1、Designation: E 1329 00 (Reapproved 2005)Standard Practice forVerification and Use of Control Charts in SpectrochemicalAnalysis1This standard is issued under the fixed designation E 1329; the number immediately following the designation indicates the year oforiginal adoption or, in the case of revisi
2、on, the year of last revision. A number in parentheses indicates the year of last reapproval. Asuperscript epsilon (e) indicates an editorial change since the last revision or reapproval.1. Scope1.1 This practice covers procedures for determining if aspectrochemical analysis is under statistical con
3、trol.1.2 Criteria are presented for determining when correctiveaction is required.1.3 Control will be effected by using verifiers to testinstrument response. It is recommended, although not required,that this be accompanied by the plotting of control charts.1.4 The preparation of control charts is d
4、escribed.1.5 LimitationsThe procedures that are described do notapply to analyses that require a calibration each time a set ofanalyses is run. Reference is made specifically to opticalemission spectroscopy, but the practice has a more generalapplication.1.6 This practice does not apply to validatio
5、n proceduresthat monitor the correctness of calibration.2. Referenced Documents2.1 ASTM Standards:2E 135 Terminology Relating to Analytical Chemistry forMetals, Ores, and Related MaterialsE 158 Practice for Fundamental Calculations to ConvertIntensities into Concentrations in Optical Emission Spec-t
6、rochemical Analysis3E 305 Practice for Establishing and Controlling Spectro-chemical Analytical Curves3E 456 Terminology Relating to Quality and StatisticsE 876 Practice for Use of Statistics in the Evaluation ofSpectrometric Data32.2 Other ASTM Documents:MNL 7 Manual on Presentation of Data and Con
7、trol ChartAnalysis43. Terminology3.1 DefinitionsFor definitions of terms used in this prac-tice, refer to Terminologies E 135 and E 456 and PracticeE 876. Refer also to the glossary of terms and symbolsappearing in MNL 7.43.2 Definitions of Terms Specific to This Standard:3.2.1 control limitsin cont
8、rol charts, the upper and lowerlimits of a statistic that are not expected to be exceeded,designated as UCL and LCL respectively in this practice. Forthe statistic that is the average of more than one reading ordetermination, the upper and lower limits will be equidistantfrom a central line (CL) rep
9、resenting the expected average. Forthe statistic of either standard deviation or range, the upperlimit will be farther from the central line if the lower limit iszero.3.2.2 normalizationa procedure for correcting readings toa common basis. A special case of normalization is standard-ization in which
10、 readings are made to conform to an existingcalibration. Normalization permits gathering data in differentperiods of time and correcting for drift in a way that may beindependent of standardization routines.3.2.3 variationdifference in an observed value from anorm.3.2.3.1 assignable causevariation w
11、hich can be identifiedand corrected. It may be the result of a condition of aninstrument or a method of operation. For example, signalintensities may be affected because a spectrometer is notprofiled properly.3.2.3.2 chance or common causerandom variation whichconsistently affects a system, contribu
12、ting to the imprecision ina predictable way. In the application of control charts, theassumption is made that chance causes of variation arenormally distributed.4. Significance and Use4.1 Consistency in analysis depends on being aware of asignificant change in instrumental response, such as that cau
13、sedby drift or changes in analytical precision, or both, and takingcorrective action. The usual corrective action for drift isstandardization. Standardization, however, when there is noreal need, can only broaden the spread of subsequent analyses.1This practice is under the jurisdiction of ASTM Comm
14、ittee E01 on AnalyticalChemistry for Metals, Ores, and Related Materials and is the direct responsibility ofSubcommittee E01.22 on Laboratory Quality.Current edition approved May 1, 2005. Published June 2005. Originallyapproved in 1990. Last previous edition approved in 2000 as E 1329 00.2For refere
15、nced ASTM standards, visit the ASTM website, www.astm.org, orcontact ASTM Customer Service at serviceastm.org. For Annual Book of ASTMStandards volume information, refer to the standards Document Summary page onthe ASTM website.3Withdrawn.4ASTM Manual Series, ASTM, 6th edition, 1990.1Copyright ASTM
16、International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959, United States.One purpose of this practice is to set guidelines that will avoid“ overstandardization.”4.2 To control manufacturing processes, there must beconfidence that a consistent material is being produced and
17、thatthe analysis of the material is reliable. For assurance that thematerial meets specification, a purchaser may require thesupporting record of control charts to assess that properanalytical control has been maintained.4.3 Ideally, variations in analytical results may be held tochance causes. The
18、concept of a confidence interval or limitson a control chart is based on what can be expected when allnormal precautions are exercised. When results appear to goout of control, the analyst should consider and correct whatmight be an assignable cause. As experience is accumulated,however, it may not
19、seem unusual for readings to drift withtime as optics degrade, detector response changes, or excita-tions modify as, for example, when deposits build up on acounter electrode (a correctable assignable cause), or thelonger range effects as an X-ray tube deteriorates.5. Problems in Applying Control5.1
20、 A complication in effecting verification control or inusing control charts with spectrochemical analyses is that themeasurements being taken are not absolute. Determinationsdepend upon comparisons of one measurement to another: therelative intensity of an analytical line to the relative intensity o
21、fan internal standard line in optical emission spectroscopy; theinterrelationship of counts in X-ray spectroscopy under somespecified condition of maintaining a fixed intensity from anirradiating source and holding to a consistent response from adetector with or without pulse height analyzers and wi
22、th orwithout an external monitor; the nonlinear relationship ofemulsion blackening to radiation in photographic measure-ments; and the relative response in integrating for fixed timeswith ostensibly constant radiation sources. Added to these isthe complication of background signal in all techniques.
23、5.2 It is important to recognize that there are several sourcesof random variation, including variations from the measuringmethod as well as inhomogeneity in the specimens. The devicebeing used to test analytical response is the analytical systemitself. This differs from normal statistical process c
24、ontrol wherean independent and usually more accurate measuring device isused to verify the process variability.6. Verifiers6.1 It is recommended that readings for all potential verifi-ers as well as standardants be established by measuring themalong with the calibrants.6.1.1 Ideally, the full set of
25、 potential standardants andverifiers should be run before and after a series of calibrants topermit normalizing all calibration data to a common basis. Toachieve the best normalization of data, readings should berecorded for all elements of interest on every standardant andverifier, even if there is
26、 no knowledge of expected concentra-tions. Unless there is a marked change in the before and aftermeasurements, the averages of a set of before and afterreadings will be used for normalization.NOTE 1If there appears to be a drift between readings of standardantsobtained before and after a set of cal
27、ibrants has been run, an instrumentproblem may have to be investigated and corrected or the operationalenvironment improved. Reliable calibration data can be obtained only ifan instrument shows a stable operation. Practice E 876 describes ways totest for drift.6.1.1.1 Unless a curve fitting routine
28、is being used thatrequires “standardizing” before running a set of referencematerials it is recommended that no normalization be doneuntil all calibration data has been recorded. Strictly speakingstandardization, as defined in Terminology E 135, only can bedone after a calibration has been establish
29、ed. If a normalizationto some prescribed set of readings is done as if it were astandardization before each time a set of reference materials isrun, the resulting record of readings can be treated as if nostandardization had been done.6.1.2 Choose one set of averages of before and afterreadings of 6
30、.1.1 as the norm. A grand overall average of thesets may be used if that seems like a reasonable median of allsets. Exclude any readings for a element in a reference materialthat does not show comparable repeatability to what wasobserved for that element in other materials. For higher levelreadings,
31、 the comparison should be made to observed relativerepeatabilities.6.1.2.1 For an ideal normalization of readings, determinethe regression fit of a set of observed readings, x, to expectedreadings, y. This linear regression, which is also supported byPractice E 305, commonly is done on electronic ca
32、lculators orcomputers by the following equations to determine a slope, m,and a constant, k, which can be used to correct observedreadings to an established norm:m 5n(xy ( x(y!n(x2! (x!2(1)andk 5 (y m(x!/n (2)where the summations of functions of x and y are as fol-lows:x = the observed average readin
33、gs of an element in acalibration set,y = the expected normal readings for that element, andn = the number of pairs of x and y readings.6.1.2.2 Apply the appropriate m and k corrections to theaverages of the verifiers and standardants, as well as to thecalibrants in each calibration set, as follows:R
34、N5 mRO1 k (3)where:RN= normalized reading, andRO= observed average reading.The grand averages of the normalized readings of thestandardants and verifiers will become the values used forstandardizing.6.1.3 If the analytical system only can support the earlyconvention of “two-point” standardization, a
35、nd if the onlypermissible normalization is a quasi-standardization, beforecollecting calibration data it is still advisable to record allreadings for all elements in all reference materials to establisha full record of what can be expected for all the referenceE 1329 00 (2005)2materials (see 8.6). T
36、he initial set of “normal” readings arereasonable starting points. Neither the preferred method ofusing a regression fit nor the recommendation of waiting untilall data have been logged before assigning normal values areinfallible. Modification of these values always should be anoption as more exper
37、ience is gained. It is expected, however,that the preferred methods will arrive at the ideal normal valuesearlier.6.1.3.1 If the operating system is based on two-point stan-dardization, Eq 3 still would be used to normalize or standard-ize readings. The generation of slope and constant corrections,h
38、owever, would be as follows:m 5 HR LR!/HO LO! (4)andk 5 HR mHO! (5)where:HR= Reference or normal reading of the high standardant,LR= Reference or normal reading of the low standardant,HO= Observed reading of the high standardant, andLO= Observed reading of the low standardant.6.1.4 If data are later
39、 transformed by a slope and intercept togive a different scaling for the calibration, the same transfor-mation must be applied to the readings of standardants andverifiers.6.2 If a verifier (or a new standardant) is established after acalibration has been defined, the expected reading can beestablis
40、hed as follows:6.2.1 Shortly after a standardization, run the verifier inreplicate and keep a record of its average reading. Averageabout ten such observations made after new standardizations toobtain a good representation of the expected reading.6.2.1.1 Normalization coefficients are determined by
41、mak-ing a linear regression fit of normal readings as a function ofobserved readings, such as is done in Practice E 305 inestablishing a straight line relationship by the method of leastsquares. The “normal” set of readings can be either overallaverages or a set that appears to be a median of all se
42、ts. The“slope” of this regression becomes the proportional factor, m,and the “intercept” the constant, k.6.2.2 If a verifier has to be established in a shorter time thanthe requirements of 6.2.1, a set of standardants and selectedcalibrants can be run along with the verifier. The data may thenbe ana
43、lyzed as described in 6.2.1.1, with the expected readingsof standardants and calibrants used as the normal readings.This should be repeated at least two more times. Average thecorrected verifier readings to obtain a good estimate of theexpected reading.6.2.2.1 The estimate of standard deviation for
44、the verifiercan be improved by pooling with readings that are similar or ifit can be defined by an overall pattern of deviation withintensity.6.3 Final statements of performance of a verifier should bein terms of concentration. Standard deviations in terms ofreading can be converted to an equivalent
45、 standard deviation interms of concentration by multiplying by the slope of the of thecalibration equation at the point of the verifier reading. Detailsare given in Annex A1.6.3.1 If a deliberate change is made in the slope of acalibration curve after the collection of data, such as might bedone in
46、the transformation in 6.1.4, the effective standarddeviation of the reading will be the previous observed standarddeviation divided by the factor used to change the slope of thecurve. Thus, if a standard deviation has been calculated asbeing 0.6 when a curve slope (change of concentration dividedby
47、change in reading) at some point was 0.4, it would become0.3 if the curve was made twice as steep, that is, when the slopeat the same point was changed to 0.8.7. Use of Confidence Interval to Control SpectrochemicalAnalysis7.1 Practice E 876 uses Students t-table to establish therange of reading or
48、concentration around an average that willinclude the true reading or concentration at some confidencelevel. The calculation includes the standard deviation of themeasurement. To be effective, the standard deviation should beestimated with at least 16 df. The interval straddling theaverage will be 6t
49、s/=n , where t is a factor from the t-tablefor some probability level, s is the estimate of standarddeviation, and n is the number of readings taken for oneobservation. If control of a method depends upon observing anintensity reading, the confidence interval may be in terms of anintensity reading. If a method uses a computer to displayconcentration, the confidence interval should be in terms ofconcentration.7.1.1 If the confidence interval is used to judge when drifthas occurred, it will be appropriate to use a confidence level of95 % to anticipate when cont
