WO2016034189A1

WO2016034189A1 - Method of adaptive sampling in a spectrophotometer and a spectrophotometer implementing the method

Info

Publication number: WO2016034189A1
Application number: PCT/EP2014/068466
Authority: WO
Inventors: Bengt Nordlund
Original assignee: Foss Analytical AS
Current assignee: Foss Analytical AS
Priority date: 2014-09-01
Filing date: 2014-09-01
Publication date: 2016-03-10
Anticipated expiration: 2017-03-01

Abstract

A method of adapting the number of sub-samples analysed by a spectrophotometer comprises acquiring spectral data from an initial plurality of sub-samples (4); processing the acquired spectral data to obtain a value for a sample parameter for each of the sub-samples of the initial plurality (6); determining a variation, particularly a standard deviation, in the obtained values for each of the initial plurality of sub-samples (8) as a Quality Index; and optionally obtaining spectral data from one or more new sub-samples depending on whether or not the Quality Index fulfils a predetermined condition.

Description

Method of Adaptive Sampling in a Spectrophotometer and a Spectrophotometer

Implementing the Method

[0001] The present invention relates to a method of adaptive sampling in a

spectrophotometer and to a spectrophotometer implementing the method. In particular the invention relates to the conditional acquisition of additional spectra in a spectrophotometer.

[0002] Spectrophotometers are widely used in the indirect analysis of one or

more sample properties, such as chemical or physical properties of the sample. The spectrophotometer generally comprises a spectral data acquisition module which is configured to illuminate sample material with a beam of electromagnetic radiation; to detect that beam after it has interacted with the sample material; and to record wavelength dependent variations in one or more properties of that interacted beam, and a data processor. The data processor may be located geographically remote from the spectral data acquisition module and interconnected by a

telecommunications link or it may be collocated with, often integrated in a same housing with, the spectral data acquisition module. The data processor employs a mathematical model to convert the recorded wavelength dependant variations in the properties of the interacted beam into a determination of a sample property, typically relating to the presence and/or amount of a constituent, even an unintentional or inappropriate constituent, of the sample.

[0003] It is well known that different constituents of a multi-constituent sample interact differently with optical radiation, particularly within the infra-red region of the electromagnetic spectrum, to produce more or less distinctive 'spectral fingerprints' in a wavelength dependent intensity variation spectrum. Stretching and bending vibrations of chemical bonds of sample constituents, for example, can provide characteristic absorption bands within the electromagnetic spectrum, particularly the near and mid infra red portions of the spectrum. Different particle size, which for ground cereal grains is related to hardness, may cause a sample to exhibit differential light scattering and/or absorption characteristics which again can be monitored through the interaction of optical radiation with the sample to provide information on a property of interest within the sample.

Chemometric analysis of optical spectra which derive from the detection of electromagnetic radiation in one or more wavelength regions from within ultra violet to infra red portions of the electromagnetic spectra (referred to herein as Optical radiation') by the spectrophotometer after it has interacted with a sample is now commonly employed as a means to derive quantitative or qualitative information about a property of the sample.

Chemometric analysis is a so-called 'indirect' technique, meaning that the constituent related information is not directly available from the recorded spectral data. Rather a mathematical model or 'calibration' must be established by linking spectral features of reference samples with information regarding a property of interest of those samples, which information is obtained for each reference sample using other, typically direct, analysis techniques. However, advantageously, chemometric analysis offers the ability to mathematically extract the relevant information about the property of interest of the sample through the development of a mathematical model that subsequently can be used for quantitative property prediction of new samples as well as for detection of deviating samples not taken into account in the calibration samples.

[0004] It is known from US 7698105 to provide such a spectrophotometer in

which an adaptive spectral data collection method is employed in order to optimise the number of spectra used in the determination of a property of the sample through chemometric analysis. Here a value of the property which is determined from spectra recorded from an initial set of spectra from a single sample region is used to decide whether or not one or more further spectra are to be acquired from that same sample region.

[0005] Often the amount of sample from which an individual spectrum is recorded is much smaller than the total amount of sample which is provided for analysis. This may be because the radiation employed to interact with the sample needs to be focussed into a relatively small spot size, which in turn illuminates a relatively small sub-sample, in order to have sufficient intensity for the interacted beam to generate a useful signal at the detector. Additionally or alternatively this may be because the sample material itself interacts strongly with the beam so that a sub-sample must be taken to restrict the amount of sample which is available to interact in order for the interacted beam to generate a useful signal at the detector.

[0006] As the sample becomes increasingly heterogeneous then the limitation on the amount of sample interacting with the beam becomes a greater issue. The property of the sample which is obtained from the spectrum of the sub-sample may not provide a proper representation of that property for the entire sample. This may be particularly problematic when the property to be determined is an analyte level since the distribution of the analyte itself may vary from sample to sample, even within the same material type.

[0007] To combat this problem it is known to make a determination of the sample property from the spectra collected from a plurality of sub-samples. Often the number of sub-samples to be collected by the spectrophotometer is fixed for a particular sample property. This number is typically selected to reflect an expected maximum heterogeneity within a sample material and so for a majority of samples of a particular material a smaller number of sub-samples would be optimal for generating an accurate measure of the sample property and for a few samples a larger number of sub-samples would be optimal. Thus the time taken to make a determination of a sample property is often longer than necessary.

[0008] It is an aim of the present invention to provide an adaptive sub-sampling scheme which is responsive to the degree of heterogeneity of the sample.

[0009] Accordingly, in a first aspect of the present invention there is provided a method of adaptive sampling in a spectrophotometer comprising the steps of: collecting spectral data from each of a plurality of sub-samples of a sample using the spectrophotometer; processing the collected spectral data in a data processor associated with the spectrophotometer to determine for each of the plurality of sub-samples a sample property;

comparing in the data processor the sample property for each of the plurality of sub-samples to obtain a Quality Index, said Quality Index being a measure of a variation in the determined sample property between the plurality of sub-samples; evaluating in the data processor whether the Quality Index fulfils a preset condition; and effecting a collection of spectral data from at least one additional sub-sample of the sample dependent on an outcome of the evaluation.

[0010] By effecting additional sub-sampling based on the variation in a measured sample property between sub-samples then the number of sub-samples required in order to obtain a value of the property which is acceptably representative of the sample may be optimised.

[001 1] In a second aspect of the present invention there is provided a

spectrophotometer comprising a spectral data acquisition module and a data processor, the data processor being configured to receive spectral data acquired by the spectral data acquisition module and adapted, for example through execution of program code implementing an algorithm, to cause the implement in the spectrophotometer the method according to the first aspect of the present invention.

[0012] As the spectrophotometer is adapted to implement the method according to the first aspect of the invention then the benefits and advantages associated with the method will be associated with the spectrophotometer of the second aspect.

[0013] The foregoing, as well as additional objects, features and advantages of the present invention, will be better understood through a consideration of the following illustrative and non-limiting detailed description of

embodiments of the present invention, made with reference to the drawings of the appended figures, of which:

Fig. 1 shows a flow chart illustrating an exemplary

embodiment of a method of adaptive sub-sampling

according o the present invention: and

Fig. 2 Shows an illustrative block diagram of a

spectrophotometer according to the present invention. [0014] Referring to Fig. 1 , here illustrated is an exemplary flow chart 2 illustrating an adaptive sub-sampling method according to the present invention for implementation in a spectrophotometer. A spectrophotometer is configured to collect spectral data from a sub-sample 4. The spectrometer, at step 4 is configured to collect spectral data from an initial plurality of sub- samples. The number of sub-samples making up this initial plurality typically depends upon the nature of the sample material and on the sample property to be determined from the collected spectra. The spectral data from each of the sub-samples of the initial plurality of sub-samples is processed in a data processor to determine a value of a sample property, at step 6, for each of these sub-samples. Typically, this processing comprises applying to the spectral data a chemometric mathematical model which links features in the spectral data to the sample property. A Quality Index is determined by the data processor, at step 8 from a comparison between the plurality of values of the sample property determined at step 6. The comparison is such that the Quality Index which is calculated at step 8 provides a measure of variation in the values of the sample property derived from the initial plurality of sub-samples. At least the Quality Index is compared in the data processor against one or more conditions at step 10. The comparison 10 determines if spectral data from one or more additional sub-sample(s) is to be collected and added to the spectral data already collected.

[0015] According to one embodiment of the method of the present invention

optionally, depending on the comparison 10 showing that the Quality Index represents an unacceptable variation in the values of the sample property determined at step 6, the spectrophotometer is instructed to collect additional spectral data from an additional sub-sample and the steps 4 to 10 are repeated with the spectral data thus far collected. This process may be repeated a number, n, times until other conditions at step 10 are fulfilled.

[0016] In addition to a comparison of the Quality Index at step 10 the data

processor may, for example, be configured to also perform a comparison of the number of sub-samples (or the number of additional sub-samples given that the number of the initial plurality of sub-samples is known) with a predetermined maximum and to prevent the collection of spectral data from further sub-samples if that maximum is exceeded.

[0017] In another embodiment, optionally and depending on the comparison of the Quality Index with preset conditions at step 10 showing that the Quality Index represents an unacceptable variation in the values of the sample property determined at step 6 the spectrophotometer is instructed to collect spectral data from a further plurality of sub-samples 4' and to compute, at step 12, a value of the sample property using the spectral data collected from the population of initial and further pluralities of sub- samples.

[0018] Additionally or alternatively a further condition validated at step 10 is

whether or not an outlier on a sub-sample exists An outlier is a sub-sample whose spectral data or value of the sample property falls outside the statistical variation in the population employed in the generation of the chemometric mathematical model employed in step 6 to determine the value of the sample property and may be, for example and as known in the art: a Global H outlier; a Leverage outlier; a Neighbourhood H outlier; a Range outlier; a Residual outlier; or a Standard Deviation outlier. If an outlier does exist then the data processor may be programmed to instruct the spectrophotometer to collect spectral data from the predetermined maximum number of sub-samples and use this data in the generation of the value of the sample property at step 12. In other embodiments additional conditions may be employed to trigger the collection of spectral data from the maximum number of sub-samples. Such conditions may for example include instrument temperature or environmental humidity level.

[0019] Ultimately the result of the comparison 10 is the generation and output 12 of a value of the sample property which is calculated in the data processor based on the chemometric mathematical model to some, preferably, all of the previously collected spectral data. In some applications spectral data from an outlier may be replaced with the spectral data which was acquired from an additional sub-sample following the comparison at step 10. [0020] The Quality Index determined at step 8 may be determined as any one or more of known statistical measures of variation within a data population, such as standard deviation, variance or average absolute deviation.

Preferably the Quality Index is calculated at step 8 as the standard deviation of the values of the determined sample property from step 6. Assuming that a plurality, p, of sub-samples have been analysed using the spectrophotometer then the standard deviation is calculated at step 8 as the square root of the average of the squared differences of the p values of the sample property from their average value.

[0021] Fig. 2 illustrates a block diagram of a spectrophotometer 14 adapted to operate according to the method represented in Fig. 1. The

spectrophotometer 14 comprises a plurality of functional units including a sampling unit 16; a spectrum generator 18 and a control unit 20 including a data processor 22. It will be appreciated that one or more of these functional units 16, 18, 20, 22 may comprise a plurality of sub-units configured to co-operate to provide the functionality of the appropriate functional unit as described below. As illustrated, this embodiment of a spectrophotometer 14 according to the present invention comprises a spectral data acquisition module which here consists of the sampling unit 16 and the spectrum generator 18. The sampling unit 16 and the spectrum generator 18 are interoperably configured to provide spectral data for processing by the data processor 22 which here is provided as a sub-unit of the control unit 20. In the present embodiment the spectral data acquisition module 16,18 and the data processor 22 are collocated in a single housing 24. In other embodiments the control unit 20, or at least the data processor 22, and the spectral data acquisition module 16,18 may be located in one or more separate housings and operably interconnected. Indeed one or more of these housings may be wirelessly interconnected with one or more of the other housings and perhaps held at a

geographically separate location. A sample container 26 for holding a sample material may be integral with the spectrophotometer or may, as illustrated in Fig. 1 , be provided separately. Sub-samples of the material within the sample container 26 are, in use, presented to the sampling unit 16 for analysis in the spectrophotometer 14. This presentation may be done manually or may be automated for example, a flow conduit may provide an interconnection between the sample holder 26 and the sampling unit 16 and a flow of sample within the conduit may be

controlled, such as via suitable valve means instructed from the control unit 20, so as to present a sub-sample to the sampling unit 16 for analysis. In other embodiments the sample holder 26 filled with sample may be manually presented to the sampling unit 16 and a relative movement between one or more elements of the sampling unit 16 and the sample holder 26 may be automatically effected under control of the control unit 20 so as to present different regions of the sample as sub-samples for analysis.

[0022] The sampling unit 16 of the present embodiment is configured to provide an optical interface with a sub-sample; to direct optical radiation towards the interfaced sub-sample for interaction therewith; and to detect the directed optical radiation after its interaction with the sub-sample.

Optionally the sampling unit 16 may include a source of optical radiation selected to emit optical energy from within the wavelength region in which the sample material is expected to have an interaction, for example one or more regions extending between and including the ultra-violet and the infra-red portions of the electromagnetic spectrum. The sampling unit 16 may also include a detector device for converting incident optical radiation into an intensity dependent electrical output which is made accessible to the data processor 22.

[0023] The spectrum generator 18 is configured to cooperate with the sampling unit 16 to generate spectral data upon instruction from the control unit 20, which spectral data here comprises values of intensity of interacted optical radiation indexed against a wavelength indicator. The spectrum generator 18 includes a wavelength selection device such as a spectrometer or interferometer for this purpose. In one embodiment the spectrum

generator 18 may include a source of optical radiation selected to emit energy across the a relatively wide band within the wavelength region in which the sample material is expected to have an interaction. The source of radiation is configured to emit its optical energy towards the

spectrometer which operates to provide a relatively narrower band output selected by wavelength. This output is provided to the sampling unit 16 for direction towards the interfaced sub-sample. In this case a signal representing the intensity of interacted, wavelength dispersed, radiation which is detected by the sampling unit 16 may be passed directly to the data processor 22 of the control unit 20 where it is stored indexed against the wavelength indicator. Alternatively, optical radiation after its interaction with the interfaced sub-sample is supplied from the sampling unit 16 as an input to the spectrum generator 18. The spectrum generator 18 is then configured to operate to output a signal representing the wavelength dependent intensity of the interacted optical radiation for access by, and preferably storage in, the data processor 22 as the spectral data.

[0024] The data processor 22 has associated with it a memory in which is stored a chemometric mathematical model by which spectral data is linked to a sample property. The data processor 22 is configured via program code to access this model, apply it to the received spectral data and to thereby determine a value of the sample property. In embodiments where the sample property relates to a constituent of the sample in the holder 26 the value is typically the concentration or the absolute amount of that constituent.

[0025] The spectrophotometer 14 thus far described constitutes a

spectrophotometer commonly known in the art. According to the present invention the data processor 22 is further configured via program code to cause the spectrophotometer 14 to operate according to an adaptive sampling protocol which implements the method described in relation to Fig. 1 in the spectrophotometer 14.

[0026] Thus the data processor 22 is caused to initiate the collection of spectral data 4 from each of an initial plurality of sub-samples of the sample. A minimum number for the plurality of sub-samples to be analysed is made accessible to the data processor 22. This number may be entered into an associated memory directly, such as via a human machine interface like a keyboard or touch screen), or it may be derived in the data processor 22 from, for example, a knowledge of material to be analysed and the sample property to be determined (which items of information may be entered into an associated memory by a user). For this purpose a lookup table may be provided for access by the data processor 22 which contains the minimum number of sub-samples to be obtained for different material/property combinations. After this collection of spectral data from each of the initial plurality of different sub-samples the data processor 22 is caused determine a sample property 6 by applying the chemometric model to the spectral data collected from each of the plurality of sub-samples in turn and thereby determine a same plurality of values of the sample property which is associated with the chemometric model 6. The data processor 22 then determines a Quality Index 8, being a measure of a variation, in one embodiment a standard deviation, in the determined sample property between the plurality of sub-samples. The Quality Index is then compared in the data processor against one or more conditions 10 relating to an acceptable value of this Index and the data processor 22 directs the optional operation of the spectrophotometer 14 depending on the outcome of this comparison 10. Should the outcome indicate that the Quality Index indicates an unacceptable variation in the values of the sample property as determined for each sub-sample from which spectral data was collected then the spectrophotometer 14 is directed to collect additional spectral data from one or more additional sub-samples 4, 4'. The process 4, 6,8,10 up to and including the conditional comparisonI O may, in one embodiment be repeated a plurality, n, of times, preferably limited by a comparison 10 of the number of sub-samples from which spectral data has been collected with a predetermined maximum number. In another embodiment the conditional comparison(s) 10 is performed only once and the data processor 22 is configured to optionally direct a collection of spectral data from a further plurality of sub-samples 4' dependent on the outcome of the comparison(s). On completion of the collection of spectral data from a sufficient plurality of sub-samples (as determined from the comparison(s) 10) the data processor 22 is caused to calculate the spectral data and the chemometric mathematical model a value of the sample property and to report that value 12.

[0027] In a particular and non-limiting embodiment of the present invention the spectrophotometer 14 is adapted to analyse cereal grains (processed or unprocessed) to determine constituent levels in the cereal such as moisture, ash, protein content from the application of a chemometric mathematical model to wavelength dependent intensity variations (spectral data) within the infrared portion of the electromagnetic spectrum. The data processor 22 is configured to compare a standard deviation of values of each sample property derived from the spectral data of each of an initial plurality of sub-samples and to direct the collection of additional spectral data from each of an additional plurality sub-samples if the standard deviation in the values of any of the sample properties exceeds a maximum for that property. A value of each of the sample properties is then derived in the data processor 22 using the spectral data collected from all of the sub-samples and the appropriate chemometric

mathematical model (s).

[0028] The spectrophotometer 14 may be adapted to analyse liquids such as

wine, milk, juices and intermediated products in a production process involving these liquids to determine constituent levels as the sample property.

[0029] Suitable program code may be implemented in software or firmware and generated using one of a variety of know programming techniques, which program code when run by the data processor 22 causes the data processor 22 to operate as described above.

Claims

1. A method of adaptive sampling in a spectrophotometer comprising the steps of: collecting spectral data from each of a plurality of sub-samples of a sample using the spectrophotometer (4; 4');

processing the collected spectral data in a data processor associated with the spectrophotometer to determine for each of the plurality of sub-samples a sample property (6);

comparing in the data processor the sample property for each of the plurality of sub-samples to obtain a Quality Index, said Quality Index being a measure of a variation in the determined sample property between the plurality of sub-samples (8);

evaluating in the data processor whether the Quality Index fulfils a preset condition( 10); and

effecting a collection of spectral data (4; 4') from at least one additional sub- sample of the sample dependent on an outcome of the evaluation.

2. A method as claimed in Claim 1 wherein the Quality Index is a standard

deviation of the determined sample property of each of the plurality of sub- samples.

3. A method as claimed in claim 2 wherein the preset condition is whether the standard deviation exceeds a predetermined value.

4. A method as claimed in claim 3 wherein the preset condition evaluated in the data processor (10) further includes whether a maximum number of sub- samples is reached.

5. A spectrophotometer (14) comprising a spectral data acquisition module

(16,18); and a data processor (22) configured to receive spectral data acquired by the spectral data acquisition module (16,18); wherein the data processor (22) is adapted to cause implementation in the spectrophotometer (14) of the method according to any one of the preceding claims.