geochemical reference material compositions: Rocks, minerals, sediments, soils, carbonates, refractories and ore used in research and industry: CRC Press Inc. Pyzdek, T., 1989, What every engineer should know about quality control. ASQC Quality Press, N.Y. 251 p.

  • Analytical Solutions Ltd., 1214-3266 Yonge Street, Toronto, ON M4N 2L6

    www.explorationgeochem.com

    24

    SELECTED BIBLIOGRAPHY (Continued) Ramsey, M.H., Thompson, M. and Hale, M., 1992, Objective evaluation of precision requirements for geochemical analysis using robust analysis of variance: Jour. Geochem. Expl., v. 44, p. 23-36. Stanley, C., 1999, Treatment of geochemical data: Some pitfalls in graphical analysis, in Quality Control in Mineral Exploration: A short course presented during the 19th International Geochemical Exploration Symposium, April 11, 1999. Steger, H.F., 1998. Uses of matrix reference materials. For presentation at the IUPAC/ISO/REMCO workshop on Reference Materials, Berlin, April 22-23, 1999 and the workshop of the conference of the Canadian Mineral Analysts, Kirkland Lake, ON, September 17, 2001, and for publication in the respective workshop proceedings. Project: MMSL No. 600637CCRMP DIVISION REPORT MMSL 98-024 (OP&J) Thompson, M., 1983, Control procedures in geochemical analysis, in Howarth, R.J., ed., Statistics and Data Analysis in Geochemical Prospecting: Handbook of Exploration Geochemistry, v.2, p. 39-58, Elsevier. Thompson, M., 1992, Data quality in applied geochemistry: the requirements and how to achieve them: Jour. Geochem. Expl., v. 44, p. 3-22. Thompson, M. and Howarth, R.J., 1978, A new approach to the estimation of analytical precision: Jour. Geochem. Explor., v. 9, p. 23-30. Vaughn, R.C., 1990. Quality Assurance. Iowa State University Press, Ames, Iowa. Verma, S.P., 1997. Sixteen statistical tests for outlier detection and rejection in evaluation of international geochemical reference materials: example of Microgabbro PM-S. Geostandards Newsletter, Vol.21, No.1, pp. 59-75.