Technical Field

The present invention relates to a method of analysing a sample using spectroscopy such as vibrational spectroscopy, for example, FTIR spectroscopy. In particular, the method may be used to calibrate a spectrometer or may be used to inform a user about a composition or characteristics of a process and provide information for controlling the process in real-time.

Background

Industrial processes may be monitored by periodically taking samples and analysing them. It is common to collect a sample from the process using, for example, a dip probe, which is then sent to a laboratory for analysis. Various methods are used to analyse such samples including spectroscopy and chromatography.

In some industrial processes with starting materials coming directly from natural resources the amount of product and by-products can be hugely variable. Moreover, byproducts themselves may be useful for other purposes and have significant value and it may be desirable that such by-products are not simply lost or become waste.

One such example process is the kraft process of making paper pulp from wood pulp. The process removes lignin, hemicelluloses and other extractives. A major by-product of the process is black liquor which comprises the lignin residues, hemicellulose and inorganic chemicals used in the process. A proportion of the black liquor is made of solids of which part are organic chemicals and the remainder are inorganic chemicals. The organic matter is made up of components from the degradation of the wood which are now water/alkali soluble. The black liquor contains more than half the energy content of the equivalent starting wood and may be burned to produce energy to recover some of the chemicals used in the processing. An important by-product recovered from the black liquor is tall oil which may be used for making vegan soap, and unlike many oils used for vegan soap it is not derived from a food source. Other important by-products recovered from the black liquor include roisin acids, which are difficult to fabricate artificially and are used for cosmetics, adipic acid and pine oil.

Tall oil is particularly valuable and it is desirable that it is removed from the black liquor to be used for other purposes and is not burned to produce energy. To separate the tall oil a settling tank is used to settle off the tall oil but allow other by-products in the black liquor to pass through and be burned at a boiler. To test for tall oil and avoid the loss of a valuable resource, samples of extractives are taken. Conventionally, the extractives are analysed using the following process:

1) Take a sample from the black liquor and evaporate the sample to dryness;

2) Weigh the solids left over in the dried sample;

3) Wash the solids with acetone;

4) Weigh the washed solids; and

5) Determine the weight lost (in g/kg relative to the starting sample) and that therefore has been dissolved by the acetone wash.

The amount dissolved by the acetone wash may be used as an indicator of the level of tall oil in the sample. The process may be sampled periodically such as every six hours and the relative level or timing at which the tall oil is removed from the settling tank adjusted.

A problem with the above-mentioned sampling and analysis is that it is performed off-line from the process such that by the time results have been obtained the process may have changed. The analysis also provides only minimal information on the status of the tall oil separation. For example, it is not possible to determine if the species removed by the acetone wash are very soluble or only partially soluble in the acetone.

There are many other industrial processes where it is desirable to obtain more information in real-time to allow better control or optimisation. The paper pulp and black liquor process described above is just one example.

Another example process which is also related to paper processing is that of processing recycled paper. Stickies are tacky substances contained in the paper pulp and process water systems during the recycling process and have the potential to contaminate the process and equipment. It is believed the stickies are formed of glues, dyes and other residues on the paper when it arrives at the recycling facility. In principle, the presence of dyes could be assessed using liquid chromatography-mass spectrometry (LCMS). However, the resulting chromatogram is often too complex to provide useful results. Scavenger chemicals can be used to remove stickies. It is desirable to find analysis tools that can provide real-time data on the amount of various chemicals, such as the scavenger chemicals, at various points in the process.

ASAs (Alkenyl succinic anhydrides) are used in paper sizing to make the paper surface hydrophobic and prevent the paper from being susceptible to absorbing water. However, ASA can also form stickies. One way to limit the formation of stickies at the paper sizing process is to slow down the rate at which the ASA is input to the process. To allow the input rate of ASA to be optimally set it would be desirable to be able to measure in realtime the ASA levels through the production process. Conventional methods for calibrating a spectrometer use standard chemicals with known spectroscopic responses. While this technique works well for spectrometers used in a laboratory, the process of calibration for spectrometers installed in a process is more difficult, especially as local temperature and pressure changes in process installations can change the spectral response. Moreover, for spectrometers that are retro-fitted to process installations it is not always possible to use standard chemicals for calibration in a live running process as this would require halting of the process, clearing down the process, flowing through the process the standard chemicals to set the calibration and then returning the process to normal running. As well as the processing time lost and cost, some processes can take a lengthy amount of time to get back up and running. Hence, other calibration methods that are not as disruptive to the process are desired.

Summary of the Invention

The present invention provides methods of analysing a sample and method of calibrating a spectrometer. For example, the present invention provides a method of analysis which uses a treatment step, in combination with models, to extract more information on process samples than can be obtained from usual spectroscopic methods. The present invention further provides a method of calibrating a spectrometer, such as a spectrometer that is built in to a process, that does not require the process to be taken offline. Furthermore, this calibration is based on, and valid for, the actual conditions of the process, namely the process temperature and pressure.

The methods of analysis include “dry” methods that receive data collected from molecular vibrational spectroscopy, such as infrared absorption spectroscopy or Raman spectroscopy, and process the data to determine a characteristic or property of a process material, or provide calibration data which may be uploaded to a process spectrometer coupled to a process. The “dry” methods may be performed by a computer or processing means. The present invention also provides “mixed wet and dry” methods which include the dry methods along with step such as collecting samples from the process.

The present invention provides a method of analysing a sample obtained from a process. The process may be an industrial process such as an industrial chemical process or an industrial biotechnological process. The method comprises: receiving or generating data corresponding to vibrational spectra, such as infrared absorption spectra or Raman spectra, of one or more first samples obtained from the process; performing an analysis, such as multivariate decomposition, MD, analysis on the data corresponding to the vibrational spectra of the one or more first samples to generate a model such as an MD model; receiving or generating data corresponding to a vibrational spectrum of a treated second sample, wherein the treatment comprises a first treatment to change the composition of the second sample; evaluating, fitting or comparing the vibrational spectrum of the treated second sample against the model; and determining a characteristic or property of the treated second sample based on the evaluation or comparison of the vibrational spectrum of the treated second sample against the model. The second sample may be collected at substantially the same time as, or collected as part of, the first sample. The characteristic or property may be representative of chemical species present in the process. By the term vibrational spectra we mean spectra generated by molecular vibrational spectroscopy.

The step of evaluating the vibrational spectrum of the treated second sample against the MD model may comprise comparing the vibrational spectrum of the treated second sample, or data based thereon, with the MD model, or projecting the vibrational spectrum of the treated second sample, or data based thereon, onto the MD model, or applying the MD model to the vibrational spectrum of the treated second sample, or data based thereon. The method may further comprise calculating a residual or score between the vibrational spectrum of the treated second sample and the MD model to determine the characteristic of the treated second sample.

The vibrational spectra of the one or more first samples and the vibrational spectrum of the treated second sample may be generated by infrared spectroscopy, near infrared spectroscopy, mid infrared spectroscopy, UV-visible spectroscopy or Raman spectroscopy, and may include FTIR spectroscopy. The MD model may be considered to be a representation of spectra of the materials of the one or more first samples.

The above-described embodiment is directed to generating the MD model based on the spectra of the one or more first samples and evaluating the spectrum of the treated second sample against the model. The spectra of the one or more first samples may comprise multiple spectra of a single first sample, single spectra of each multiple first samples, or multiple spectra of each of multiple samples. The spectra of one or more first samples may be considered together as a set of first spectra. In other embodiments, the MD model may be generated based on spectra of the treated second sample and the spectra of the one or more first samples may be evaluated against the MD model. In such a case the spectra may comprise multiple spectra of a single treated second sample, single spectra of each multiple treated second samples, or multiple spectra of each of multiple treated second samples. The spectra of one or more first samples may be considered together as a set of first spectra, and the spectra of one or more treated second samples may be considered together as a set of second spectra. The set of spectra used to generate the MD model preferably includes at least three spectra.

The method may comprise treating the second sample obtained or collected from the process, wherein the treatment comprises the first treatment to change the composition of the second sample. In this regard the present invention provides a method of analysing a sample obtained from a process such as an industrial process or an industrial biotechnological process. The sample may be considered to be an analyte. The method may comprise the steps of the dry or computing steps of the preceding paragraphs as well as steps of collecting samples. The method comprises: generating or obtaining vibrational spectra, such as Fourier Transform Infra-red, FTIR, spectra of one or more first samples obtained from the process. The first sample may be considered to be a training sample because it is used for generating a model. Preferably, at least three vibrational spectra are generated for the first sample or samples. This may be three spectra from one sample if only one sample is obtained, use single spectra from multiple samples, or multiple spectra each from multiple samples. The method further comprises performing multivariate decomposition, MD, analysis on the spectra of the one or more first samples to generate an MD model; and treating a second sample obtained from the process, wherein the treatment comprises a first treatment to change the composition of the second sample or to remove species or types of chemical species from the second sample. The method further comprises: generating a vibrational spectrum of the treated second sample; evaluating, fitting or comparing the vibrational spectrum of the treated second sample against the MD model; and determining a characteristic of the treated second sample based on the evaluation of the vibrational spectrum of the treated second sample against the MD model. The step of evaluating the vibrational spectrum of the treated second sample against the MD model may comprise comparing the vibrational spectrum of the treated second sample, or data based thereon, with the MD model, or projecting the vibration spectrum of the treated second sample, or data based thereon, onto the MD model, or applying the MD model to the vibration spectrum of the treated second sample, or data based thereon. The method may further comprise calculating a residual or score between the vibration spectrum of the treated second sample and the MD model to determine the characteristic or property of the treated second sample. As set out above in relation to the dry or processing steps, the MD model may alternatively be generated based on spectra of the treated second sample and the spectra of the one or more first samples may be evaluated against the MD model. In such a case the spectra may comprise multiple spectra of a single treated second sample, single spectra of each multiple treated second samples, or multiple spectra of each of multiple treated second samples.

The vibrational spectra of the one or more first samples and the vibrational spectrum of the treated second sample may be generated by infrared spectroscopy, near infrared spectroscopy, mid infrared spectroscopy, UV-visible spectroscopy or Raman spectroscopy and may include FTIR spectroscopy. The step of comparing or evaluating may quantify the characteristic of the treated second sample based on the treatment performed.

Alternatively, other types of molecular vibrational spectroscopy may be used. A model may be a statistical model that is used to find a fit to the data using reduced parameters compared to the starting data. For example, the model may resolve or decompose the data to this more limited number of parameters. New data can then be readily compared to existing data using the model. In the method the step of projecting the vibrational spectra or FTIR spectra may alternatively be considered to be applying the FTIR spectra to the model. The step of projecting may use data from the vibrational spectrum or FTIR spectrum or data that has already gone through some processing such mean centring.

The first sample or samples and the second sample may be collected from the process at substantially the same time. That is, they may be collected together such as a single sample that is subsequently split up or they may be collected separately but over timescales for which the process is substantially invariant. This may depend on the process and may, for example be a small number of minutes such as 2, 5 or 10 minutes. Multiple second samples may be collected and used in the method. The multiple second samples may be collected at the same respective times as first samples.

The step of generating vibrational spectra of the first sample or samples may comprise generating at least three vibrational or FTIR spectra of the first sample. The MD analysis may be principal component analysis or multivariate curve resolution and the MD model may be a PCA or MCR model.

The vibrational spectra may cover a range, or be limited over a range, in the mid infra-red such as 600-4000cm ^-1 or 800-1800cm ^-1 .

The method may further comprise thermostatically controlling the temperature of the first sample such that the vibrational spectra of the first sample(s) are generated at substantially the same temperature, for example room temperature or at around 20°C, such as between 18 and 22°C.

The method may further comprise: treating one or more further second samples obtained from the process, wherein the treatment is the first treatment and the one or more further second samples have been obtained from the process at one or more times after the second sample; generating a vibrational spectrum of each of the one or more treated further second samples and adding the vibrational spectra to a set of second spectra; evaluating or comparing the vibrational spectrum of the one or more treated further second samples, such as those added to the set of second spectra, against the MD model or revising the MD model based on the updated set of second spectra, depending on whether the set of first spectra or the set of second spectra have been used to generate the MD model, and re-evaluating the set of first spectra against the MD model; and determining or quantifying the characteristic or property of the one or more treated further second samples based on the evaluation. The further second sample is a sample collected in the same manner as the first sample but collected at a later time. The step of evaluating may comprise calculating one or more residuals or scores between the vibrational spectrum of the treated further second samples and the MD model if the model is based on the one or more first samples, or calculating one or more residuals or scores between the vibrational spectra of the one more first samples and the MD model if the MD model is based on the set of second spectra to determine the characteristic or property of the one or more treated further second samples.

The method may further comprise comparing the residuals or scores to determine a change in the characteristic over time. It is preferable that the vibrational spectra of each of one or more treated further second samples are evaluated against the MD model which has been generated using one or more first samples. This is because if the model is instead built on spectra from treated second samples and revised using spectra from treated further second samples, it is more difficult to detect changes in the residuals or scores.

If the multivariate decomposition, MD, analysis is performed on the set of first spectra to generate the MD model, and the evaluation, against the MD model, is an evaluation of the set of second spectra against the MD model, then the method may further comprise: treating a third sample obtained from the process, wherein the treatment comprises a second treatment different from the first treatment; generating a vibrational spectrum of the treated third sample; evaluating the vibrational spectrum of the treated third sample against the MD model and determining or quantifying a second characteristic of the treated third sample based on the second treatment performed, wherein the characteristic of the treated second sample is a first characteristic and is different to the second characteristic of the treated third sample. The step of evaluating may comprise comparing the vibrational spectrum of treated third sample with the MD model or projecting the vibrational spectrum of treated third sample onto the MD model, and calculating a residual or score between the vibrational spectrum of the treated third sample and the MD model to determine the second characteristic.

The method may further comprise: treating one or more further third samples obtained from the process, wherein the treatment comprises the second treatment; generating a vibrational spectrum of each of the one or more treated further third samples; evaluating the vibrational spectrum of the one or more treated further third samples against the MD model; and determining or quantify the second characteristic of the treated one or more further third samples based on the evaluation. The step of evaluating may comprise comparing the vibrational spectra of the one or more further treated third samples with the MD model or projecting the vibrational spectra of the one or more further treated third samples onto the MD model, and calculating one or more residuals or scores between the vibrational spectrum of the one or more further treated third samples and the MD model to determine the second characteristic of the one or more further treated third samples

The method may further comprise: comparing the residuals or scores obtained from the treated second sample and/or one or more treated further second samples against the treated third sample and/or one or more treated further third samples to determine a combined difference in the characteristics. The method may further comprise grouping the samples according to the determined characteristics. These methods may be used to determine whether one or two characteristics of a process material are changing with time. By taking further samples and performing different treatments changes to other characteristics may be monitored.

The present invention provides a method of determining the relative amounts of polar and non-polar organic molecules in a series of samples, the samples collected and analysed periodically, the method comprising steps out above in relation to second and third samples which respectively undergo first and second treatments, wherein the second samples are treated by a first treatment comprising washing with dichloromethane and the third samples are treated by a second treatment comprising washing with hexane. The method may further comprise at least one of: providing an indication whether the amount of polar organic molecules in the process at the time of sampling is in excess of, or below, a relative threshold, wherein the amount of polar organic molecules is based on an evaluation or comparison, such as calculating the residual or score, between the vibrational spectrum of the treated second sample and the MD model; and providing an indication whether the amount of non-polar organic molecules in the process at the time of sampling is in excess of, or below, a relative threshold, wherein the amount of non-polar organic molecules is based on an evaluation or comparison, such as calculating the residual or score, between the vibrational spectrum of the treated third sample and the MD model.

The method may further comprise: grouping the samples in to samples comprising: high levels of polar organic molecules and low levels of non-polar organic molecules; high levels of polar organic molecules and high levels of non-polar organic molecules; low levels of polar organic molecules and high levels of non-polar organic molecules; and low levels of polar organic molecules and low levels of non-polar organic molecules, wherein a determination of high or low status may be made in comparison to a measurement threshold.

The present invention provides a method of analysing black liquor or other process material for polar and non-polar organic molecules, the method comprising: generating vibrational spectra, such as FTIR spectra, of one or more first samples obtained from the process; performing multivariate decomposition, MD, analysis such as principal component analysis, PCA, on the spectra of one or more first samples to generate an MD model, such as a PCA model; treating a second sample obtained from the process, wherein the treatment comprises a first treatment of washing the second sample with dichloromethane to remove polar organic molecules; generating a vibrational spectrum, such as an FTIR spectrum, of the treated second sample; treating a third sample obtained from the process, wherein the treatment comprises a second treatment of washing the third sample with hexane to remove non-polar organic molecules; generating a vibrational spectrum, such as an FTIR spectrum, of the treated third sample; evaluating the vibrational spectra of the treated second sample and the treated third sample against the MD model or PCA model; and determining a relative level of polar organic molecules removed from the second sample based on the evaluation and determining a relative level of non-polar organic molecules removed from the third sample based on the evaluation. The evaluation step may comprises calculating a residual or score between the vibrational spectrum, such as the FTIR spectrum, of the second sample and the PCA model; and calculating a residual or score between the vibrational spectrum, such as the FTIR spectra, of the third sample and the PCA model.

The first, second and third samples maybe collected from the process at substantially the same time, or collected together from the process. The MD analysis may be PCA analysis or multivariate curve resolution and the MD model may be a PCA model or an MCR model.

The present invention provides a method of determining the relative amounts of polar and non-polar organic molecules in a series of samples of black liquor resulting from digesting wood pulp into paper pulp, the method comprising steps set out above.

The methods set out herein may be applied to many other processes and the first and/or second treatments may comprises mechanical, chemical or physical means.

The present invention provides a method of determining characteristics of a process stream, the method comprising: collecting a sample from a process stream; and performing the method steps set out above.

The present invention provides a method of maintaining a characteristic of a process, the method comprising: performing the methods set out above; and if the characteristic is outside a target range, then adjusting a process control parameter to bring the characteristic back into the target range. For example, controlling a process parameter may comprise: opening or closing a valve; adding or, increasing or decreasing, the amount of a chemical added to the process. In general this aspect may be considered to be controlling a PID to bring the process back to a normal or optimised state.

The present invention provides a method of calibrating a process spectrometer coupled to a process, such as a spectrometer mounted in-situ at a process with a probe extending into or against the process material. The method comprises: generating an vibrational spectrum of a process material within the process using the process spectrometer; performing a method of analysing a sample obtained from a process as set out in any of the preceding paragraphs, wherein the sample may be one or more first samples obtained or collected from the process and the evaluation step provides one or more parameters describing a characteristic or property of the one or more first samples; generating a second model linking the vibrational spectrum of the process material generated using the process spectrometer and the one or more parameters, wherein the second model represents a calibration for estimating a value of the characteristic or property of the process material based on a vibrational spectrum generated by the process spectrometer. The calibration or model may then be uploaded to the process spectrometer. Accordingly, the present invention provides a method comprising: acquiring or generating a vibrational spectrum of a process material/stream within the process using the process spectrometer; collecting one or more first samples of process material from the process or process stream; generating vibrational spectra of the one or more first samples collected from the process or process stream; performing multivariate decomposition, MD, analysis on the spectra of the one or more first samples to obtain an MD model; treating a second sample collected from the process, wherein the treatment comprises a first treatment to change the composition of the second sample, and the second sample is collected at substantially the same time as, or collected as part of, the first sample; generating a vibrational spectrum of the treated second sample; projecting the vibrational spectrum of the treated second sample on to the MD model; comparing the vibrational spectrum of the treated second sample against the MD model to determine one or more parameters based on the comparison; generating a second model, such as using regression analysis, linking the vibrational spectrum of the process material generated using the process spectrometer and the one or more parameters, wherein the second model represents a calibration for estimating a value of the characteristic of the process material based on a vibrational spectrum generated by the process spectrometer; and uploading the model to the process spectrometer. Some of the steps included here are substantially similar to the method of analysing a sample set out in the preceding paragraphs. Accordingly, optional aspects applicable to that method are equally applicable here. The step of comparing may comprise calculating a residual or score between the vibrational spectrum of the treated second sample and the MD model. The vibrational spectra may be generated by infrared spectroscopy, near infrared spectroscopy, mid infrared spectroscopy, UV-visible spectroscopy or Raman spectroscopy and may include FTIR spectroscopy. The above-described embodiment is directed to generating the MD model based on the spectra of the one or more first samples and evaluating the spectrum of the treated second sample against the model. The spectra of the one or more first samples may comprise multiple spectra of a single first sample, single spectra of each multiple first samples, or multiple spectra of each of multiple samples. The spectra of one or more first samples may be considered together as a set of first spectra. In other embodiments, the MD model may be generated based on spectra of the treated second sample and the spectra of the one or more first samples may be evaluated against the MD model. In such a case the spectra may comprise multiple spectra of a single treated second sample, single spectra of each multiple treated second samples, or multiple spectra of each of multiple treated second samples. The spectra of one or more first samples may be considered together as a set of first spectra, and the spectra of one or more treated second samples may be considered together as a set of second spectra.

The step of generating vibrational spectra of the one or more first samples may comprise generating at least three vibrational spectra of at least one of the one or more first samples. The MD analysis may be principal component analysis, PCA, or multivariate curve resolution, MCR, analysis and the MD model may be a PCA or MCR model.

The vibrational spectra may cover a range, or be limited to a range, in the mid infrared such as 600-4000cnr ¹ or 800-1800cm ^-1 .

The method may further comprise the steps of: acquiring or generating a further vibrational spectrum of the process material/stream within the process using the process spectrometer; collecting one or more further first samples of process material from the process or process stream; generating vibrational spectra of each of the further first samples collected from the process or process stream; performing multivariate decomposition, MD, analysis on the spectra of the further first samples and first sample to obtain an MD model; treating one or more further second samples collected from the process, wherein the treatment comprises the first treatment to change the composition of the second sample, and the one or more further second samples are collected at substantially the same time as, or collected as part of, the further first sample; generating a vibrational spectrum of each of the treated one or more further second samples; comparing the vibrational spectra of the one or more treated further second samples against the MD model to determine one or more parameters based on the comparison; and revising the second model linking the vibrational spectrums generated using the process spectrometer and the parameters. The step of comparing may comprise calculating a further residual or score between the vibrational spectra of the treated further second samples and the MD model.

The method may further comprise repeating the preceding steps with additional further first samples and additional further second samples until the first and/or second models meet a closeness or confidence of fit requirement to the collected spectra.

The present invention further provides for use of the calibration set out above to convert raw data collected by an in-situ process spectrometer to a generate an estimate of a characteristic of the process stream in real-time.

The present invention provides computer readable media, such as non-transitory computer readable media, having stored thereon instructions, such as computer program code, that when executed cause a computer, analyser, spectrometer or processor to perform computational steps set out above and/or below. The present invention further provides corresponding computer program code or signals.

Accordingly, the present invention provides a computer program for analysing data relating to a sample obtained from a process, the computer program comprising computer program code for performing the steps of: receiving or generating data corresponding to vibrational spectra of one or more first samples received or obtained from the process; performing multivariate decomposition, MD, analysis on the data corresponding to the vibrational spectra of the one or more first samples to generate an MD model; receiving or generating data corresponding to a vibrational spectrum of a treated second sample, wherein the second sample has been collected at substantially the same time as, or collected as part of, the first sample, and the treatment comprises a first treatment to change the composition of the second sample; evaluating the vibrational spectrum of the treated second sample against the MD model; and determining or quantifying a characteristic of the treated second sample based on the evaluation performed. The step of evaluating may comprise comparing the vibrational spectrum of the treated second sample with the MD model and calculating a residual or score between the vibrational spectrum of the treated second sample and the MD model to determine a characteristic of the treated second sample based on the first treatment performed.

The MD analysis may be principal component analysis, PCA, or multivariate curve resolution analysis, MCR, and the MD model may be a PCA model or an MCR model.

The present invention further provides a computer program for calibrating a process spectrometer such as an in-situ process spectrometer, the computer program comprising computer program code for performing the steps of: receiving or generating data corresponding to vibrational spectra of one or more first samples obtained from the process or process stream; performing multivariate decomposition, MD, analysis on the spectra of the one or more first samples to obtain an MD model; receiving or generating data corresponding to a vibrational spectrum of a treated second sample, wherein the second sample has been collected at substantially the same time as, or collected as part of, the first sample, and the treatment comprises a first treatment to change the composition of the second sample; comparing the vibrational spectrum of the treated second sample against the MD model to determine one or parameters based on the comparison; generating a second model, such as using regression analysis, linking the data corresponding to the vibrational spectrum generated using the process spectrometer and the one or more parameters, wherein the second model represents a calibration for estimating a value of the characteristic of the process material or process stream based on a vibrational spectrum generated by the process spectrometer; and uploading the model to the process spectrometer. The step of comparing may comprise calculating a residual or score between the vibrational spectrum of the treated second sample spectrum and the MD model.

The MD analysis may be principal component analysis, PCA, and the MD model may be a PCA model, or the MD analysis may be multivariate curve resolution, MCR, analysis and the MD model may be an MCR model.

The present invention comprises apparatus comprising: a vibrational spectrometer, such as an FTIR spectrometer; and a computer comprising a processor and memory, wherein the apparatus is configured to store and process any of the computer programs or methods set out above. The computer may be part of the spectrometer or a separate device. The computer may further comprise an input for receiving data describing the vibrational spectra described herein and may comprise an output for outputting a calculation result such as a characteristic or property of the sample.

The above-described embodiments of computer programs are directed to generating the MD model based on the spectra of the one or more first samples and evaluating the spectrum of the treated second sample against the model. The spectra of the one or more first samples may comprise multiple spectra of a single first sample, single spectra of each multiple first samples, or multiple spectra of each of multiple samples. The spectra of one or more first samples may be considered together as a set of first spectra. In other embodiments, the MD model may be generated based on spectra of the treated second sample and the spectra of the one or more first samples may be evaluated against the MD model. In such a case the spectra may comprise multiple spectra of a single treated second sample, single spectra of each multiple treated second samples, or multiple spectra of each of multiple treated second samples. The spectra of one or more first samples may be considered together as a set of first spectra, and the spectra of one or more treated second samples may be considered together as a set of second spectra.

In the methods, apparatus, computer programs, code and computer readable media, the infrared absorption spectra may be any suitable mid-infrared absorption spectra, but is preferably FTIR spectra.

Brief Description of the Drawings

Embodiments of the present invention, along with aspects of the prior art, will now be described with reference to the accompanying drawings, of which: figure 1 is a schematic diagram of a spectrometer coupled to a process; figure 2 is a block diagram of a method of obtaining information on a process based on spectroscopic measurements and treatment of samples; figure 3a is a schematic diagram showing FTIR absorption spectra of a number of samples; figure 3b is 3D plot showing a set of data that has been applied to PCA space; figure 4 is a block diagram of a method of obtaining information on a process based on spectroscopic measurements and treatment of samples using two different treatments; figure 5 is a 1-D graph showing residuals for samples processed by washing with DCM and hexane; figure 6 is a 2-D graph showing residuals for samples processed by washing with DCM and hexane; figures 7a and 7b respectively show a dendogram and table relating to clustering of samples, whereas figure 7c is a copy of figure 6 with an example of clustering illustrated; figure 8 is a block diagram showing method steps for calibrating a process spectrometer; figure 9 comprises three plots that illustrate the projection of the data from the DCM treated samples and the hexane treated samples to a PCA model generated from the raw samples; figure 10 shows four plots relating to the fit of data to the hexane treated samples; figure 11 is a schematic graph showing how an identified characteristic, such as the amount of polar organic molecules, can be used to monitor and control the process in realtime; and figure 12 shows a plot providing a comparison between the extractives method and separate washing of samples using DCM and hexane.

Detailed Description

Figure 1 shows an example spectrometer 20 coupled to a process 10 for measuring a vibrational spectrum of the process. This may for example be an infrared absorption spectrum, such as an FTIR spectrum, or a Raman spectrum. The spectrometer may be coupled to any part of a process in which there is product, batch, other material of the process or by-product of the process. The part of the process may include a pipe or tank. Spectrometer 20 is coupled to the process by a probe 26. Light or electromagnetic radiation, which may be infra-red, is emitted from a source which may be part of interferometer 22. The infra-red radiation may preferably be mid infra-red such as in the range 600-4000cm ^-1 or 800-1800cm ^-1 .The probe 26 directs the light to the process and couples it back to the interferometer for measurement. The end of the probe may comprise a window that allows the light to interact with the process material, such as to allow the process material to absorb the light at particular wavelengths. The interferometer 22 generates a pattern, such as an interference pattern on a detector. The signal from the detector representing the interference pattern is sent to electronics 24 for processing. Display 30 may provide an output spectrum, such as an FTIR spectrum, showing characteristics of the process. Electronics may comprises a memory and processor, such as forming a computing device, that processes that data. Alternatively, 30 may be a computing device including a display. In other embodiments the spectrometer and display may be incorporated together as one. Aspects of the present invention relate to the processing and calibration of the spectrometer 20.

Preferably, the spectrometer comprises an interferometer such as the one described in WO 2011/086357 A1 or US 9046412 B, which are hereby incorporated by reference. Alternatively, the spectrometer may be different but preferably is a form of molecular spectroscopy such as vibrational spectroscopy.

In alternative embodiments the spectrometer may be located away from the process to perform off-line analysis. In both cases the spectrometer may be an FTIR spectrometer such as the IRmadillo™ spectrometer by the applicant or any other suitable process spectrometer.

Various aspects of the analysis may be performed by a computer which forms part of the spectrometer or may be coupled to the spectrometer. For example, the computer may receive the spectra such as the FTIR spectra and perform the analyses described below. Accordingly, computer programs may be provided for performing the analyses.

Figure 2 is a block diagram illustrating a method of obtaining information on a process based on spectroscopic measurements and treatment of samples.

The method of figure 2 commences by taking a sample at step 100 such as from a process. This may include dipping a probe into a process to collect a sample from the process. Next at step 110 the sample is input into a spectrometer, such as an FTIR spectrometer, to generate a vibrational spectrum. This may be an infrared absorption spectrum, such as an FTIR spectrum. The spectrum may be collected over a wavelength range such as 600-4000 cm ^-1 or 800-1800 cm' ¹ . The step of generating a spectrum is repeated because preferably at least three spectra are used to build a principal component analysis model. More preferably, around ten or more spectra are collected. It is also desirable to build up the model using multiple samples. The samples may be collected over an extended period of time. However, in such a case it should be borne in mind that the process and components in the process may vary over a time period.

Once sufficient data has been collected the data may be used to build a Multivariate Decomposition (MD) model such as a Principal Component Analysis (PCA) model. Other types of model may be used instead such any that result in decomposition of the data into components linked to the identification of components or characteristics of the sample. For example, models may be generated using multivariate curve resolution, MCR. PCA performs an unsupervised search for eigenvector solutions, which in PCA analysis are known as Principal Components. MCR is supervised and as a result the number of eigenvectors may be set by the user. In the present methods, the number of eigenvectors may be set as eight or more. We now describe the process of generating a model using the preferred example of Principal Component Analysis and FTIR spectra.

In spectroscopy large amounts of data can be generated when many samples are measured such as in continuous or real-time monitoring in a production environment. For example, an FTIR spectrum absorption or transmittance spectrum can be collected at regular intervals such as once per minute. Over a period of hours or days the data that is built up can become large. It is often desirable to analyse the data from different samples to determine changes in the chemical composition. This can be difficult when the sample being analysed is a complex mixture of different chemicals.

Principal Component Analysis (PCA) is a statistical method of analysing large amounts of data. In the case of spectroscopy, an FTIR spectrum of a sample may comprise data points over a frequency range from, for example, 1800cm ^-1 to 800 cm ^-1 . The data points may be every 1cm ^-1 or more or less. With the data points occurring at every 1cm ^-1 over this range results in 1001 data points. With samples being taken at regular time intervals in a process the amount of data rapidly becomes large. PCA aims to describe the information using fewer variables than in the original data set. The first variable generated by PCA is the first principal component (PC1) and this aims to describe most of the variation in the data. The second variable (PC2), known as the second principal component, aims to describe as much as possible of the remaining variability that has not been described by the first principal component. The second principal component (PC2) is orthogonal to the first principal component (PC1) in PCA space. Principal Component Analysis is a known tool for analysing large, related data sets. For the convenience of the reader we describe below in further detail how the process is applied to such data sets. Figure 3a is a diagram showing schematic FTIR absorption spectra of a number of samples. Some variation between spectra for the samples can be seen. The analysis may include pre-treatment of the data by centring the data about a mean and optionally multiplicative scatter correction or de-trending. Principal Component Analysis then fits the first variable or first principal component (PC1), to describe as much of the variation in the data as possible.

Mathematically PCA analysis can be described using the equation:

X = T P ^T + E where X is the starting matrix of data in which, for example, each row contains the FTIR spectra values for a particular sample and the columns represent the spectra values across the various samples at a particular wavelength. Hence, for seven spectra, each having 1001 data points, the matrix may have 1001 columns and 7 rows. P is the loadings matrix which identifies the principal component direction in PCA space. T is known as the scores matrix and represents the coordinates of the samples in PCA space. Usually the number of principal components is much less than the number of samples or data points. When plotted in PCA space the data points may not lie exactly on the principal components. The differences between the principal components and the data makes up the residuals matrix which is identified by E in the above equation.

Figure 3b shows a set of data that has been applied to PCA space. The principal components PC1 and PC2 are shown and together form a plane. It is difficult to represent more than two principal components on a two-dimensional paper or screen figure but there would usually be more than two principal components. Towards the top of figure 3b a data point D1 which lies away from the plane formed by the principal components is projected on to the plane by a line. This line represents the residual E of the fit of the data to PC1 and PC2.

Returning to the method of figure 2, a PCA model for the FTIR data for the various samples is generated at step 120. The PCA model will have a number of principal components and the distance of each data sample from the principal components, as mentioned above, are known as residuals.

At step 140 of figure 2 samples that have been collected from the process are treated in some way to change them. This treatment will be related to a characteristic or component of interest in the process. The treatment may be a mechanical treatment such as filtering, a chemical treatment such as washing or reacting with a chemical, a physical treatment such as grinding or heating, or another type of treatment. When the samples have been treated the FTIR spectra of the treated samples are collected 150 and processed in the same way as the data that was input to produce the PCA model. The data can then be compared or evaluated against the model, such as by projecting 130 the data on to the PCA model and the difference between the data and the principal components determined. In the example of figure 2 using PCA analysis the difference is known as residuals. The residuals, as determined at 160, may be a quantitative indicator of the characteristic or component of interest in the process. In one example, the treatment may be to wash the samples to remove a species of interest from the samples. As an alternative, the scores of the data projected against the model may be used. The scores provide an indication of the fit of the data to the model. The closeness of fit, as determined by the scores, or the level of difference, as determined by the residuals, provide different ways of comparing the data to the model. The choice of which to use may depend on the number of samples, how the model is generated and the type of confidence of fit analysis it is desired to use in assessing the fit of the model.

In one example, the characteristic of interest is to quantify the amount or polar organic molecules there are in the samples. In this case the samples are washed with the dichloromethane (DCM) which removes polar organic molecules. The calculated residuals for the treated samples will provide an indication of the amount of polar organic molecules present. A variation between samples will allow, for example, samples that have high levels of polar organic molecules to be determined. This may be tracked back to the time at which the sample was collected from the process and the associated process conditions or batch determined. Action on future batches to increase or decrease the amount of polar organic molecules as desired for optimisation of the process.

Figure 4 shows a related method to that of figure 2 but for which two treatments are performed on the samples. Similarly to figure 2 samples are collected from a process at step 200. The FTIR spectra are collected of the samples and Principal Component Analysis, step 210, is performed on the spectra to generate a PCA model, at step 220. Preferably, multiple FTIR spectra are collected of each sample, such as a minimum of three but preferably ten. Multiple samples are also preferably used for generating the model. The samples are then each divided such that for each sample a first part can be treated using a first method and a second part can be treated separately using a second method. In other words there are two sets of samples; one for each treatment method. In figure 4, the first set of samples are treated, at step 230-1 , with a first method treatment which may be to remove a first compound or species of interest or change the sample in some way. For example, as discussed above this may be a treatment by washing the samples with DCM to remove polar organic molecules. Spectra of the samples treated by the first method are generated, at step 240-1 , and the spectra are processed and projected on to the PCA model at 250-1 . The second set of samples are treated by a second method different to the first method. This second method of treatment may be to remove a second compound or species of interest or change the sample in some way but different to the first treatment method. For example, the second treatment method may be to treat the samples by washing with hexane to remove non-polar organic molecules. Spectra of the second set of treated samples are collected, at step 240-2, and the spectra are processed and projected on to PCA model at step 250-2.

As shown at step 260-1 , a residual, often called a Q-residual or an x-residual , can be calculated for the samples treated by the first method, and similarly for the second samples at 260-2. In the following we will refer only to Q-residuals for convenience. The value of the Q-residual from the first treated samples can be used as an indicator, as shown at step 270-1 , of the level of the first chemical or species of interest, for example, that removed by the washing with DCM, namely polar organic molecules. Similarly, the value of the Q-residual from the second treated samples can be used as an indicator, as shown at step 270-2, of the level of the second chemical or species of interest, for example, that removed by the washing with hexane, namely non-polar organic molecules. Also as shown at step 280 the two sets of residuals can be used together such as to provide two-dimensional cluster information on the samples

Figure 5 shows example residuals information calculated for 10 samples where two types of washing have been performed. The residuals information is shown as a onedimensional bar chart, that is, the relative amounts for each sample are shown on the same axis. The two types of washing performed are the two examples mentioned above, namely: washing with DCM and washing with hexane. The bar-chart shows that three measurements/spectra were taken for each sample for projection onto the model. As can be seen in figure 5, sample 10 has high levels for the DCM washing indicating that sample 10 had high levels of polar organic molecules. Similarly, sample 2 also has high levels of polar organic molecules, but sample 2 also has the highest level for the hexane washing indicating it also has a high level of non-polar organic molecules. Figure 6 is a two-dimensional plot of the example data shown in figure 5 for 10 samples. Each sample is plotted based on the residual from the first washing and the residual from the second washing. In other words the data is plotted with the residuals for the DCM washing plotted along the x axis and the residuals for the hexane washing plotted along the y axis. In figure 6 it is easier to see that samples 2 and 10 are outliers in comparison to the rest of the samples. Sample 2 has higher residuals for both polar and non-polar organic molecules. Sample 10 has a high residual from the DCM washing indicating a high level of polar organic molecules. Figure 6 also shows that sample 1 was relatively high in non-polar organic molecules, and this was not readily apparent in figure 5. Hence, this 2-dimensional plot makes it easier to identify samples that differ, or are outliers, from the majority of samples. The remaining samples, namely samples 3-9 are clustered together with low values of residuals from both washings.

Although figures 5 and 6 relate to two treatments, they may be expanded to more treatments although actually displaying the information graphically becomes difficult beyond three dimensions.

The use of two or more treatments allows the user to differentiate between samples and start to understand what compounds and species are present, not just the relative quantities.

Although the 2-dimensional plot shows clustering of samples, it is not always possible to plot and readily determine which samples are clustered together. This is especially true with an increasing number of samples and more different treatments that may add further dimensions to plots such as that of figure 6. Clustering algorithms may be used for more complex cases. Figure 7a shows clustering that has been performed by an algorithm using the data from figure 6. The clustering algorithm used here is an HCA (Hierarchical Cluster Analysis) clustering using Ward’s method and PCA compression. Each sample comprises three measurements. Hence, the 10 samples result in 30 data points which are numbered as 1-30 with the numbers 1-3 corresponding to sample 1 , numbers 4-6 corresponding to sample 2 and so on up to numbers 28-30 corresponding to sample 10. The clustering algorithm operates on the Principal Component Analysis processed data to look at the distances between data points in PCA space. The closer the distance the more likely they are to be considered a cluster. Graphically clustering may be represented by a dendogram such as that shown in figure 7a. A dendogram links data points together by a line to represent their proximity. Clustering is then determined by the distance between the data points and clusters. By drawing a line at a particular distance the cluster groups may be determined. Hence, in figure 7a if we draw a line at a weighted distance of 8 along the x axis, three or four cluster groups may be identified. The top group in the figure comprises data points 26, 25, 21 , 20, 27, 19, 12, 11 and 10. These are the data points of samples 4, 7 and 9 and are identified as cluster 3 in the table at figure 7b. The middle group in figure 7a comprises data points 29, 30, 28, 6, 5 and 4. These are the data points of samples 2 and 10 and are identified as cluster 2 in the table at figure 7b. Based on the dendogram sample 2 and sample 10 may actually form their own separate groups. This can be seen in figure 7c, which is the same as figure 6, and the distance between samples 2 and 10 can be seen to be relatively large. Returning to figure 7a, the lower group in the figure comprises data points 24, 23, 22, 14, 15, 13, 18, 17, 16, 8, 9, 7, 3, 2 and 1 . These are the data points of samples 1 , 3, 5, 6 and 8 and are identified as cluster 1 in the table at figure 7b.

Although we have described how the analysis and treating of samples can be used to provide quantitative and qualitative information on the contents of samples taken from a process, the methods may also be applied to calibration of a spectrometer that is located in-situ at a process as we will now describe.

Calibration

Figure 8 is a block diagram of a method for calibrating an in-situ process spectrometer, that is a spectrometer that is coupled to the process for directly measuring the process. The method uses both the in-situ process spectrometer and a laboratorybased or remotely located spectrometer. This could include a spectrometer located at the process location, that is at the process line, but not connected to the process.

Once a spectrometer has been installed at the process and can start acquiring spectra then the spectrometer can be set up to monitor for the presence of chemical compounds, types of compound or species of interest. The spectrometer may be set up to access the process through a window such as a diamond window. Spectra of the process are collected by the in-situ spectrometer, as indicated at step 310 of figure 8. The data is stored, as indicated at step 320 of figure 8.

The user will also need to take a sample from the process for off-line analysis using the laboratory spectrometer. This is indicated at step 300 in figure 8. The time at which the sample is taken is noted so that it can be correlated with a spectra taken by the in-situ spectrometer at the same time. Since the constituents in the process may change with time the time-stamp is used to check that the two measurements are measuring the same constituents. There are two aspects that can be taken into account by the calibration. The first is that the sample that is taken off-line and measured at the laboratory may change because of a different temperature and pressure at the laboratory instrument compared to in the process. The second aspect is to set up the spectrometer to analyse the spectra generated by the in-situ device to monitor for chemicals or characteristics of interest.

The sample that has been collected is analysed by the laboratory or at-line spectrometer, as indicated at step 400. The temperature of this analysis is not important but later when a spectrum is collected based on treated samples the temperature at which that scan is taken must be the same. Preferably, the temperature of the original scan and the treated scan would be thermostatically controlled to 20°C, although it may be higher if required. For example, some samples may only be liquid at higher temperatures.

In order to be able to build a PCA model multiple spectra must be recorded, this should be at least three and preferably ten spectra. Spectra from multiple samples may also be used to build the model. As indicated at step 420 of figure 8, the PCA model can then be built, such as in the manner set out earlier in the description. The PCA model can be built to cover the wavelength range over which the spectrometer operates or it could be a reduced range. For example, the wavelength range may be 680-4000cm ^-1 or preferably 800-1800cm- ¹ .

Before processing, the data from the spectrometer may be pre-processed. This may include mean-centring and may also include any of: standard normal variate (SNV) processing, multiplicative scatter correction (MSC) or de-trending.

The fit of the PCA model to the at least three or more preferably ten spectra can be analysed using Hotelling’s T-Squared analysis. As described above, the residuals represent the difference between a sample and its projection though the model. Hotelling’s T-Squared represents a measure of the variation in each sample within the model, that is, it indicates how far each sample is from the centre, or zero scores, of the model. Since all of the spectra are collected from the same sample then there should be very little variation.

Similar to the steps of figures 2 and 4 the sample then goes through the treatment. This is shown 440 in figure 8. The treatment may be to remove chemicals of interest, for example, polar and non-polar organic chemicals as described earlier in this application or may include other treatment processes. In the case of the removal of polar and non-polar organics, this is performed by washing the hexane and DCM respectively. Both of these solvents are immiscible with water so can be used to wash aqueous samples. The method now continues much as for figures 2 and 4. The next step is to collect spectra of the treated samples, which is indicated at step 450. This is done at the same temperature as the spectra of the raw samples were collected at step 400. The spectra then undergo any preprocessing and are projected on to the PCA model generated from the raw samples. This is shown at step 430 in figure 8. As described above for the raw samples, at least three and preferably ten spectra are collected of the raw sample to allow the PCA model to be generated. The same number of repeat spectra are collected for the treated sample at this point. The steps of taking a sample and treating it may be repeated to build up sufficient data, as required. This indicated at step 500 in figure 8.

Figure 9 comprises three plots that illustrate the projection of the data from the DCM treated samples and the hexane treated samples to the PCA model generated from the raw samples. Figure 9a shows the residuals for the DCM and hexane washed samples plotted with their Hotelling’s T-squared value. In the plot the three diamonds are the data points for the DCM treated samples and the squares are the data points for the hexane treated samples. The residuals for the hexane washed samples can be seen to larger. For both groups of samples the Hotelling’s T-squared value is very small and much less than 1 . This indicates that overall the samples are broadly in line with the untreated samples. The high values for the residuals indicated a significant spectral difference. This is that a chemical has been removed. Figures 9b and 9c provide information on scores (T) for the treated samples in relation to the PCA model. The values of Hotelling’s T-squared and/or the residuals are recorded and used later in the calibration method. The recording of the residuals is indicated at step 460 in figure 8.

We now move to the phase of the method of using the collected data to build the calibration for the process spectrometer. The residuals data from step 460 of figure 8 are combined with data from the raw samples from step 320 in figure 8. The two groups of data are combined to build a model or algorithm linking the raw data to do the residuals. As a reminder the residuals are a measure of how much of a component of interest has been removed from the sample. Written mathematically we are looking to build a function or model linking x which represents the data from the raw samples with y which is the residuals data and may be represented as: y = f(x)

Such a function or model may be built using many techniques, for example regression analysis. A preferred technique used here is that of partial least squares (PLS). Figure 10 shows four plots for the fitting of data to the hexane treated samples. Figure 10a is a similar plot to figure 9a which shows the residuals for hexane treated samples. There is more variation in the residuals. There is also a greater variation in Hotelling’s T-squared. This difference is expected since some chemicals or species have been removed by the hexane washing.

Figure 10b provides an indication of which samples are outliers in the distribution of y-values, which are studentized residuals. If the samples lie outside of the dashed lines (95% confidence limits) it suggests there may be unexplained variance in the data. The x- axis indicates “leverage” which is a measure of how much a given point affects the calibration. A high leverage means a single point might have a disproportionally high impact on the calibration. The leverage on the x-axis of figure 10b shows no data is unduly influencing and the y-axis data shows reasonably clustered data with no data outside confidence limits. This provides good confidence that the data is reliable.

Figure 10c illustrates the match between the model’s prediction for hexane washed samples and the actual data collected. In this case a strong calibration is achieved, with an R ² of cross validation of 0.928, and an average error (defined as root mean squared error of cross validation - RMSECV - divided by the median measured value x 100) is:

6.98 x 10 ²⁸

X 100 = 23 %

3.06 x 10 ²⁹ which is a reasonable value (“good” calibrations are < 10 %, but anything < 30 % is usable). These figures are found in figure 10c, except for the median value which is 3.06x10 ²⁹ as indicated in the above-equation. Figure 10d shows the scores for the data in relation to the latent variables of the fit determined by the PLS. The latent variables (LV) are similar to the principle components for PCA analysis. As can be seen there is a spread of the data between LV1 and LV2 which is expected for a set of data of this kind. The dashed oval line represents a 95 % confidence limit. The majority of the data is within the limit and the few data points that are outside the limit are so close to the 95% limit that it is not a problem to include them.

The calibration process described above can be run in real time, enabling the use of statistical difference modelling on a spectrometer installed in a process where off-line reference data is not available. Hence, calibration data can be built to give real-time information on constituents in a process stream such as whether the process has a high amount of polar or non-polar organic molecules.

Accordingly, the final step in the method is to upload the calibration to the in-situ process spectrometer, as set out at step 480 of figure 8. A result of the spectrometer which is in-situ at the process and calibrated to provide information on characteristics of the process in real-time is that the process can be adjusted to compensate for undesirable changes in the process constituents or can be adjusted to draw off high value components to prevent them from being lost. Figure 11 provides a plot of an example where the characteristic is the amount of polar residual in the process. As shown in the example of figure 11 the amount changes with time. It is desirable to keep the amount of polar residual below a high-level limit. For example, this may be an amount measured in a paper pulp making process and the polar residual is an indicator of the amount of tall oil that is about to be sent to a boiler to be burned. It is desirable that the tall oil is removed from the process because it is valuable and can be sold. It may be known that the settling time needs to be increased to allow the tall oil to separate out to prevent the tall oil being lost. Hence, the process parameter of settling time may be increased to prevent loss of the tall oil. In other processes other process parameters may be adjusted. Following the adjustment to the settling time the amount of polar residual reduces and this can be seen in the example figure 11 where the signal level reduces to below the acceptable limit.

Other processes may use other characteristics and other measures to bring the process parameters back to an acceptable level. For example, chemical agents may be added to disperse stickies in the paper-recycling industry.

In general the output from the spectrometer, or an analyser connected thereto, may be provided to a PID controller to iterate the process back to acceptable limits.

Other example methods for removing and characterising constituents in a process may include:

1 . Washing edible oils with a) water and/or b) phosphoric acid. The former removes soaps, hydratable phospholipids and some water soluble fatty acids and the latter removes non-hydratable phospholipids. These could be useful to determine an amount of soap, phospholipids and/or amount of downstream gum formation;

2. To remove heavy metals from diesel oil it may be possible to put the oil through an ion exchange column or use EDTA (Ethylenediaminetetraacetic acid);

3. Remove silver from a solution by precipitating in out by adding hydrochloric acid or sodium chloride; and

4. Heating organic materials to remove volatiles. Comparison

We previously discussed the process of obtaining extractives by washing black liquor from the paper-pulp making process with acetone and then drying the result. Figure 12 is a plot providing a comparison between the extractives method and separate washing of samples using DCM and hexane followed by determination of residuals using PCA as discussed above.

The horizontal axis in figure 12 shows the amount of extractive in grams extracted per kg of black liquor for a given sample. The vertical axis shows the calculated residuals for samples taken at the same times as those for the extractives. Residuals for DCM washing and hexane washing are shown. Ten samples were taken over a period, for example, over two weeks. For each set of data the FTIR spectra were collected and processed three times such that there are three data points for each sample. The results suggest that some of the samples show a reasonable correlation between extractives and residuals. In other words, this shows that the chemicals removed in the washing phase are the same as those in the extractives procedure. One of the samples has a much higher extractives readout (30.3 g / kg) which suggests that the chemicals that cause the high extractive readout are more soluble in water than either DCM or hexane.

The process of determining residuals can be used instead of the extractives process, and because the residuals process can be performed in real-time it can allow the process to be adjusted to optimum outputs more responsively.

The person skilled in the art will readily appreciate that various modifications and alterations may be made to the above described methods, apparatus and computer programs. For example, different spectroscopic methods, decompositional analysis or multivariate analysis may be performed. The spectroscopic methods have principally been described as using FTIR spectroscopy but other mid-infrared absorption spectroscopy may alternatively be used. Furthermore, the methods described herein may be applied to many other industrial process, industrial biotechnological processes or the like.