Animal feeding stuffs, cereals and milled cereal products — Guidelines for the application of near infrared spectrometry

ISO 12099:2010 gives guidelines for the determination by near infrared spectroscopy of constituents such as moisture, fat, protein, starch, and crude fibre as well as parameters such as digestibility in animal feeding stuffs, cereals and milled cereal products. The determinations are based on spectrometric measurement in the near infrared spectral region.

Aliments des animaux, céréales et produits de mouture des céréales — Lignes directrices pour l'application de la spectrométrie dans le proche infrarouge

L'ISO 12099:2010 fournit des lignes directrices pour la détermination par spectrométrie dans le proche infrarouge de constituants tels que l'eau, les matières grasses, les protéines, l'amidon et la cellulose brute, ainsi que des paramètres tels que la digestibilité des aliments pour animaux, des céréales et des produits céréaliers moulus. Les déterminations sont basées sur des mesurages spectrométriques dans le domaine du proche infrarouge.

General Information

Status
Withdrawn
Publication Date
06-Jun-2010
Withdrawal Date
06-Jun-2010
Current Stage
9599 - Withdrawal of International Standard
Completion Date
10-Aug-2017
Ref Project

Relations

Buy Standard

Standard
ISO 12099:2010
English language
12 pages
sale 15% off
Preview
sale 15% off
Preview
Standard
ISO 12099:2010 - Animal feeding stuffs, cereals and milled cereal products -- Guidelines for the application of near infrared spectrometry
English language
30 pages
sale 15% off
Preview
sale 15% off
Preview
Standard
ISO 12099:2010 - Aliments des animaux, céréales et produits de mouture des céréales -- Lignes directrices pour l'application de la spectrométrie dans le proche infrarouge
French language
32 pages
sale 15% off
Preview
sale 15% off
Preview

Standards Content (Sample)

МЕЖДУНАРОДНЫЙ ISO
СТАНДАРТ 12099
Первое издание
2010-06-15

Корма для животных, зерновые и
молотые зерновые продукты.
Руководство по применению
спектрометрии в ближней
инфракрасной области спектра
Animal feeding stuffs, cereals and milled cereal products — Guidelines
for the application of near infrared spectrometry



Ответственность за подготовку русской версии несёт GOST R
(Российская Федерация) в соответствии со статьёй 18.1 Устава ISO
Ссылочный номер
ISO 12099:2010(R)
©
ISO 2010

---------------------- Page: 1 ----------------------
ISO 12099:2010(R)
Отказ от ответственности при работе в PDF
Настоящий файл PDF может содержать интегрированные шрифты. В соответствии с условиями лицензирования, принятыми
фирмой Adobe, этот файл можно распечатать или смотреть на экране, но его нельзя изменить, пока не будет получена
лицензия на установку интегрированных шрифтов в компьютере, на котором ведется редактирование. В случае загрузки
настоящего файла заинтересованные стороны принимают на себя ответственность за соблюдение лицензионных условий
фирмы Adobe. Центральный секретариат ISO не несет никакой ответственности в этом отношении.
Adobe – торговый знак фирмы Adobe Systems Incorporated.
Подробности, относящиеся к программным продуктам, использованным для создания настоящего файла PDF, можно найти в
рубрике General Info файла; параметры создания PDF были оптимизированы для печати. Были приняты во внимание все
меры предосторожности с тем, чтобы обеспечить пригодность настоящего файла для использования комитетами-членами
ISO. В редких случаях возникновения проблемы, связанной со сказанным выше, просьба проинформировать Центральный
секретариат по адресу, приведенному ниже.


ДОКУМЕНТ ЗАЩИЩЕН АВТОРСКИМ ПРАВОМ


© ISO 2010
Все права сохраняются. Если не указано иное, никакую часть настоящей публикации нельзя копировать или использовать в
какой-либо форме или каким-либо электронным или механическим способом, включая фотокопии и микрофильмы, без
предварительного письменного согласия ISO, которое должно быть получено после запроса о разрешении, направленного по
адресу, приведенному ниже, или в комитет-член ISO в стране запрашивающей стороны.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Опубликовано в Швейцарии

ii © ISO 2010 – Все права сохраняются

---------------------- Page: 2 ----------------------
ISO 12099:2010(R)
Содержание Страница
Предисловие .iv
Введение .v
1  Область применения .1
2  Термины и определения .1
3  Принцип.2
4  Аппаратура.2
5  Калибровка и первоначальная валидация .2
6  Статистики для измерения рабочих характеристик.5
7  Отбор проб.12
8  Проведение анализа .12
9  Проверка стабильности прибора.13
10  Проведение проверки качества калибровки .13
11  Прецизионность и точность.15
12  Протокол испытания.15
Приложение A (информативное) Руководство по конкретным стандартам NIR.17
Приложение B (информативное) Примеры рисунков .18
Приложение C (информативное) Дополнительные термины и определения .24
Библиография.33

© ISO 2010 – Все права сохраняются iii

---------------------- Page: 3 ----------------------
ISO 12099:2010(R)
Предисловие
Международная организация по стандартизации (ISO) является всемирной федерацией национальных
организаций по стандартизации (комитетов-членов ISO). Разработка международных стандартов
обычно осуществляется техническими комитетами ISO. Каждый комитет-член, заинтересованный в
деятельности, для которой был создан технический комитет, имеет право быть представленным в этом
комитете. Международные правительственные и неправительственные организации, имеющие связи с
ISO, также принимают участие в работах. ISO работает в тесном сотрудничестве с Международной
электротехнической комиссией (IEC) по всем вопросам стандартизации в области электротехники.
Международные стандарты разрабатываются в соответствии с правилами, установленными в
Директивах ISO/IEC, Часть 2.
Основная задача технических комитетов состоит в подготовке международных стандартов. Проекты
международных стандартов, одобренные техническими комитетами, рассылаются комитетам-членам
на голосование. Их опубликование в качестве международных стандартов требует одобрения, по
меньшей мере, 75 % комитетов-членов, принимающих участие в голосовании.
Следует иметь в виду, что некоторые элементы данной части ISO 16065 могут быть объектом
патентных прав. Организация ISO не должна нести ответственность за идентификацию какого-либо
одного или всех патентных прав.
ISO 12099 был подготовлен Европейским комитетом по стандартизации (CEN), Техническим комитетом
TC 327, Корма для животных. Методы отбора проб и анализа, совместно с Техническим комитетом
ISO TC 34, Пищевые продукты, Подкомитетом SC 10, Корма для животных, в соответствии с
Соглашением по техническому сотрудничеству между ISO и CEN (Венское Соглашение).
iv © ISO 2010 – Все права сохраняются

---------------------- Page: 4 ----------------------
ISO 12099:2010(R)
Введение
Настоящий международный стандарт разработан на основе международного стандарта
[15]
ISO 21543⎪IDF 201 , подготовленного Техническим комитетом ISO/TC 34, Пищевые продукты,
Подкомитетом SC 5, Молоко и молочные продукты, и Международной Молочной Федерацией (IDF).
© ISO 2010 – Все права сохраняются v

---------------------- Page: 5 ----------------------
МЕЖДУНАРОДНЫЙ СТАНДАРТ ISO 12099:2010(R)

Корма для животных, зерновые и молотые зерновые
продукты. Руководство по применению спектрометрии в
ближней инфракрасной области спектра
1 Область применения
В настоящем международном стандарте дано руководство по определению методом спектроскопии в
ближней инфракрасной области спектра таких составляющих, как влага, жиры, белки, крахмал и сырая
клетчатка, а также таких параметров, как перевариваемость в отношении кормов для животных,
зерновых и молотых зерновых продуктов.
Определения основаны на спектрометрическом измерении в ближней инфракрасной области спектра.
2 Термины и определения
Применительно к данному международному стандарту используются следующие термины и
определения.
2.1
прибор, работающий в ближней инфракрасной области спектра
near infrared instrument
ИК-анализатор
NIR-анализатор
NIR instrument
прибор, с помощью которого при использовании в установленных условиях, можно прогнозировать
содержание компонентов (2.3) и технологические параметры (2.4) в матрице через взаимосвязь с
поглощением в ближней инфракрасной области спектра
ПРИМЕЧАНИЕ В контексте данного международного стандарта матрицами являются корма для животных,
зерновые и молотые зерновые продукты.
2.2
корма для животных
animal feeding stuff
любое вещество или продукт, включая добавки, обработанные, частично обработанные или
необработанные, предназначенные для кормления скота
ПРИМЕРЫ Сырые материалы, фураж, кормовая мука, комбикорм и другие конечные продукты, а также корма
для домашних питомцев.
2.3
содержание компонентов
constituent content
массовая доля веществ, определенная подходящим стандартизованным или подтвержденным
химическим методом
ПРИМЕЧАНИЕ 1 Массовую долю часто выражают в процентах.
ПРИМЕЧАНИЕ 2 Примерами определяемых компонентов являются влага, жиры, белки, сырая клетчатка,
нейтральное детергентное волокно и кислое детергентное волокно. В отношении подходящих методов см.,
например, Ссылки [1] – [16].
© ISO 2010 – Все права сохраняются 1

---------------------- Page: 6 ----------------------
ISO 12099:2010(R)
2.4
технологический параметр
technological parameter
свойство или функциональная характеристика матрицы, которые можно определить подходящим
стандартизованным или подтвержденным методом(ами)
ПРИМЕР Перевариваемость.
ПРИМЕЧАНИЕ 1 В контексте данного международного стандарта матрицы представляют собой корма для
животных, зерновые и молотые зерновые продукты.
ПРИМЕЧАНИЕ 2 Можно разработать и выполнить валидацию NIR-методов в отношении других параметров и
матриц, при условии соблюдения процедуры настоящего международного стандарта. Единицы измерения
определяемых параметров должны соответствовать единицам, используемым в стандартных (контрольных)
методах.
3 Принцип
Спектральные данные в ближней инфракрасной области собирают и трансформируют в концентрации
компонента или параметра по калибровочным моделям, разработанным по репрезентативным пробам
рассматриваемых продуктов.
4 Аппаратура
4.1 ИК-анализаторы (в ближней ИК-области спектра), на основе измерения диффузионного
−1
отражения или пропускания в диапазоне длин волн от 770 нм до 2 500 нм (от 12 900 см до
−1
4 000 см ), или сегментах этого диапазона или при выбранных длинах волн или волновых числах.
Оптический принцип может быть дисперсионным (например, дифракционные монохроматоры),
интерферометрическим или нетепловым (например, светоиспускающие диоды, лазерные диоды и
лазеры). Прибор должен быть оснащен диагностической системой для тестирования
фотометрического шума и воспроизводимости, точности длины волны или волнового числа и
прецизионности длины волны или волнового числа (для сканирующего спектрофотометра).
Анализатор должен измерять пробу достаточно большого объема или площади поверхности, чтобы
устранить всякое заметное влияние неоднородности за счет химического состава или физических
свойств анализируемой пробы. Длину оптического пути в пробе (толщина слоя пробы) при измерениях
пропускания следует оптимизировать согласно рекомендациям изготовителя по отношению к
интенсивности сигнала, чтобы получить линейность и максимальное соотношение сигнал/шум. При
измерении отражения, для устранения эффектов высушивания поверхность взаимодействующего слоя
пробы необходимо отделить кварцевым окошком или другим подходящим материалом.
4.2 Подходящее устройство для размалывания или измельчения, для подготовки пробы (при
необходимости).
ПРИМЕЧАНИЕ Изменения условий размалывания или измельчения могут повлиять на NIR-измерения.
5 Калибровка и первоначальная валидация
5.1 Общие положения
Перед применением прибор подлежит калибровке. Ввиду разнообразия калибровочных систем,
которые можно применять к ИК-анализаторам, невозможно привести конкретную методику калибровки.
2 © ISO 2010 – Все права сохраняются

---------------------- Page: 7 ----------------------
ISO 12099:2010(R)
Для пояснения методов калибровки см., например, Ссылку [17] и соответствующие инструкции
изготовителя. Для валидации, важно иметь достаточное количество репрезентативных проб,
охватывающих такие варианты, как:
a) различные комбинации и диапазоны состава компонентов соответствующего максимального и
минимального содержания;
b) сезонные, географические и генетические эффекты на фураж, кормовое сырье и зерновые;
c) различная техника и условия обработки;
d) различные условия хранения;
e) различная температура пробы и прибора;
f) различные варианты приборов (различия между приборами).
ПРИМЕЧАНИЕ Для надежной валидации требуется не менее 20 проб.
5.2 Стандартные (контрольные) методы
Следует пользоваться принятыми на международном уровне стандартными методами определения
содержания влаги, жиров, белков и других компонентов и параметров. См. Ссылки [1] – [16] в
отношении примеров.
Стандартный метод, используемый для калибровки, должен находиться с статистически управляемом
режиме, т.е. для любой пробы изменчивость должна включать случайные изменения воспроизводимой
системы. Важно знать прецизионность стандартного метода.
5.3 Выбросы
В большинстве случаев в процессе калибровки и валидации наблюдаются статистические выбросы.
Выбросы могут относиться к NIR-данным (данным, полученным в ближней ИК-области спектра)
(спектральные выбросы, здесь и далее называемые «x-выбросы») или погрешностям в опорных
данных (данных, полученных в стандартных методах), или пробам (результатам) с различными
взаимосвязями между опорными данными и NIR-данными (здесь и далее называемые «y-выбросы»)
(см. Рисунки B.1 – B.5).
Для целей валидации пробы (результаты) не считаются выбросами в том случае, если:
a) они находятся в рабочем диапазоне компонентов /параметров в калибровке;
b) они находятся в пределах спектральной изменчивости калибровочных образцов, например, как
оцененные по расстоянию Махаланобиса;
c) спектральный остаток будет ниже предельного значения, определенного в процессе калибровки;
d) прогнозируемый остаток будет ниже предельного значения, определенного в процессе калибровки.
Если проба (результат) оказалась выбросом, то ее необходимо сначала проверить, чтобы посмотреть,
является ли она x-выбросом. Если она превышает пределы для x-выбросов, определенные для
калибровки, ее следует исключить. Если это не x-выброс, тогда следует проверить и опорное значение,
и NIR-пррогнозируемое значение. Если проверка подтверждает первоначальные значения, то пробу
(результат) не следует исключать: ее следует включить в статистику валидации. Если полученные при
проверке значения показывают, что либо первоначальные опорные значения, либо NIR-
прогнозируемые значения были с погрешностью, то следует использовать новые значения.
© ISO 2010 – Все права сохраняются 3

---------------------- Page: 8 ----------------------
ISO 12099:2010(R)
5.4 Модели валидации и калибровки
5.4.1 Общие положения
Перед применением калибровочные уравнения подлежат валидации локально на независимой
анализируемой выборке, которая является репрезентативной для множества подлежащих анализу
проб. Для определения систематической погрешности (смещения) требуется не менее 10 проб; для
определения стандартной ошибки прогноза (SEP, см.6.5) требуется не менее 20 проб. Валидация
должна выполняться для каждого типа проб, компонента или параметра и температуры. Калибровка
действительна только для данных вариантов, т.е. типов проб, диапазона и температуры,
использованных в валидации.
Результаты, полученные на независимой анализируемой выборке, наносят на график, опорные
значения против NIR-значений, и разности против опорных результатов, чтобы получить визуальное
впечатление о результативности калибровки. Рассчитывают SEP (см. 6.5) и исследуют график
остатков по данным, скорректированным на среднюю систематическую ошибку (смещение) в
отношении выбросов, т.е. проб (результатов), остаток для которых превышает ± 3s .
SEP
Если процесс валидации показывает, что модель не может дать приемлемой статистики, тогда такую
модель использовать не следует.
ПРИМЕЧАНИЕ Приемлемость модели зависит от таких критериев, как исполнение стандартного метода,
охваченного диапазона и цели анализа, а решение принимают участвующие стороны.
Следующий этап заключается в подгонке NIR-данных, y , и опорных данных, y , методом линейной
NIR ref
регрессии (y = by + a), чтобы получить статистику, которая описывает результаты валидации.
ref NIR
5.4.2 Поправка на систематическую погрешность (смещение)
Данные также исследуют в отношении смещения между методами. Разность между средними NIR-
прогнозируемыми и опорными значениями, значительно отличающаяся от нуля, показывает на
систематическую погрешность калибровки. Систематическую погрешность можно устранить путем
корректировки константы (см. 6.3) в калибровочном уравнении.
5.4.3 Регулировка наклона
Если наклон, b, значительно отличается от 1, калибровка смещена.
Подгонки наклона или отсекаемого отрезка при калибровке обычно не рекомендуется, если калибровка
применяется не к новым типам проб или приборов. Если повторное исследование калибровки не
выявило выбросов, особенно выбросов с высокой балансировкой, предпочтительно расширить
калибровочную выборку, чтобы включить больше проб. Однако, если наклон отрегулирован, тогда
следует испытать калибровку на новой независимой анализируемой выборке.
5.4.4 Расширение калибровочной выборки
Если точность калибровки не оправдывает ожидания, калибровочную выборку следует расширить с
включением большего числа проб или выполнить новую калибровку. В любом случае, если выполняют
новую калибровку на расширенной выборке, процесс валидации следует повторить на новой выборке
для валидации. Если необходимо, расширение калибровочной выборки следует продолжить, пока не
будут получены приемлемые результаты на выборке для валидации.
5.5 Изменение условий измерения и работы прибора
Если дополнительная калибровка не проведена, локальную валидацию NIR-метода, устанавливающую
точность этого метода, обычно не считается достоверной в случае изменения условий испытания.
4 © ISO 2010 – Все права сохраняются

---------------------- Page: 9 ----------------------
ISO 12099:2010(R)
Например, калибровки, проведенные для определенного множества проб, могут оказаться
недействительными за границами этого множества, хотя диапазон концентраций аналита не
изменяется. Калибровка, проведенная на травянистом силосе из одного района, может не дать такую
же точность на силосе из другого района, если параметры генетики, выращивания и переработки не
совпадают.
Изменения в технике представления пробы или условиях измерений (например, температуры), не
учтенные в калибровочной выборке, также могут повлиять на результат.
Калибровки, выполненные на определенном приборе, не всегда можно переносить непосредственно
на идентичный анализатор, работающий по тому же принципу. Может потребоваться выполнение
регулировки систематической погрешности, наклона или отсекаемого отрезка, по калибровочных
характеристикам. В большинстве случаев необходимо стандартизовать оба прибора друг
относительно друга до переноса калибровочных характеристик (Ссылка [17]). Можно использовать
разные методы стандартизации для переноса калибровки между приборами разного типа, при условии,
что пробы измеряются одинаковым образом (отражение, пропускание) и что область спектра обычная.
Если изменились условия, рекомендуется выполнить дополнительную валидацию.
Калибровку рекомендуется проверять всякий раз, когда меняют или ремонтируют основную часть
анализатора (оптическую систему, детектор).
6 Статистики для измерения рабочих характеристик
6.1 Общие положения
Рабочие характеристики модели прогнозирования должны быть определены по выборке проб для
валидации. Эта выборка состоит из проб, которые независимы от калибровочной выборки. На
предприятии, это новые партии, в сельском хозяйстве – это новый урожай или новое место
проведения эксперимента.
Такую выборку проб необходимо внимательно проанализировать стандартными методами.
Необходимо особое внимание уделить анализу проб для валидации, а прецизионность результатов
более важна для выборки для валидации, чем для проб, используемых на этапе калибровки.
Число проб для валидации должно быть не менее 20, чтобы вычислить статистику с определенным
доверием.
6.2 Построение графика по результатам
Важно визуализировать результаты в графиках, например, в графике зависимости опорных значений
от прогнозируемых или остатков от прогнозируемых значений.
Остатки определяются как:
)
ey=−y (1)
ii i
где
y i-тое опорное значение;
i
)
y i-тое прогнозируемое значение, полученное при использовании многофакторной NIR-модели.
i
Способ вычисления разностей дает положительную систематическую погрешность, если
прогнозируемые значения слишком высоки, и отрицательную, если прогнозируемые значения слишком
низки по сравнению с опорными значениями.
График данных дает промежуточное впечатление о корреляции, систематической погрешности,
наклоне и показывает наличие очевидных выбросов (см. Рисунок 1).
© ISO 2010 – Все права сохраняются 5

---------------------- Page: 10 ----------------------
ISO 12099:2010(R)

Обозначение
1 линия 45° (идеальная линия с систематической погрешностью, e = 0 , наклоном, b = 1) a отрезок
2 линия 45° , смещенная систематической погрешностью, e e систематическая погрешность
3 линия линейной регрессии с отсекаемым отрезком y , a y прогнозируемое значение в
ref NIRS
ближней ИК-области спектра
4 выбросы y опорное значение
ref
ПРИМЕЧАНИЕ Выбросы имеют сильное влияние на расчет наклона и подлежат исключению, если
результаты предполагается использовать для выверки.
Рисунок 1 — График рассеяния для выборки для валидации, y = f(a + by )
ref NIRS
6.3 Систематическая погрешность
Большей частью систематическая погрешность или систематическая ошибка – это то, что наблюдается
с NIR-моделями. Систематическая погрешность может возникать за счет того, что: новые пробы
определенного типа не были предварительно просмотрена моделью, за счет дрейфа прибора,
отклонений в мокрых химических методах, изменений в процессах, и изменений при подготовке проб.
При количестве независимых проб, n, систематическая погрешность (или смещение) является средней
разностью, e , и ее можно определить как:
n
1
(2)
ee=
∑ i
n
i=1
где e остаток, определенный в Формуле (1), или
i
nn
⎛⎞
1
) )
⎜⎟
ey=−y=y−y (3)
ii
∑∑
⎜⎟
n
ii==11
⎝⎠
где
6 © ISO 2010 – Все права сохраняются

---------------------- Page: 11 ----------------------
ISO 12099:2010(R)
y i-тое опорное значение;
i
)
y i-тое прогнозируемое значение, полученное при использовании многофакторной NIR-модели;
i
и
)
y среднее из прогнозируемых значений;
y среднее из опорных значений.
Значимость систематической погрешности проверяют с помощью t- критерия. Расчет доверительных
пределов систематической погрешности (BCL), T , определяет предельные значения для выполнения
b
формулы приемки или браковки на небольшой выборке проб из нового множества.
ts
1/−α2 SEP
()
(4)
T =±
b
n
где
α вероятность появления ошибки типа I;
t соответствующее t-значение Стьюдента для двустороннего критерия со степенями свободы,
связанными с SEP, и выделенная вероятность ошибки типа I (см. Таблицу 1);
n число независимых проб;
s стандартная ошибка прогноза (см. 6.5).
SEP
ПРИМЕР При n = 20, и s = 1, пределы будут равны
SEP
2, 09 × 1
T =± = ± 0, 48 (5)
b
20
Это означает, что систематическая погрешность, тестированная на 20 пробах, должна быть выше 48 %
от стандартной ошибки прогноза, которая должна считаться отличной от нуля.
Таблица 1 — Значения t-распределения с вероятностью, α = 0,05 = 5 %
n t n t n t n t
5 2,57 11 2,20 17 2,11 50 2,01
6 2,45 12 2,18 18 2,10 75 1,99
7 2,36 13 2,16 19 2,09 100 1,98
8 2,31 14 2,14 20 2,09 200 1,97
9 2,26 15 2,13 30 2,04 500 1,96
10 2,23 16 2,12 40 2,02 1 000 1,96
1)
ПРИМЕЧАНИЕ Можно использовать функцию Excel TINV.

1) Excel – торговое наименование продукта, поставляемого компанией Microsoft. Эта информация дается для
удобства пользователей данного документа и не указывает на предпочтение со стороны ISO в отношении этого
продукта. Можно использовать равноценные продукты, при условии получения аналогичных результатов.
© ISO 2010 – Все права сохраняются 7

---------------------- Page: 12 ----------------------
ISO 12099:2010(R)
6.4 Среднеквадратическая ошибка прогноза (RMSEP)
Среднеквадратическая ошибка прогноза (RMSEP), s , (C.3.6) математически выражается
RMSEP
следующим образом:
n
2
e
i

i=1
(6)
s =
RMSEP
n
где
e остаток i-той пробы;
i
n число независимых проб.
Это значение можно сравнить с SEC (C.3.3) и SECV (C.3.4).
RMSEP включает случайную ошибку (SEP) и систематическую погрешность или смещение. Она также
включает погрешность стандартных методов (также как SEC и SECV).
(1n −)
22
s =+se
(7)
RMSEP SEP
n
где
n число независимых проб;
s стандартная ошибка прогноза (см. 6.5);
SEP
e систематическая погрешность или смещение.
Не существует прямого теста для RMSEP. По этой причине выделяют систематическую ошибку,
систематическую погрешность или e , и случайную ошибку, SEP или s .
SEP
6.5 Стандартная ошибка прогноза (SEP)
Стандартная ошибка прогноза SEP, s , или стандартное отклонение остатков, которое выражает
SEP
точность рутинных NIR-результатов (NIR), скорректированных по средней разности (смещению) между
рутинным NIR- и стандартным методом, может быть рассчитано по Формуле (8):
n
2
ee−
()
∑ i
i=1
s = (8)
SEP
n −1
где
n число независимых проб;
e остаток i-той пробы;
i
e систематическая погрешность или смещение.
Ошибку SEP следует связать с SEC (C.3.3) или SECV (C.3.4) для проверки надежности калибровочной
модели для выбранной выборки для валидации.
8 © ISO 2010 – Все права сохраняются

---------------------- Page: 13 ----------------------
ISO 12099:2010(R)
Доверительные пределы необъяснимой ошибки (UECL), T , рассчитывают из F-критерия (отношение
UE
2 дисперсий) (см. Ссылку [19] и Таблицу 2).
Ts= F (9)
UE SEC (α:vM, )
где
s стандартная ошибка калибровки (C.3.3);
SEC
α вероятность появления ошибки типа I;
ν = n − 1 числитель степеней свободы, связанных с SEP анализируемой выборки, в которой
n число проб в процессе валидации;
M = n − p − 1 знаменатель степеней свободы, связанных с SEC (стандартной ошибкой
c
калибровки);
В котором
n число проб для калибровки,
c
p число членов или факторов PLS в модели.
ПРИМЕЧАНИЕ 1 SEC можно заменить на SECV , которая является лучшей статистикой, чем SEC; поскольку
зачастую SEC слишком оптимистична, s > s .
SECV SEC
ПРИМЕР При n = 20, α = 0,05, M = 100, и s = 1,
SEC
T = 1, 30 (10)
UE
Это означает, что для 20 проб SEP может быть принята, т.е. примерно на 30 % больше, чем SEC.
2)
ПРИМЕЧАНИЕ 2 Можно использовать функцию Excel f FINV.
F-критерий нельзя использовать для сравнения двух калибровок на одной и той же выборке для
валидации. Требуется (как здесь), чтобы работало две независимые выборки. Другой критерий
требуется для сравнения двух и более моделей на одной и той же выборке данных.

2) Excel – торговое наименование продукта, поставляемого компанией Microsoft. Эта информация дается для
удобства пользователей данного документа и не указывает на предпочтение со стороны ISO в отношении этого
продукта. Можно использовать равноценные продукты, при условии получения аналогичных результатов.
© ISO 2010 – Все права сохраняются 9

---------------------- Page: 14 ----------------------
ISO 12099:2010(R)
Таблица 2 — F-значения и корни квадратные из F-значений как функция степеней свободы
числителя, связанного с SEP, и знаменателя, связанного с SEC
[см. определения под Формулой (9)]
F √[F ]
(α: ν, M) (α: ν, Μ)
Степени Степени свободы (SEC) Степени Степени свободы (SEC)
свободы свободы
50 100 200 500 1 000 50 100 200 500 100
(SEP) (SEP)
5 2,40 2,31 2,26 2,23 2,22 5 1,55 1,52 1,50 1,49 1,49
6 2,29 2,19 2,14 2,12 2,11 6 1,51 1,48 1,46 1,45 1,45
7 2,20 2,10 2,06 2,03 2,02 7 1,48 1,45 1,43 1,42 1,42
8 2,13 2,03 1,98 1,96 1,95 8 1,46 1,43 1,41 1,40 1,40
9 2,07 1,97 1,93 1,90 1,89 9 1,44 1,41 1,39 1,38 1,37
10 2,03 1,93 1,88 1,85 1,84 10 1,42 1,39 1,37 1,36 1,36
11 1,99 1,89 1,84 1,81 1,80 11 1,41 1,37 1,36 1,34 1,34
12 1,95 1,85 1,80 1,77 1,76 12 1,40 1,36 1,34 1,33 1,33
13 1,92 1,82 1,77 1,74 1,73 13 1,39 1,35 1,33 1,32 1,32
14 1,89 1,79 1,74 1,71 1,70 14 1,38 1,34 1,32 1,31 1,30
15 1,87 1,77 1,72 1,69 1,68 15 1,37 1,33 1,31 1,30 1,29
16 1,85 1,75 1,69 1,66 1,65 16 1,36 1,32 1,30 1,29 1,29
17 1,83 1,73 1,67 1,64 1,63 17 1,35 1,31 1,29 1,28 1,28
18 1,81 1,71 1,66 1,62 1,61 18 1,30 1,31 1,29 1,27 1,27
19 1,80 1,69 1,64 1,61 1,60 19 1,34 1,30 1,28 1,27 1,26
29 1,69 1,58 1,52 1,49 1,48 29 1,30 1,26 1,23 1,22 1,22
49 1,60 1,48 1,42 1,38 1,37 49 1,27 1,22 1,19 1,17 1,17
99 1,53 1,39 1,32 1,28 1,26 99 1,24 1,18 1,15 1,13 1,12
199 1,48 1,34 1,26 1,21 1,19 199 1,22 1,16 1,12 1,10 1,09
499 1,46 1,31 1,22 1,16 1,13 499 1,21 1,14 1,11 1,08 1,07
999 1,45 1,30 1,21 1,14 1,11 999 1,20 1,14 1,10 1,07 1,05
6.6 Наклон
)
Наклон, b, простой регрессии y = a + b yчасто сообщают в публикациях NIR.
Необходимо отметить, что наклон должен вычисляться с опорными значениями как зависимая
переменная, а с прогнозируемыми NIR-значениями как независимая переменная, если вычисленный
наклон будет использоваться для регулировки результатов NIR (как в случае обратной
многофакторной регрессии, используемой для построения модели прогноза).
Из подбора методом наименьших квадратов наклон рассчитывают как:
)
s
yy
b = (11)
2
s

y
где
...

INTERNATIONAL ISO
STANDARD 12099
First edition
2010-06-15

Animal feeding stuffs, cereals and milled
cereal products — Guidelines for the
application of near infrared spectrometry
Aliments des animaux, céréales et produits de mouture des céréales —
Lignes directrices pour l'application de la spectrométrie dans le proche
infrarouge




Reference number
ISO 12099:2010(E)
©
ISO 2010

---------------------- Page: 1 ----------------------
ISO 12099:2010(E)
PDF disclaimer
This PDF file may contain embedded typefaces. In accordance with Adobe's licensing policy, this file may be printed or viewed but
shall not be edited unless the typefaces which are embedded are licensed to and installed on the computer performing the editing. In
downloading this file, parties accept therein the responsibility of not infringing Adobe's licensing policy. The ISO Central Secretariat
accepts no liability in this area.
Adobe is a trademark of Adobe Systems Incorporated.
Details of the software products used to create this PDF file can be found in the General Info relative to the file; the PDF-creation
parameters were optimized for printing. Every care has been taken to ensure that the file is suitable for use by ISO member bodies. In
the unlikely event that a problem relating to it is found, please inform the Central Secretariat at the address given below.


COPYRIGHT PROTECTED DOCUMENT


©  ISO 2010
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland

ii © ISO 2010 – All rights reserved

---------------------- Page: 2 ----------------------
ISO 12099:2010(E)
Contents Page
Foreword .iv
Introduction.v
1 Scope.1
2 Terms and definitions .1
3 Principle.2
4 Apparatus.2
5 Calibration and initial validation .2
6 Statistics for performance measurement .4
7 Sampling.11
8 Procedure.11
9 Checking instrument stability .12
10 Running performance check of calibration .12
11 Precision and accuracy .14
12 Test report.14
Annex A (informative) Guidelines for specific NIR standards.15
Annex B (informative) Examples of figures.16
Annex C (informative) Supplementary terms and definitions .22
Bibliography.29

© ISO 2010 – All rights reserved iii

---------------------- Page: 3 ----------------------
ISO 12099:2010(E)
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO 12099 was prepared by the European Committee for Standardization (CEN) Technical committee TC 327,
Animal feeding stuffs — Methods of sampling and analysis, in collaboration with ISO Technical Committee
TC 34, Food products, Subcommittee SC 10, Animal feeding stuffs, in accordance with the Agreement on
technical cooperation between ISO and CEN (Vienna agreement).
iv © ISO 2010 – All rights reserved

---------------------- Page: 4 ----------------------
ISO 12099:2010(E)
Introduction
[15]
This International Standard has been drafted using, as a basis, ISO 21543⎪IDF 201 , prepared by Technical
Committee ISO/TC 34, Food products, Subcommittee SC 5, Milk and milk products, and the International
Dairy Federation (IDF).

© ISO 2010 – All rights reserved v

---------------------- Page: 5 ----------------------
INTERNATIONAL STANDARD ISO 12099:2010(E)

Animal feeding stuffs, cereals and milled cereal products —
Guidelines for the application of near infrared spectrometry
1 Scope
This International Standard gives guidelines for the determination by near infrared spectroscopy of
constituents such as moisture, fat, protein, starch, and crude fibre as well as parameters such as digestibility
in animal feeding stuffs, cereals and milled cereal products.
The determinations are based on spectrometric measurement in the near infrared spectral region.
2 Terms and definitions
For the purposes of this International Standard, the following terms and definitions apply.
2.1
near infrared instrument
NIR instrument
apparatus which, when used under specified conditions, predicts constituent contents (2.3) and
technological parameters (2.4) in a matrix through relationships to absorptions in the near infrared range
NOTE In the context of this International Standard, the matrices are animal feeding stuffs, cereals and milled cereal
products.
2.2
animal feeding stuff
any substance or product, including additives, whether processed, partially processed or unprocessed,
intended to be used for oral feeding to animals
EXAMPLES Raw materials, fodder, animal flour, mixed feed and other end products, and pet food.
2.3
constituent content
mass fraction of substances determined using the appropriate, standardized or validated chemical method
NOTE 1 The mass fraction is often expressed as a percentage.
NOTE 2 Examples of constituents determined include moisture, fat, protein, crude fibre, neutral detergent fibre, and
acid detergent fibre. For appropriate methods, see, e.g., References [1] to [16].
2.4
technological parameter
property or functionality of a matrix that can be determined using the appropriate standardized or validated
method(s)
EXAMPLE Digestibility.
NOTE 1 In the context of this International Standard, the matrices are animal feeding stuffs, cereals and milled cereal
products.
© ISO 2010 – All rights reserved 1

---------------------- Page: 6 ----------------------
ISO 12099:2010(E)
NOTE 2 It is possible to develop and validate NIR methods for other parameters and matrices than listed, as long as
the procedure from this International Standard is observed. The measuring units of the parameters determined have to
follow the units used in the reference methods.
3 Principle
Spectral data in the near infrared (NIR) region are collected and transformed to constituent or parameter
concentrations by calibration models developed on representative samples of the products concerned.
4 Apparatus
4.1 Near-infrared instruments, based on diffuse reflectance or transmittance measurement covering the
−1 −1
NIR wavelength region, 770 nm to 2 500 nm (12 900 cm to 4 000 cm ), or segments of this or at selected
wavelengths or wavenumbers. The optical principle may be dispersive (e.g. grating monochromators),
interferometric or non-thermal (e.g. light-emitting diodes, laser diodes, and lasers). The instrument should be
provided with a diagnostic test system for testing photometric noise and reproducibility, wavelength or
wavenumber accuracy and wavelength or wavenumber precision (for scanning spectrophotometers).
The instrument should measure a sufficiently large sample volume or surface to eliminate any significant
influence of inhomogeneity derived from chemical composition or physical properties of the test sample. The
sample pathlength (sample thickness) in transmittance measurements should be optimized according to the
manufacturer's recommendation with respect to signal intensity for obtaining linearity and maximum
signal/noise ratio. In reflectance measurements, a quartz window or other appropriate material to eliminate
drying effects should preferably cover the interacting sample surface layer.
4.2 Appropriate milling or grinding device, for preparing the sample (if needed).
NOTE Changes in grinding or milling conditions can influence NIR measurements.
5 Calibration and initial validation
5.1 General
The instrument has to be calibrated before use. Because a number of different calibration systems can be
applied with NIR instruments, no specific procedure can be given for calibration.
For an explanation of methods for calibration development see, for example, Reference [17] and appropriate
manufacturers' manuals. For the validation, it is important to have a sufficient number of representative
samples, covering variations such as:
a) combinations and composition ranges of major and minor sample components;
b) seasonal, geographic and genetic effects on forages, feed raw materials and cereals;
c) processing techniques and conditions;
d) storage conditions;
e) sample and instrument temperature;
f) instrument variations (differences between instruments).
NOTE For a solid validation at least 20 samples are needed.
2 © ISO 2010 – All rights reserved

---------------------- Page: 7 ----------------------
ISO 12099:2010(E)
5.2 Reference methods
Internationally accepted reference methods for determination of moisture, fat, protein, and other constituents
and parameters should be used. See References [1] to [16] for examples.
The reference method used for calibration should be in statistical control, i.e. for any sample, the variability
should consist of random variations of a reproducible system. It is essential to know the precision of the
reference method.
5.3 Outliers
In many situations, statistical outliers are observed during calibration and validation. Outliers may be related to
NIR data (spectral outliers, hereafter referred to as x-outliers) or errors in reference data or samples with a
different relationship between reference data and NIR data (hereafter referred to as y-outliers)
(see Figures B.1 to B.5).
For the purpose of validation, samples are not to be regarded as outliers if:
a) they are within the working range of the constituents/parameters in the calibration(s);
b) they are within the spectral variation of the calibration samples, e.g. as estimated by Mahalanobis
distance;
c) the spectral residual is below a limit defined by the calibration process;
d) the prediction residual is below a limit defined by the calibration process.
If a sample appears as an outlier then it should be checked initially to see if it is an x-outlier. If it exceeds the
x-outlier limits defined for the calibration it should be removed. If it is not an x-outlier, then both the reference
value and the NIR predicted value should be checked. If these confirm the original values then the sample
should not be deleted and the validation statistics should include this sample. If the repeat values show that
either the original reference values or the NIR predicted ones were in error then the new values should be
used.
5.4 Validation of calibration models
5.4.1 General
Before use, calibration equations shall be validated locally on an independent test set that is representative of
the sample population to be analysed. For the determination of bias, at least 10 samples are needed; for the
determination of standard error of prediction (SEP, see 6.5) at least 20 samples are needed. Validation shall
be carried out for each sample type, constituent or parameter, and temperature. The calibration is valid only
for the variations, i.e. sample types, range and temperature, used in the validation.
Results obtained on the independent test set are plotted, reference against NIR, and residuals against
reference results, to give a visual impression of the performance of the calibration. The SEP is calculated
(see 6.5) and the residual plot of data corrected for mean systematic error (bias) is examined for outliers, i.e.
samples with a residual exceeding ± 3s .
SEP
If the validation process shows that the model cannot produce acceptable statistics, then it should not be used.
NOTE What is acceptable depends on such criteria as the performance of the reference method, the range covered,
and the purpose of the analysis and is up to the parties involved to decide.
The next step is to fit NIR data, y , and reference data, y , by linear regression (y = by + a) to produce
NIR ref ref NIR
statistics that describe the validation results.
© ISO 2010 – All rights reserved 3

---------------------- Page: 8 ----------------------
ISO 12099:2010(E)
5.4.2 Bias correction
The data are also examined for bias between the methods. If the difference between means of the NIR
predicted and reference values is significantly different from zero then this indicates that the calibration is
biased. A bias may be removed by adjusting the constant term (see 6.3) in the calibration equation.
5.4.3 Slope adjustment
If the slope, b, is significantly different from 1, the calibration is skewed.
Adjusting the slope or intercept of the calibration is generally not recommended unless the calibration is
applied to new types of samples or instruments. If a reinvestigation of the calibration does not detect outliers,
especially outliers with high leverage, it is preferable to expand the calibration set to include more samples.
However, if the slope is adjusted, the calibration should then be tested on a new independent test set.
5.4.4 Expansion of calibration set
If the accuracy of the calibration does not meet expectations, the calibration set should be expanded to
include more samples or a new calibration performed. In all cases, when a new calibration is developed on an
expanded calibration set, the validation process should be repeated on a new validation set. If necessary,
expansion of the calibration set should be repeated until acceptable results are obtained on a validation set.
5.5 Changes in measuring and instrument conditions
Unless additional calibration is performed, a local validation of an NIR method stating the accuracy of the
method can generally not be considered valid if the test conditions are changed.
For example, calibrations developed for a certain population of samples may not be valid for samples outside
this population, although the analyte concentration range is unchanged. A calibration developed on grass
silages from one area may not give the same accuracy on silages from another area if the genetic, growing
and processing parameters are different.
Changes in the sample presentation technique or the measuring conditions (e.g. temperature) not included in
the calibration set may also influence the analytical results.
Calibrations developed on a certain instrument cannot always be transferred directly to an identical instrument
operating under the same principle. It may be necessary to perform bias, slope or intercept adjustments to
calibration equations. In many cases, it is necessary to standardize the two instruments against each other
before calibration equations can be transferred (Reference [17]). Standardization procedures can be used to
transfer calibrations between instruments of different types provided that samples are measured in the same
way (reflectance, transmittance) and that the spectral region is common.
If the conditions are changed, a supplementary validation should be performed.
The calibrations should be checked whenever any major part of the instrument (optical system, detector) has
been changed or repaired.
6 Statistics for performance measurement
6.1 General
The performances of a prediction model shall be determined by a set of validation samples. This set consists
of samples which are independent of the calibration set. In a plant, it is new batches; in agriculture, it is a new
crop or a new experiment location.
This set of samples shall be carefully analysed following the reference methods. Care is essential in analysing
validation samples and the precision of these results is more important for the validation set than for the
samples used at the calibration phase.
4 © ISO 2010 – All rights reserved

---------------------- Page: 9 ----------------------
ISO 12099:2010(E)
The number of validation samples shall be at least 20 to compute the statistics with some confidence.
6.2 Plot the results
It is important to visualize the results in plots, i.e. reference vs predicted values or residuals vs predicted
values.
The residuals are defined as:

ey=−y (1)
ii i
where
y is the ith reference value;
i

y is the ith predicted value obtained when applying the multivariate NIR model.
i
The way the differences are calculated gives a positive bias when the predictions are too high and a negative
one when the predictions are too low compared to the reference values.
A plot of the data gives an immediate overview of the correlation, the bias, the slope, and the presence of
obvious outliers (see Figure 1).

Key
1 45° line (ideal line with bias, e = 0 and slope, b = 1) a intercept
2 45° line displaced by bias, e e bias
3 linear regression line with y -intercept, a y near infrared spectroscopy predicted values
ref NIRS
4 outliers y reference value
ref
NOTE The outliers have a strong influence on the calculation of the slope and should be removed if the results are to
be used for adjustments.
Figure 1 — Scatter plot for a validation set, y = f(a + by )
ref NIRS
© ISO 2010 – All rights reserved 5

---------------------- Page: 10 ----------------------
ISO 12099:2010(E)
6.3 The bias
Most of the time, a bias or systematic error is what is observed with NIR models. Bias can occur due to: new
samples of a type not previously seen by the model, drift of the instrument, drift in wet chemistry, changes in
the process, and changes in the sample preparation.
With the number of independent samples, n, the bias (or offset) is the mean difference, e , and can be defined
as:
n
1
ee= (2)
i

n
i=1
where e is the residual as defined in Equation (1), or
i
nn
⎛⎞
1  
⎜⎟
ey=−y=y−y (3)
∑∑ii
⎜⎟
n
ii==11
⎝⎠
where
y is the ith reference value;
i

y is the ith predicted value obtained when applying the multivariate NIR model;
i
and

y is the mean of the predicted values;
y is the mean of the reference values.
The significance of the bias is checked by a t-test. The calculation of the bias confidence limits (BCLs), T ,
b
determines the limits for accepting or rejecting equation performance on the small set of samples chosen from
the new population.
ts
SEP
1/−α2
()
T =± (4)
b
n
where
α is the probability of making a type I error;
t is the appropriate Student t-value for a two-tailed test with degrees of freedom associated with
SEP and the selected probability of a type I error (see Table 1);
n is the number of independent samples;
s is the standard error of prediction (see 6.5).
SEP
EXAMPLE With n = 20, and s = 1, the BCLs are
SEP
2, 09 × 1
T =± =± 0, 48 (5)
b
20
6 © ISO 2010 – All rights reserved

---------------------- Page: 11 ----------------------
ISO 12099:2010(E)
This means that the bias tested with 20 samples must be higher than 48 % of the standard error of prediction
to be considered as different from zero.
Table 1 — Values of the t-distribution with a probability, α = 0,05 = 5 %
n t n t n t n t
5 2,57 11 2,20 17 2,11 50 2,01
6 2,45 12 2,18 18 2,10 75 1,99
7 2,36 13 2,16 19 2,09 100 1,98
8 2,31 14 2,14 20 2,09 200 1,97
9 2,26 15 2,13 30 2,04 500 1,96
10 2,23 16 2,12 40 2,02 1 000 1,96
1)
NOTE The Excel function TINV can be used.

6.4 Root mean square error of prediction (RMSEP)
The RMSEP, s , (C.3.6) is expressed mathematically as
RMSEP
n
2
e
∑ i
i=1
s = (6)
RMSEP
n
where
e is the residual of the ith sample;
i
n is the number of independent samples.
This value can be compared with SEC (C.3.3) and SECV (C.3.4).
RMSEP includes the random error (SEP) and the systematic error (bias). It also includes the error of the
reference methods (as do SEC and SECV).
(1n −)
22
s =+se (7)
RMSEP SEP
n
where
n is the number of independent samples;
s is the standard error of prediction (see 6.5);
SEP
e is the bias or systematic error.
There is no direct test for RMSEP. This is the reason for separating the systematic error, bias or e , and the
random error, SEP or s .
SEP

1) Excel is the trade name of a product supplied by Microsoft. This information is given for the convenience of users of
this document and does not constitute an endorsement by ISO of the product named. Equivalent products may be used if
they can be shown to lead to comparable results.
© ISO 2010 – All rights reserved 7

---------------------- Page: 12 ----------------------
ISO 12099:2010(E)
6.5 Standard error of prediction (SEP)
The SEP, s , or the standard deviation of the residuals, which expresses the accuracy of routine NIR results
SEP
corrected for the mean difference (bias) between routine NIR and reference method, can be calculated by
using Equation (8):
n
2
ee−
()
∑ i
i=1
s = (8)
SEP
n −1
where
n is the number of independent samples;
e is the residual of the ith sample;
i
e is the bias or systematic error.
The SEP should be related to the SEC (C.3.3) or SECV (C.3.4) to check the validity of the calibration model
for the selected validation set.
The unexplained error confidence limits (UECLs), T , are calculated from an F-test (ratio of 2 variances)
UE
(see Reference [19] and Table 2).
Ts= F (9)
UE SEC (α:vM, )
where
s is the standard error of calibration (C.3.3);
SEC
α is the probability of making a type I error;
ν = n − 1 is the numerator degrees of freedom associated with SEP of the test set, in which n is
the number of samples in the validation process;
M = n − p − 1 is the denominator degrees of freedom associated with SEC (standard error of
c
calibration);
in which
n is the number of calibration samples,
c
p is the number of terms or PLS factors of the model.
NOTE 1 SEC can be replaced by SECV which is a better statistic than SEC; very often SEC is too optimistic,
s > s .
SECV SEC
EXAMPLE With n = 20, α = 0,05, M = 100, and s = 1,
SEC
T = 1, 30 (10)
UE
This means that, with 20 samples, a SEP can be accepted that is up to 30 % larger than the SEC.
2)
NOTE 2 The Excel function FINV can be used.

2) Excel is the trade name of a product supplied by Microsoft. This information is given for the convenience of users of
this document and does not constitute an endorsement by ISO of the product named. Equivalent products may be used if
they can be shown to lead to the same results.
8 © ISO 2010 – All rights reserved

---------------------- Page: 13 ----------------------
ISO 12099:2010(E)
The F-test cannot be used to compare two calibrations on the same validation set. It needs (as here) two
independent sets to work. Another test is required to compare two or more models on the same data set.
Table 2 — F-values and square root of the F-values as a function of the degrees of freedom of the
numerator associated with SEP and of the denominator associated with SEC
[see definitions under Equation (9)]
F √[F ]
(α: ν, M) (α: ν, Μ)
Degrees Degrees of freedom (SEC) Degrees Degrees of freedom (SEC)
of of
50 100 200 500 1 000 50 100 200 500 100
freedom freedom
(SEP) (SEP)
5 2,40 2,31 2,26 2,23 2,22 5 1,55 1,52 1,50 1,49 1,49
6 2,29 2,19 2,14 2,12 2,11 6 1,51 1,48 1,46 1,45 1,45
7 2,20 2,10 2,06 2,03 2,02 7 1,48 1,45 1,43 1,42 1,42
8 2,13 2,03 1,98 1,96 1,95 8 1,46 1,43 1,41 1,40 1,40
9 2,07 1,97 1,93 1,90 1,89 9 1,44 1,41 1,39 1,38 1,37
10 2,03 1,93 1,88 1,85 1,84 10 1,42 1,39 1,37 1,36 1,36
11 1,99 1,89 1,84 1,81 1,80 11 1,41 1,37 1,36 1,34 1,34
12 1,95 1,85 1,80 1,77 1,76 12 1,40 1,36 1,34 1,33 1,33
13 1,92 1,82 1,77 1,74 1,73 13 1,39 1,35 1,33 1,32 1,32
14 1,89 1,79 1,74 1,71 1,70 14 1,38 1,34 1,32 1,31 1,30
15 1,87 1,77 1,72 1,69 1,68 15 1,37 1,33 1,31 1,30 1,29
16 1,85 1,75 1,69 1,66 1,65 16 1,36 1,32 1,30 1,29 1,29
17 1,83 1,73 1,67 1,64 1,63 17 1,35 1,31 1,29 1,28 1,28
18 1,81 1,71 1,66 1,62 1,61 18 1,30 1,31 1,29 1,27 1,27
19 1,80 1,69 1,64 1,61 1,60 19 1,34 1,30 1,28 1,27 1,26
29 1,69 1,58 1,52 1,49 1,48 29 1,30 1,26 1,23 1,22 1,22
49 1,60 1,48 1,42 1,38 1,37 49 1,27 1,22 1,19 1,17 1,17
99 1,53 1,39 1,32 1,28 1,26 99 1,24 1,18 1,15 1,13 1,12
199 1,48 1,34 1,26 1,21 1,19 199 1,22 1,16 1,12 1,10 1,09
499 1,46 1,31 1,22 1,16 1,13 499 1,21 1,14 1,11 1,08 1,07
999 1,45 1,30 1,21 1,14 1,11 999 1,20 1,14 1,10 1,07 1,05

6.6 Slope

The slope, b, of the simple regression y = a + by is often reported in NIR publications.
Notice that the slope must be calculated with the reference values as the dependent variable and the
predicted NIR values as the independent variable, if the calculated slope is intended to be used for adjustment
of NIR results (like in the case of the inverse multivariate regression used to build the prediction model).
© ISO 2010 – All rights reserved 9

---------------------- Page: 14 ----------------------
ISO 12099:2010(E)
From the least squares fitting, the slope is calculated as:

s
yy
b = (11)
2
s

where

s is the covariance between reference and predicted values;
yy
2
s is the variance of the n predicted values.
ˆ
y
The intercept is calculated as:
ay=−byˆ (12)
where

y is the mean of the predicted values;
y is the mean of the reference values;
b is the slope.
As for the bias, a t-test can be calculated to check the hypothesis that b =1
2
sn(1−)

tb=− 1 (13)
obs
2
s
res
where
n is the number of independent samples;
2
s is the variance of the n predicted values;

s is the residual standard deviation as defined in Equation (14).
res
n
2
ya−+byˆ
()
∑ii
i=1
s = (14)
res
n − 2
in which
n is the number of independent samples,
a is the intercept Equation (12),
b is the slope Equation (11),
y is the ith reference value,
i

y is the ith predicted value obtained when applying the multivariate NIR model.
i
10 © ISO 2010 – All rights reserved

---------------------- Page: 15 ----------------------
ISO 12099:2010(E)
(RSD is like the SEP when the predicted values are corrected for slope and intercept. Do not confuse bias and
intercept — see also Figure 1.) The bias equals the intercept only when the slope is exactly one.
The slope, b, is considered as different from 1 when
ttW
obs (1−α / 2)
where
t is the observed t-value, calculated according to Equation (13);
obs
t is the t-value obtained from Table 1 for a probability of α = 0,05 (5 %).
(1-α/2)
Too narrow a range or an uneven distribution leads to inappropriate correction of the slope even when the
SEP is correct. The slope can only be adjusted when the validation set covers a large part of the calibration
range.
EXAMPLE For n = 20 samples with a residual standard deviation [Equation (14)] of 1, a standard deviation of the
predicted values of s = 2 and a calculated slope of b = 1,2, the observed t value is 1,7 and then the slope
...

NORME ISO
INTERNATIONALE 12099
Première édition
2010-06-15


Aliments des animaux, céréales et
produits de mouture des céréales —
Lignes directrices pour l'application de la
spectrométrie dans le proche infrarouge
Animal feeding stuffs, cereals and milled cereal products — Guidelines
for the application of near infrared spectrometry




Numéro de référence
ISO 12099:2010(F)
©
ISO 2010

---------------------- Page: 1 ----------------------
ISO 12099:2010(F)
PDF – Exonération de responsabilité
Le présent fichier PDF peut contenir des polices de caractères intégrées. Conformément aux conditions de licence d'Adobe, ce fichier
peut être imprimé ou visualisé, mais ne doit pas être modifié à moins que l'ordinateur employé à cet effet ne bénéficie d'une licence
autorisant l'utilisation de ces polices et que celles-ci y soient installées. Lors du téléchargement de ce fichier, les parties concernées
acceptent de fait la responsabilité de ne pas enfreindre les conditions de licence d'Adobe. Le Secrétariat central de l'ISO décline toute
responsabilité en la matière.
Adobe est une marque déposée d'Adobe Systems Incorporated.
Les détails relatifs aux produits logiciels utilisés pour la création du présent fichier PDF sont disponibles dans la rubrique General Info
du fichier; les paramètres de création PDF ont été optimisés pour l'impression. Toutes les mesures ont été prises pour garantir
l'exploitation de ce fichier par les comités membres de l'ISO. Dans le cas peu probable où surviendrait un problème d'utilisation,
veuillez en informer le Secrétariat central à l'adresse donnée ci-dessous.


DOCUMENT PROTÉGÉ PAR COPYRIGHT


©  ISO 2010
Droits de reproduction réservés. Sauf prescription différente, aucune partie de cette publication ne peut être reproduite ni utilisée sous
quelque forme que ce soit et par aucun procédé, électronique ou mécanique, y compris la photocopie et les microfilms, sans l'accord écrit
de l'ISO à l'adresse ci-après ou du comité membre de l'ISO dans le pays du demandeur.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Publié en Suisse

ii © ISO 2010 – Tous droits réservés

---------------------- Page: 2 ----------------------
ISO 12099:2010(F)
Sommaire Page
Avant-propos .iv
Introduction.v
1 Domaine d'application .1
2 Termes et définitions .1
3 Principe.2
4 Appareillage .2
5 Étalonnage et validation initiale.2
6 Statistiques pour le mesurage des performances.5
7 Échantillonnage.12
8 Mode opératoire.12
9 Vérification de la stabilité de l'appareil .13
10 Contrôle de performance de l'étalonnage.13
11 Fidélité et exactitude.15
12 Rapport d'essai.16
Annexe A (informative) Lignes directrices pour des normes NIR spécifiques .17
Annexe B (informative) Exemples de figures.18
Annexe C (informative) Termes et définitions supplémentaires.24
Bibliographie.31

© ISO 2010 – Tous droits réservés iii

---------------------- Page: 3 ----------------------
ISO 12099:2010(F)
Avant-propos
L'ISO (Organisation internationale de normalisation) est une fédération mondiale d'organismes nationaux de
normalisation (comités membres de l'ISO). L'élaboration des Normes internationales est en général confiée
aux comités techniques de l'ISO. Chaque comité membre intéressé par une étude a le droit de faire partie du
comité technique créé à cet effet. Les organisations internationales, gouvernementales et non
gouvernementales, en liaison avec l'ISO participent également aux travaux. L'ISO collabore étroitement avec
la Commission électrotechnique internationale (CEI) en ce qui concerne la normalisation électrotechnique.
Les Normes internationales sont rédigées conformément aux règles données dans les Directives ISO/CEI,
Partie 2.
La tâche principale des comités techniques est d'élaborer les Normes internationales. Les projets de Normes
internationales adoptés par les comités techniques sont soumis aux comités membres pour vote. Leur
publication comme Normes internationales requiert l'approbation de 75 % au moins des comités membres
votants.
L'attention est appelée sur le fait que certains des éléments du présent document peuvent faire l'objet de
droits de propriété intellectuelle ou de droits analogues. L'ISO ne saurait être tenue pour responsable de ne
pas avoir identifié de tels droits de propriété et averti de leur existence.
L'ISO 12099 a été élaborée par le comité technique CEN/TC 327, Aliments des animaux — Méthodes
d'échantillonnage et d'analyse, du Comité européen de normalisation (CEN) en collaboration avec le comité
technique ISO/TC 34, Produits alimentaires, sous-comité SC 10, Aliments des animaux, conformément à
l'Accord de coopération technique entre l'ISO et le CEN (Accord de Vienne).

iv © ISO 2010 – Tous droits réservés

---------------------- Page: 4 ----------------------
ISO 12099:2010(F)
Introduction
[15]
La présente Norme internationale a été rédigée en utilisant pour base l'ISO 21543⏐FIL 201 , qui a été
élaborée par le comité technique ISO/TC 34, Produits alimentaires, sous-comité SC 5, Lait et produits laitiers,
et par la Fédération internationale de laiterie (FIL).

© ISO 2010 – Tous droits réservés v

---------------------- Page: 5 ----------------------
NORME INTERNATIONALE ISO 12099:2010(F)

Aliments des animaux, céréales et produits de mouture des
céréales — Lignes directrices pour l'application de la
spectrométrie dans le proche infrarouge
1 Domaine d'application
La présente Norme internationale fournit des lignes directrices pour la détermination par spectrométrie dans le
proche infrarouge de constituants tels que l'eau, les matières grasses, les protéines, l'amidon et la cellulose
brute, ainsi que des paramètres tels que la digestibilité des aliments pour animaux, des céréales et des
produits céréaliers moulus.
Les déterminations sont basées sur des mesurages spectrométriques dans le domaine du proche infrarouge.
2 Termes et définitions
Pour les besoins du présent document, les termes et définitions suivants s'appliquent.
2.1
appareil proche infrarouge
appareil NIR
appareillage qui, lorsqu'il est utilisé dans des conditions spécifiées, permet de prédire les teneurs en
constituants (2.3) et les paramètres technologiques (2.4) dans une matrice, par le biais de relations avec
les absorptions dans la gamme du proche infrarouge
NOTE Dans le cadre de la présente Norme internationale, les matrices sont les aliments des animaux, les céréales
et les produits de mouture des céréales.
2.2
aliments des animaux
tout produit ou substance, y compris les additifs, traité, traité en partie ou non traité, destiné à être utilisé pour
l'alimentation des animaux par ingestion
EXEMPLES Matières premières, fourrage, farine animale, aliments mélangés pour bétail et autres produits finaux,
aliments pour animaux domestiques.
2.3
teneur en constituants
fraction massique des substances déterminée à l'aide de la méthode chimique appropriée, normalisée ou
validée
NOTE 1 La fraction massique est souvent exprimée en pourcentage.
NOTE 2 Les exemples de constituants déterminés comprennent l'eau, les matières grasses, les protéines, la cellulose
brute, les fibres détergentes neutres et les fibres détergentes acides. Pour les méthodes appropriées, voir les
Références [1] à [16] par exemple.
© ISO 2010 – Tous droits réservés 1

---------------------- Page: 6 ----------------------
ISO 12099:2010(F)
2.4
paramètre technologique
propriété ou fonctionnalité d'une matrice qui peut être déterminée à l'aide de la (des) méthode(s)
appropriée(s), normalisée(s) ou validée(s)
EXEMPLE Digestibilité.
NOTE 1 Dans le cadre de la présente Norme internationale, les matrices sont les aliments des animaux, les céréales
et les produits de mouture des céréales.
NOTE 2 Il est possible de développer et de valider des méthodes NIR pour des paramètres et des matrices autres que
ceux énumérés, sous réserve que le mode opératoire spécifié dans la présente Norme internationale soit respecté. Les
unités de mesure des paramètres déterminés doivent suivre les unités utilisées dans les méthodes de référence.
3 Principe
Les données spectrales dans le proche infrarouge (NIR) sont collectées et transformées en concentrations de
constituants ou de paramètres par des modèles d'étalonnage réalisés sur des échantillons représentatifs des
produits concernés.
4 Appareillage
4.1 Appareils proche infrarouge, basés sur les mesurages en réflexion diffuse ou en transmission
−1
couvrant le domaine de longueurs d'onde du proche infrarouge de 770 nm à 2 500 nm (12 900 cm à
−1
4 000 cm ), ou segments de ce dernier ou à des longueurs d'onde ou des nombres d'onde sélectionnés. Le
principe optique peut être dispersif (monochromateurs à réseau par exemple), interférométrique ou non
thermique (par exemple diodes électroluminescentes, diodes laser et lasers). Il convient que l'appareil soit
doté d'un système effectuant des tests diagnostiques pour évaluer le bruit photométrique et la reproductibilité,
ainsi que l'exactitude et la fidélité des longueurs d'onde ou des nombres d'onde (pour les spectrophotomètres
à balayage).
Il convient que l'appareil mesure un volume ou une surface d'échantillon suffisamment important(e) pour
éliminer toute influence significative d'un manque d'homogénéité découlant de la composition chimique ou
des propriétés physiques de l'échantillon pour essai. Dans les mesurages en transmission, pour obtenir la
linéarité et un rapport signal/bruit maximal, il convient d'optimiser l'intensité du signal en adaptant le chemin
optique à chaque type d'échantillon (épaisseur des échantillons) conformément aux recommandations du
fabricant. Dans les mesurages en réflexion, il convient de préférence qu'une fenêtre en quartz ou autre
matériau adéquat pour éliminer les effets de séchage recouvre la surface d'interaction de l'échantillon.
4.2 Dispositif de mouture ou de broyage approprié, pour la préparation de l'échantillon (si nécessaire).
NOTE Des changements dans les conditions de mouture ou de broyage peuvent influencer les mesurages NIR.
5 Étalonnage et validation initiale
5.1 Généralités
L'appareil doit être étalonné avant d'être utilisé. Dans la mesure où un grand nombre de systèmes différents
peut être utilisé pour l'étalonnage des appareils NIR, aucun mode opératoire spécifique ne peut être indiqué
pour l'étalonnage.
2 © ISO 2010 – Tous droits réservés

---------------------- Page: 7 ----------------------
ISO 12099:2010(F)
Pour une explication des méthodes de développement de l'étalonnage se reporter, par exemple, à la
Référence [17] et aux manuels d'instruction des fabricants respectifs. Pour la validation, il est important de
disposer d'un nombre suffisant d'échantillons représentatifs, couvrant des variations telles que:
a) les combinaisons et les gammes de concentration des constituants principaux et secondaires de
l'échantillon;
b) les effets saisonniers, géographiques et génétiques sur les fourrages, les matières premières
alimentaires et les céréales;
c) les techniques et les conditions de transformation;
d) les conditions de stockage;
e) la température de l'échantillon et de l'appareil;
f) les variations des appareils (différence entre appareils).
NOTE Pour garantir une bonne validation, au moins 20 échantillons sont nécessaires.
5.2 Méthodes de référence
Il convient d'utiliser des méthodes de référence reconnues à l'échelle internationale pour la détermination de
la teneur en eau, en matières grasses, en protéines et autres constituants et paramètres. Voir les
Références [1] à [16] pour des exemples.
Pour l'étalonnage, il convient d'utiliser la méthode de référence en contrôle statistique, c'est-à-dire que, pour
chaque échantillon, il convient que la variabilité soit composée des variations aléatoires d'un système
reproductible. Il est indispensable de connaître la fidélité de la méthode de référence.
5.3 Valeurs aberrantes
Dans de nombreuses situations, des valeurs statistiques aberrantes sont observées au cours de l'étalonnage
et de la validation. Les valeurs aberrantes peuvent porter sur des données NIR (valeurs aberrantes de spectre,
désignées ci-après valeurs aberrantes x) ou des erreurs dans les données ou les échantillons de référence
avec une relation différente entre les données de référence et les données NIR (désignées ci-après valeurs
aberrantes y) (voir les Figures B.1 à B.5).
Pour les besoins de la validation, les échantillons ne sont pas considérés comme des valeurs aberrantes:
a) s'ils se situent dans la plage de travail des constituants/paramètres de l'étalonnage ou des étalonnages;
b) s'ils se situent dans les limites de la variation spectrale des échantillons d'étalonnage, telles qu'estimées,
par exemple, par la distance de Mahalanobis;
c) si le résidu spectral se situe en dessous d'une limite définie par le procédé d'étalonnage;
d) si le résidu de prédiction se situe en dessous d'une limite définie par le procédé d'étalonnage.
S'il apparaît qu'un échantillon est une valeur aberrante, il convient alors de le vérifier initialement pour voir s'il
s'agit d'une valeur aberrante x. Si cette valeur dépasse les limites des valeurs aberrantes x définies pour
l'étalonnage, il convient de l'éliminer. S'il ne s'agit pas d'une valeur aberrante x, il convient alors de vérifier la
valeur de référence et la valeur NIR prédite. Si cette vérification confirme les valeurs initiales, il convient alors
de ne pas supprimer l'échantillon et de l'inclure dans les valeurs statistiques de validation. Si les nouvelles
valeurs indiquent que les valeurs de référence initiales ou les valeurs NIR prédites sont erronées, il convient
alors d'utiliser ces nouvelles valeurs.
© ISO 2010 – Tous droits réservés 3

---------------------- Page: 8 ----------------------
ISO 12099:2010(F)
5.4 Validation des modèles d'étalonnage
5.4.1 Généralités
Avant d'être utilisées, les équations d'étalonnage doivent être validées localement sur un ensemble
d'échantillons indépendant représentatif de la population d'échantillons à analyser. Au moins 10 échantillons
sont nécessaires pour la détermination du biais; au moins 20 échantillons sont nécessaires pour la
détermination de l'erreur type de prédiction (SEP, voir 6.5). La validation doit être effectuée pour chaque type
d'échantillon, constituant ou paramètre et température. L'étalonnage n'est valable que pour les variations,
c'est-à-dire les types d'échantillons, la gamme et la température utilisés lors de la validation.
Les résultats obtenus sur l'ensemble de validation indépendant sont reportés sur un graphique, référence en
fonction de NIR et résidus en fonction des résultats de référence, pour représenter visuellement les
performances de l'étalonnage. La SEP est calculée (voir 6.5) et les valeurs aberrantes, telles que des
échantillons avec un résidu dépassant ± 3 s , sont recherchées sur le graphique résiduel des données
SEP
corrigées pour l'erreur systématique moyenne (biais).
Si le processus de validation montre que le modèle ne peut pas produire de valeurs statistiques acceptables,
il convient alors de ne pas l'utiliser.
NOTE Ce qui est acceptable dépend de critères tels que les performances de la méthode de référence, la plage
couverte et le but de l'analyse et il appartient aux parties concernées d'en décider.
L'étape suivante consiste à ajuster les données NIR, y , et les données de référence, y , par régression
NIR ref
linéaire (y = by + a) pour produire des valeurs statistiques qui décrivent les résultats de validation.
ref NIR
5.4.2 Correction du biais
Un biais entre les méthodes est également recherché dans les données. Si la différence entre les moyennes
des valeurs NIR prédites et des valeurs de référence est significativement différente de zéro, cela indique que
l'étalonnage est biaisé. Il est possible de supprimer un biais en ajustant la constante (voir 6.3) de l'équation
d'étalonnage.
5.4.3 Ajustement de la pente
Si la pente, b, est significativement différente de 1, l'étalonnage est faussé.
L'ajustement de la pente et de l'ordonnée à l'origine de l'étalonnage n'est généralement pas recommandé,
sauf si l'étalonnage est appliqué à de nouveaux types d'échantillons ou d'appareils. Si une nouvelle étude de
l'étalonnage ne permet pas de détecter de valeurs aberrantes, notamment des valeurs aberrantes «de levier
élevé», il est préférable d'élargir l'ensemble d'étalonnage pour inclure davantage d'échantillons. Cependant, si
la pente est ajustée, il convient alors de tester l'étalonnage sur un nouvel ensemble de validation indépendant.
5.4.4 Élargissement de l'ensemble d'étalonnage
Si l'exactitude de l'étalonnage ne répond pas aux attentes, il convient d'élargir l'ensemble d'étalonnage pour
inclure un plus grand nombre d'échantillons ou de réaliser un nouvel étalonnage. Dans tous les cas, lorsqu'un
nouvel étalonnage est développé sur un ensemble d'étalonnage élargi, il convient de répéter le mode
opératoire de validation sur un nouvel ensemble de validation indépendant. Si nécessaire, il convient de
répéter l'élargissement de l'ensemble d'étalonnage jusqu'à ce que des résultats acceptables soient obtenus
sur un ensemble de validation indépendant.
5.5 Changements des conditions de mesurage et d'utilisation des appareils
Sauf si un étalonnage supplémentaire est effectué, une validation locale d'une méthode NIR indiquant
l'exactitude de la méthode ne peut généralement pas être considérée comme valide si les conditions d'essai
ont changé.
4 © ISO 2010 – Tous droits réservés

---------------------- Page: 9 ----------------------
ISO 12099:2010(F)
Par exemple, les étalonnages développés pour une certaine population d'échantillons ne peuvent pas être
valides pour des échantillons en dehors de cette population, même si la gamme de concentration de l'analyte
n'a pas changé. Il est possible que l'étalonnage développé sur des ensilages provenant d'une zone ne donne
pas la même exactitude sur des ensilages provenant d'une autre zone si les paramètres génétiques, de
croissance et de traitement sont différents.
Des changements de technique de présentation des échantillons ou de conditions de mesurage (par exemple
la température) non compris dans l'ensemble d'étalonnage peuvent également influencer les résultats
analytiques.
Les étalonnages développés sur un appareil spécifique ne peuvent pas toujours être transférés directement
vers un appareil identique fonctionnant selon le même principe. Des ajustements du biais, de la pente ou de
l'ordonnée à l'origine dans les équations d'étalonnage peuvent s'avérer nécessaires. Dans de nombreux cas,
il est nécessaire de normaliser les deux appareils l'un par rapport à l'autre avant de pouvoir transférer les
équations d'étalonnage (Référence [17]). Les modes opératoires de normalisation peuvent être utilisés pour
transférer les étalonnages entre des appareils de types différents à condition que les échantillons soient
mesurés de la même manière (réflexion, transmission) et que le domaine spectral soit commun.
Si les conditions ont changé, il convient de réaliser une validation supplémentaire.
Il convient de vérifier les étalonnages dès qu'une pièce importante de l'appareil (système optique, détecteur) a
été remplacée ou réparée.
6 Statistiques pour le mesurage des performances
6.1 Généralités
Les performances d'un modèle de prédiction doivent être déterminées par un ensemble d'échantillons de
validation. Cet ensemble est composé d'échantillons indépendants de l'ensemble d'étalonnage. Dans une
usine, il s'agit de nouveaux lots; en agriculture, il s'agit d'une nouvelle récolte ou d'un nouveau lieu
d'expérimentation.
Cet ensemble d'échantillons doit être soigneusement analysé suivant les méthodes de référence. Il est
indispensable d'apporter le plus grand soin à l'analyse des échantillons de validation et la fidélité des résultats
est plus importante pour l'ensemble de validation que pour les échantillons utilisés lors de la phase
d'étalonnage.
Le nombre d'échantillons de validation doit être au moins égal à 20 pour calculer les valeurs statistiques avec
une certaine confiance.
6.2 Représentation graphique des résultats
Il est important de visualiser les résultats sous forme de graphiques, c'est-à-dire de valeurs de référence en
fonction de valeurs prédites ou de résidus en fonction de valeurs prédites.
Les résidus sous définis par:

ey=−y (1)
ii i

ème
y est la i valeur de référence;
i

ème
y est la i valeur prédite obtenue en utilisant le modèle NIR à plusieurs variables.
i
© ISO 2010 – Tous droits réservés 5

---------------------- Page: 10 ----------------------
ISO 12099:2010(F)
La manière dont les différences sont calculées donne un biais positif lorsque les prédictions sont trop élevées
et un biais négatif lorsque les prédictions sont trop faibles par rapport aux valeurs de référence.
Une représentation graphique des données fournit un aperçu immédiat de la corrélation, du biais, de la pente
et de la présence de valeurs aberrantes évidentes (voir Figure 1).

Légende
1 droite à 45° (droite idéale avec un biais, e = 0 et une pente, b = 1) a ordonnée à l'origine
2 droite à 45° décalée d'un biais, e e biais
3 droite de régression linéaire d'ordonnée à l'origine y , a y valeurs spectrométriques prédites
ref NIRS
4 valeurs aberrantes dans le proche infrarouge
 y valeurs de référence
ref
NOTE Étant donné que les valeurs aberrantes ont une forte influence sur le calcul de la pente, il convient de les
éliminer si les résultats sont destinés à être utilisés pour des ajustements.
Figure 1 — Diagramme de dispersion pour un ensemble de validation, y = f(a + by )
ref NIRS
6.3 Biais
Avec les modèles NIR, on observe la plupart du temps un biais ou une erreur systématique. Un biais peut
avoir plusieurs causes: nouveaux échantillons d'un type totalement inconnu du modèle, dérive de l'appareil,
dérive dans la chimie par voie humide, changements dans le mode opératoire et changements dans la
préparation des échantillons.
6 © ISO 2010 – Tous droits réservés

---------------------- Page: 11 ----------------------
ISO 12099:2010(F)
n étant le nombre d'échantillons indépendants, le biais (ou décalage) est la différence moyenne, e , et peut
être défini par:
n
1
ee= (2)
i

n
i=1
où e est le résidu tel que défini dans l'Équation (1), ou
i
nn
⎛⎞
1  
⎜⎟
ey=−y=y−y (3)
∑∑ii
⎜⎟
n
ii==11
⎝⎠

ème
y est la i valeur de référence;
i

ème
y est la i valeur prédite obtenue en utilisant le modèle NIR à plusieurs variables;
i
et

y est la moyenne des valeurs prédites;
y est la moyenne des valeurs de référence.
L'importance du biais est vérifiée par un test t. Le calcul des limites de confiance du biais (BCL), T , détermine
b
les limites d'acceptation ou de rejet des performances des équations sur le petit ensemble d'échantillons
choisi dans la nouvelle population.
ts
(1−α / 2) SEP
T =± (4)
b
n

α est la probabilité d'erreur de type I;
t est la valeur t de Student appropriée pour un test bilatéral avec des degrés de liberté associés à la
SEP et la probabilité choisie d'une erreur de type I (voir Tableau 1);
n est le nombre d'échantillons indépendants;
s est l'erreur type de prédiction (voir 6.5).
SEP
EXEMPLE Avec n = 20 et s = 1, les limites de confiance du biais sont
SEP
2,09 ×1
T =± =± 0,48 (5)
b
20
Cela signifie que le biais soumis à essai avec 20 échantillons doit être supérieur à 48 % de l'erreur type de
prédiction pour être considéré comme différent de zéro.
© ISO 2010 – Tous droits réservés 7

---------------------- Page: 12 ----------------------
ISO 12099:2010(F)
Tableau 1 — Valeurs t de la distribution de Student avec une probabilité α = 0,05 = 5 %
n t n t n t n t
5 2,57 11 2,20 17 2,11 50 2,01
6 2,45 12 2,18 18 2,10 75 1,99
7 2,36 13 2,16 19 2,09 100 1,98
8 2,31 14 2,14 20 2,09 200 1,97
9 2,26 15 2,13 30 2,04 500 1,96
10 2,23 16 2,12 40 2,02 1 000 1,96
1)
NOTE La fonction TINV dans Excel peut être utilisée.

6.4 Erreur quadratique moyenne de prédiction (RMSEP)
La RMSEP, s , (C.3.6), est exprimée mathématiquement par:
RMSEP
n
2
e
∑ i
i=1
s = (6)
RMSEP
n

ème
e est le résidu du i échantillon;
i
n est le nombre d'échantillons indépendants.
Cette valeur peut être comparée à SEC (C.3.3) et SECV (C.3.4).
RMSEP comprend l'erreur aléatoire (SEP) et l'erreur systématique (biais). Elle comprend également l'erreur
des méthodes de référence (tout comme SEC et SECV).
(1n −)
22
s =+se (7)
RMSEP SEP
n

n est le nombre d'échantillons indépendants;
s est l'erreur type de prédiction (voir 6.5);
SEP
e est le biais ou erreur systématique.
Il n'existe aucun test direct pour RMSEP. C'est la raison pour laquelle on sépare l'erreur systématique, biais
ou e, de l'erreur aléatoire, SEP ou s .
SEP

1) Excel est l'appellation commerciale d'un produit fourni par Microsoft. Cette information est donnée à l'intention des
utilisateurs de la présente Norme internationale et ne signifie nullement que l'ISO approuve ou recommande l'emploi
exclusif du produit ainsi désigné. Des produits équivalents peuvent être utilisés s'il est démontré qu'ils conduisent aux
mêmes résultats.
8 © ISO 2010 – Tous droits réservés

---------------------- Page: 13 ----------------------
ISO 12099:2010(F)
6.5 Erreur type de prédiction (SEP)
La SEP, s , ou l'écart-type des résidus, qui exprime l'exactitude des résultats de routine NIR corrigés de
SEP
l'écart moyen (biais) entre les méthodes de routine NIR et de référence, peut être calculée à l'aide de
l'Équation (8):
n
2
ee−
()
∑ i
i=1
s = (8)
SEP
n −1

n est le nombre d'échantillons indépendants;
ème
e est le résidu du i échantillon;
i
e est le biais ou erreur systématique.
Il convient que la SEP soit reliée à la SEC (C.3.3) ou la SECV (C.3.4) pour vérifier la validité du modèle
d'étalonnage pour l'ensemble de validation choisi.
Les limites de confiance d'erreur inexpliquée (UECL), T , sont calculées à partir d'un test F (rapport de
UE
2 variances) (voir Référence [19] et Tableau 2).
Ts= F (9)
UE SEC (α:vM, )

s est l'erreur type d'étalonnage (C.3.3);
SEC
α est la probabilité d'erreur de type I;
ν = n − 1 degrés de liberté pour le numérateur associé à la SEP de l'ensemble de validation, dans
lequel n est le nombre d'échantillons dans le processus de validation;
M = n − p − 1 degrés de liberté pour le dénominateur associé à la SEC (erreur type d'étalonnage);
c
dans lequel
n est le nombre d'échantillons d'étalonnage,
c
p est le nombre de termes ou de facteurs PLS du modèle.
NOTE 1 SEC peut être remplacée par SECV qui est une meilleure valeur statistique que SEC; très souvent SEC est
trop optimiste, s > s .
SECV SEC
EXEMPLE Avec n = 20, α = 0,05, M = 100 et s = 1,
SEC
T = 1, 30 (10)
UE
Cela signifie qu'avec 20 échantillons, il est possible d'accepter une SEP jusqu'à 30 % supérieure à la SEC.
2)
NOTE 2 La fonction FINV dans Excel peut être utilisée.

2) Excel est l'appellation commerciale d'un produit fourni par Microsoft. Cette information est donnée à l'intention des
utilisateurs de la présente Norme internationale et ne signifie nullement que l'ISO approuve ou recommande l'emploi
exclusif du produit ainsi désigné. Des produits équivalents peuvent être utilisés s'il est démontré qu'ils conduisent aux
mêmes résultats.
© ISO 2010 – Tous droits réservés 9

---------------------- Page: 14 ----------------------
ISO 12099:2010(F)
Le test F ne peut pas être utilisé pour comparer deux étalonnages sur le même ensemble de validation. Il
nécessite (comme dans ce cas) deux
...

Questions, Comments and Discussion

Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.