Gudelnes for CCPR and RMO Blateral Key Comparsons CCPR Workng Group on Key Comparson CCPR-G5 October 10 th, 2014 These gudelnes are prepared by CCPR WG-KC and RMO P&R representatves, and approved by CCPR, to ensure that blateral Key Comparsons n CCPR or RMOs are prepared and performed n a far and unform manner and that the results be lnked to the CCPR KCs approprately. Ths document s to supplement the CIPM gude, Measurement comparsons n the CIPM MRA, CIPM-MRA-D-05 (December 2011) [1]. 1. Purposes and ntaton of a blateral comparson 1.1 A CCPR blateral KC s carred out only rght after a CCPR KC s completed and when there was a problem n the reported results wth a partcpant and the partcpant wants to correct the results. All other blateral KCs are done as RMO comparsons. 1.2 RMO blateral KCs are carred out when the CCPR KC or an RMO KC of the quantty of nterest s not avalable n the requred tme frame, and a supportng evdence for a CMC clam s urgently needed. 1.3 Wth the am of reducng the number of blateral comparsons, t s recommended that two or more blateral comparsons of the same quantty planned at a smlar tme at dfferent RMOs be combned, or an NMI seekng for a blateral comparson can be nvted to an RMO comparson of the same quantty n other RMOs planned at a smlar tme. 1.3.1 An NMI needng a comparson to underpn ts CMCs shall contact ts RMO TC char before lookng for a blateral comparson. 1.3.2 When a need for a blateral comparson s dentfed, the RMO TC char should ask the WG-CMC char and ts members whether there are any comparsons of the same quantty planned n any other RMO, n whch the NMI can partcpate or wth whch the NMI can coordnate. 1.4 These blateral comparsons are normally conducted between two laboratores. However, there can be cases where two or three blateral comparsons wth the same plot laboratory are combned as one comparson. In ths case, the comparson may be regstered as one comparson n KCDB, and there should be one comparson report. 1.4.1 If there are sgnfcant dfferences n measurement tme or measurement condtons for each partcpant, the comparson can be regstered separately for each par of partcpants, n whch case, report should be prepared separately. 1.5 If the partcpants are members of dfferent RMOs, the RMO to whch the lab requestng the comparson belongs (laboratory to be lnked) should organze the comparson, unless the other drecton s agreed by both RMOs. - 1 -
2. Plot laboratory and Lnk laboratory 2.1 The plot laboratory s responsble for developng the comparson protocol, preparng and dstrbutng transfer standards (transfer standards may be prepared by the other lab), conductng measurements of all transfer standards, and preparng the comparson report. 2.2 A lnk laboratory n a blateral CCPR KC s a partcpant of the blateral KC and also s a partcpant of the prevous CCPR KC and provdes the lnk of results between the blateral KC and CCPR KC. 2.3 A lnk laboratory n a blateral RMO KC s a partcpant of the blateral KC and also s a partcpant of the prevous or current RMO or CCPR KC and provdes the lnk of results between the blateral KC and the RMO or CCPR KC. 2.4 The partcpant whose result s to be lnked to the CCPR or RMO KC s called non-lnk laboratory. 2.5 The lnk laboratory normally serves as the plot laboratory, but t s possble for the non-lnk laboratory to serve as the plot lab. In ths stuaton the lnk laboratory wll be responsble for pre-draft A, the non-lnk laboratory wll act as plot for comparson regstraton, preparaton of Draft A and subsequent work. 3. Development of Techncal Protocol 3.1 Follow the CCPR G6: Gudelnes for RMO PR Key Comparsons. The procedures for submttng the results to a thrd party lab (5.2, 5.3) should be ncluded n the protocol. 4. Regstraton to KCDB and approval of the techncal protocol 4.1 In case of an RMO blateral KC, upon completon of the techncal protocol, the plot lab sends t to the RMO PR TC Char, who wll submt the techncal protocol to WG-KC Secretary (copy to WG-KC Char) to request for approval. 4.2 After approval of the protocol, the RMO PR TC Char sends the KCDB entry form and the protocol document to CCPR Executve Secretary. The RMO PR TC Char should obtan confrmaton of the recept of the form by BIPM. 4.3 In case of a blateral CCPR KC, the plot laboratory sends the techncal protocol to the WG-KC Secretary for approval and nforms the KDCB manager. 4.4 Blateral comparsons are regstered n KCDB wth an ID number endng wth a dot and a seral number (e.g., EURAMET PR-K4.1). The number before the dot s dentcal wth that of the man CCPR or RMO comparson. 5. Measurements 5.1 Follow the CCPR G6: Gudelnes for RMO PR Key Comparsons n general, wth exceptons for the ponts gven here (5.2-5.4). 5.2 A thrd party (WG-KC Secretary) s desgnated for the comparson, and all the measurement results, both from the non-lnk laboratory and the lnk laboratory are - 2 -
submtted to the thrd party upon completon of each measurement, to ensure blndness of the comparson. 5.3 At completon of all measurements, the thrd party sends all the data receved to the lnk laboratory, so that the lnk laboratory can start Pre-Draft A process. 5.4 The thrd party sends to both partcpants all the raw data receved after Draft A s ssued. 6. Pre-Draft A process 6.1 The lnk laboratory always takes responsblty for pre-draft A, even f the non-lnk laboratory s actng as plot for other stages. 6.2 Sectons 1, 2, and 3 of CCPR G2: Gudelnes for CCPR Comparson Report Preparaton shall be followed by the lnk laboratory, but smplfed as below. 6.3 In the verfcaton of results, the lnk laboratory sends the other partcpant s reported results (as receved from the thrd party) to ask them to verfy. 6.4 In the revew of uncertanty budget, only the non-lnk laboratory s uncertanty budget s revewed. 6.5 Relatve data should be prepared by the lnk laboratory. In the revew of Relatve Data, the lnk laboratory prepares the Relatve Data and sends to the other partcpant, and dscusses any need for removng data of unstable artfacts, before the non-lnk laboratory has seen any absolute results. 6.6 If the non-lnk laboratory s actng as plot, at the end of the pre-draft A stage, the lnk laboratory passes the results and the calculatons of the pre-draft A phase to the plot for the Draft A report preparaton. 7. Preparaton of Draft A (for Key Comparsons) 7.1 After the Pre-Draft A processes are complete, the plot laboratory prepares and dstrbutes Draft A to both partcpants, whch dscloses the absolute results of the comparson. The Draft A should tabulate all the results as well as present them n graphcal form as necessary. It s recommended that the plot laboratory also dstrbute the data of the analyses n a spreadsheet fle. The Draft A should be dstrbuted wthn sx months after completon of all the measurements of the comparson. 7.2 The results of the blateral KC are lnked to the results of the most recent CCPR KC of the same quantty. The unlateral Degree of Equvalence (DoE) of the nonlnk laboratory should be calculated usng all approprate nformaton avalable from the Key Comparson and Blateral Comparson. The analyss process should be as smple as possble and transparent. In order to reduce the checkng requrements t should follow, where possble, prevously publshed approaches. An example approach s gven n Appendx A. 7.3 Blateral DoEs are not requred. - 3 -
8. Revew of Draft A and preparaton of Draft B Follow the procedure gven n CCPR G2: Gudelnes for CCPR Comparson Report Preparaton. References 1. CIPM MRA-D-05, Measurement comparsons n the CIPM MRA, Verson 1.1 of December 2011. - 4 -
Appendx A: An example analyss approach Ths example analyss s based on the work of Ojanen et al [1]. It assumes that a sngle artfact was measured durng each comparson. When multple artfacts are used, the effectve sngle artfact can be determned usng a smple mean of dfferent artfact values of y y n Equaton (1) or a smple mean of dfferent artfact values of y y 1 n Equaton (2). Here, subscrpt denotes the non-lnk laboratory, and subscrpt denotes the lnk laboratory ( s the desgnaton of the lnk laboratory as a partcpant n the CCPR KC to be lnked). The uncertanty assocated wth ths effectve sngle artfact s gven by the uncertanty declared by the partcpant for a sngle artfact. Ths ensures that the uncertantes assocated wth random effects are not reduced by takng a mean, dependng on the number of measurements made by the partcpant n ths way the partcpants are treated more equtably. The abbrevaton KC represents here the CCPR Key Comparson that the blateral comparson s lnked to. The abbrevaton BC represents the blateral comparson. It s mportant to understand whether the comparson model s based on absolute dfferences or relatve dfferences. In an absolute-dfference model: The KCRV has unts of the key comparson quantty. All uncertanty components have the same unts. The DoE of the BC partcpant wll be the best estmate of the systematc offset of that partcpant s measurements. So f the partcpant measured the CCPR KC vrtual artfact, ths would be the offset n that partcpant s measured value from the KCRV n the same unts. In a relatve-dfference model: The KCRV s the value 1. It s the average rato of partcpant s measurements of the vrtual artfact to the KC value of the vrtual artfact. All uncertanty components are expressed relatvely, n percentages. The DoE of the BC partcpant wll be the best estmate of the systematc rato (mnus one) of that partcpant s measurements to the KCRV. So f the partcpant measured the CCPR KC vrtual artfact, ths would be the rato of that partcpant s measured value and the KCRV mnus one. It could be expressed as percentage error. The value component of the unlateral Degree of Equvalence (DoE) of the non-lnk laboratory s calculated from the unlateral DoE of the lnk laboratory n the KC and the dfference between the measurement results of the partcpants of the BC. Where multple artfacts have been used, the average (smple mean) dfference can be used nstead. Where at least one partcpant has made two measurements, the two measurements should be averaged (smple mean). For an absolute-dfference model: - 5 -
D x x ref D y y, (1) from average of artefacts where D s the unlateral DoE of the non-lnk laboratory D x xref s the unlateral DoE for the lnk laboratory, calculated durng the KC from the measurement result x and KCRV x ref. y y s the average value (of multple artfacts) of the dfference between the non-lnk laboratory s measurement result (or average of the non-lnk laboratory s measurements) and the lnk-laboratory s measurement result (or average result) for each artfact n the BC. For a relatve-dfference model: D x x ref 1 y y 1 D from average of artefacts (2) where D s the unlateral DoE of the non-lnk laboratory D x x s the unlateral DoE for the lnk laboratory, calculated durng the KC from the measurement result x and KCRV x ref. y y s the average value (of multple artfacts) of the rato between the non-lnk laboratory s measurement result (or average of the non-lnk laboratory s measurements) and the lnk-laboratory s measurement result (or average result) for each artfact n the BC, subtractng unty to obtan a DoE. ref 1 1 The uncertanty component of the unlateral DoE s gven as an expanded uncertanty 2u D U D (3) where the standard uncertanty s calculated usng 2 2 2 2 2 2 2 2 ref 12 KC,r,KC,st,r,BC BC declared uncert. KC effects lnkng qualty BC effects u D u u x w s u u u s (4) There are four contrbutons n ths uncertanty calculaton. These are exactly the same for absolute-dfference and relatve-dfference models except that for absolute-dfference models the uncertantes are absolute uncertantes (expressed n the unt of the KC quantty) and for relatve-dfference models the uncertantes are relatve uncertantes (expressed, for example, as a percentage). - 6 -
Uncertantes assocated wth non-lnk laboratory measurement results u s the declared total standard uncertanty of the non-lnk laboratory for a sngle artfact. Ths ncludes uncertantes due to both correlated and uncorrelated effects. Key comparson effects ux ref s the standard uncertanty assocated wth the Key Comparson Reference Value. Ths value s avalable from the KC report. s KC s the transfer uncertanty for the KC. Ths may be an artfact nstablty factor calculated from known effects, or t may be the addtonal s term added durng a Mandel-Paule approach n obtanng consstency of the KC results. In ether case ths term s avalable from the KC report. If t s not used n the KC report, skc 0. 1 2 s calculated from the weght w assgned to the lnkng laboratory s measurements n the calculaton of the KCRV. Ths value s provded n the KC report; however, where t s not avalable, then w 0 wll gve a conservatve estmate of the uncertantes. The multpler w Lnkng qualty Ths term consders the qualty of the lnk provded by the lnk laboratory. It ncludes the standard uncertanty assocated wth uncorrelated effects (random uncertanty) of the lnk laboratory durng the KC, u,,r,kc the standard uncertanty assocated wth reproducblty of the lnk laboratory s scale between the KC and BC, u.,st the standard uncertanty assocated wth uncorrelated effects (random uncertanty) of the lnk laboratory durng the BC, u,r,bc Blateral comparson effects Ths term consders uncertantes from the blateral comparson. Ths ncludes the standard transfer uncertanty of the blateral comparson, s BC. The transfer uncertanty may come from known effects (e.g. known artfact nstablty), or from the smlar term durng the key comparson (to account for artfact nstablty that s not vsble n the BC). Alternatvely t may be approprate to consder sbc 0. Further notes on lnkng qualty for older KCs For some of the older KCs, the declared uncertanty assocated wth the lnk laboratory s measurements durng the KC may not be avalable broken down nto uncertantes assocated wth correlated (systematc) and uncorrelated (random) effects. In ths case the term u,r,kc s not readly avalable. Then ts value durng the KC may be estmated and/or the cases (a) and (b) below may be consdered: - 7 -
(a) If the tme nterval between KC and BC s short or f the lnk laboratory can use addtonal evdence to confrm the stablty of ts scale, then t may be assumed that u,st 0 and/or u,r,kc u,r,bc (stable scale). (b) Alternatvely for some older KCs, t may be more approprate to assume that the qualty of the lnk comes from the combned standard uncertanty assocated wth the lnk 2 2 2 laboratory s measurements durng the KC. In ths case u,r,kc u,st u (scale stablty unknown), where u s the declared total standard uncertanty of the lnk laboratory durng the KC. [1] Ojanen, M, Shpak, M, Kärhä, P, Leecharoen, R and Ikonen, E 2009 Uncertanty evaluaton for lnkng a blateral key comparson wth the correspondng CIPM key comparson Metrologa 46(5) 379-403 - 8 -