Geoffrey Lyman

Embed Size (px)

Citation preview

  • 7/28/2019 Geoffrey Lyman

    1/12

    Sampling with Discrete

    Contamination

    Geoff Lyman: Materials Sampling & Consulting

    Florent Bourgeois: Materials Sampling & Consulting Europe

    University of Toulouse

  • 7/28/2019 Geoffrey Lyman

    2/12

    Motivation

    The international grain trade operates understrict quality control constraints:

    Critical analytes are:

    Mycotoxins (Ochratoxin A and deoxynivalenol (DON) GMOs

    Pesticide residues

    Heavy metals Mycotoxins and GMOs are trace components

    and are highly heterogeneous in cargos

  • 7/28/2019 Geoffrey Lyman

    3/12

    Motivation

    OTA can reach a concentration of >10 000 ppb

    on a single kernel of wheat

    The allowable limit is 5 ppb in a cargo.

    There is NO data on OTA distribution in cargos

    It is desirable to have a means of estimatingthe variance (and distribution) of sampling for

    mycotoxins and GMOs

  • 7/28/2019 Geoffrey Lyman

    4/12

    The Problem

    If OTA and GMOs occur in slugs of relatively

    high concentration in the shipment, it takes

    only a small number of slugs to put the

    shipment off spec.

    Sampling by taking increments may have a

    high probability of missing the slugs, making

    the sampling variance potentially large

  • 7/28/2019 Geoffrey Lyman

    5/12

    The Solution

    The sampling variance for stratified sampling

    can be determined by considering slugs placed

    at random over a stratified sampling scheme.

  • 7/28/2019 Geoffrey Lyman

    6/12

    The Solution

    The statistical analysis is based on a Bernoulli

    process (we hit or miss the sample increment

    with each slug)

  • 7/28/2019 Geoffrey Lyman

    7/12

    The Problem

    The result is valid when the slug size is smallerthan the extent of the stratum (a single slug

    cannot hit more than one increment)

    The result is correct to first order for randomstratified sampling and random sampling

  • 7/28/2019 Geoffrey Lyman

    8/12

    The Solution

    For increments smaller than the slug, which isthe usual case, the relative sampling SD is,

    2

    2

    1 21

    E 3

    Sc

    S

    L u v vn

    n m u vc u

    m = slugs

    n = increments

    L = consignment mass

    u = slug mass

    v = increment mass

    cS = sample concentration

  • 7/28/2019 Geoffrey Lyman

    9/12

    The Solution

    This result extends to 2 and 3 dimensions withcorrections for geometry of the slug and

    increment

    However, to get a real picture of sampling risk,we need the full distribution of the sampling

    uncertainty, not just the variance

    This demands simulation of the process

  • 7/28/2019 Geoffrey Lyman

    10/12

    The Solution

    The sampling process is easy to simulate The lot concentration is held at 2 ppb (

  • 7/28/2019 Geoffrey Lyman

    11/12

    Examples (2 ppb average)

    22.2 tonnes between increments

    20 slugs, 200 kg

    Slug conc = 1000ppb

    RSD = 229%

    100 slugs, 200 kg

    Slug conc = 200ppb

    RSD = 102%

    100 slugs, 2200 kg

    Slug conc = 18ppb

    RSD = 30%

  • 7/28/2019 Geoffrey Lyman

    12/12

    Conclusions

    High

    heterogeneity

    High probability of false negatives

    Positives overestimate average

    concentration

    Moderateheterogeneity

    False negatives still probablePositives may overestimate average

    concentration

    Low

    heterogeneity

    No false negatives

    Results nearly normally distributedVariance is still significant compared to

    analytical variance (~10% RSD)