Method for spectral determination and use of the method

Abstract

The invention relates to a method for spectroscopic determination and the use of the method. The invention discloses a method for spectroscopic determination of at least one measured variable in a medium, comprising the following steps: defining an application; providing a spectrometer; providing a raw spectrum, in particular by measuring the medium with the aid of the spectrometer; applying a calibration to the raw spectrum and obtaining a calibrated spectrum; normalizing the calibrated spectrum and obtaining a normalized spectrum, wherein the normalization is performed with the aid of spectrometer type-specific information and a transformation function, wherein the transformation function comprises the calibrated spectrum and additional measurement information as input variables, and comprises a spectrometer-specific transformation function and a calibration characteristic; selecting, adapting if necessary, or creating a model which is independent of the spectrometer and application-specific from the normalized spectrum and application-specific data; and determining the measured variable from the raw spectrum with the aid of the model. The invention also relates to the use of the method with a Raman spectrometer.

Description

Technical Field

[0001] This invention relates to a method for determining the spectrum of at least one measured variable in a medium. The invention also relates to the application of said method. Background Technology

[0002] In spectroscopy, the raw spectrum—that is, the pure measurement—is mapped onto the measured variable through a “model.” For example, the frequency-correlated effective cross section is mapped to the absolute concentration of a particular substance in the medium. This applies to all spectroscopic methods, including Raman spectroscopy.

[0003] Such models are typically only effective for the equipment configuration used to collect spectra for model training. Once one or more parts of the measurement equipment are replaced, the model loses its effectiveness and must be recreated, often involving significant experimental effort. Partial or complete replication of the experiments used to record training spectra requires substantial time and potentially high costs, for example, in the biopharmaceutical field, for providing the measurement media across all the variations required for model creation, as well as the personnel and instrumentation for proper training. Creating new models requires highly qualified experts with domain knowledge of the measurement media, spectroscopic methods, and model creation itself. Summary of the Invention

[0004] The purpose of this invention is to simplify model generation.

[0005] This objective is achieved by a method for determining the spectrum of at least one measurement variable in a medium, the method comprising the following steps: defining an application; providing a spectrometer; providing a raw spectrum, particularly by measuring the medium using the spectrometer; applying calibration to the raw spectrum and obtaining a calibration spectrum; standardizing the calibration spectrum and obtaining a standardized spectrum, wherein standardization is performed using spectrometer-type-specific information and a transformation function, wherein the transformation function includes the calibration spectrum and additional information about the measurement as input variables, and includes spectrometer-specific transformation functions and calibration characteristics; selecting, adapting, or creating, a spectrometer-independent and application-specific model based on the standardized spectrum and application-specific data; and determining the measurement variable based on the raw spectrum using the model.

[0006] This objective is achieved by a method for determining the spectrum of at least one measured variable in a medium, the method comprising the steps of: defining an application; providing a spectrometer; providing a raw spectrum, particularly by measuring the medium with the aid of the spectrometer; applying calibration to the raw spectrum and obtaining a calibration spectrum; selecting, optionally adapting, or creating a spectrometer-type-specific and application-specific model based on the calibration spectrum, applying specific data, and with the aid of at least one database, wherein the database includes models for calibration and non-standardized spectra; and determining the measured variable based on the raw spectrum with the aid of the model.

[0007] This typically simplifies model generation. For users, the protected method enables reduced costs and effort, and thus makes spectroscopic methods—such as Raman spectroscopy—accessible even to users who lack sufficient resources to create the model itself, even those that are practically difficult to establish.

[0008] The claimed invention eliminates the need to retrain the model for new instrument configurations, or at least significantly reduces the experimental effort required for model transfer. Additionally, it reduces the effort required for model adaptation (model structure, model weighting factors) for model verification and validation. Furthermore, time is saved because certification of the entire measurement process is no longer necessary or is greatly simplified (e.g., cGMP, i.e., the Food and Drug Administration's (FDA) Dynamic Good Manufacturing Practice).

[0009] One embodiment specifies that the raw spectrum is provided from a source other than the measurement by the spectrometer. In one embodiment, this is done based on literature or by means of another spectrometer (i.e., not the spectrometer used to measure the variable in the measuring medium). For example, the other spectrometer is a spectrometer of the same type / model. In one embodiment, the other spectrometer is a different type / model, particularly from a different manufacturer.

[0010] Logically, every spectrometer should first determine the raw spectrum. However, calibration from the raw data is usually already performed within the spectrometer, giving the user no direct access to the raw spectrum. In this case, processing begins directly with the calibrated spectrum. This is typically applicable to spectra from literature.

[0011] In one embodiment, without an explicit calibration step, the raw spectral output meets the stability criteria also required for the calibration spectrum. In this case, the raw spectrum corresponds to the calibration spectrum.

[0012] One embodiment specifies that the application considers the qualitative, preferably quantitative, determination of the composition of a medium by taking into account one or more substances to be determined—i.e., the measurement variables. Specific examples of this are alcohol concentration in wine production, glucose or lactate concentration in cell culture media, or amino acids in pharmaceutical product manufacturing. Further examples can be found downstream of the processing of mAbs (monoclonal antibodies) or the separation of proteins and protein aggregates (high molecular weight species) in pharmaceutical products. Similarly, methane in natural gas, , , It is possible to determine this.

[0013] One embodiment specifies that calibration includes wavelength and / or intensity. In Raman spectroscopy, for example, Raman shifts (typically expressed as wavenumbers on the x-axis of the spectrum) and intensity (typically expressed as arbitrary units (au) on the y-axis) are plotted.

[0014] One embodiment specifies that the calibration takes into account the measurement setup with a spectrometer and connected additional measuring equipment—particularly a measuring probe. Therefore, the calibration determined for the measurement setup is applied to the raw spectrum. The calibrated spectrum is then obtained. For the calibration of the measurement setup, the spectrometer can be considered on its own, or additional measuring equipment (e.g., a measuring probe, and optionally a process connection) can also be considered.

[0015] During the “normalization” of the calibration spectrum, any remaining differences (between spectrometers) after calibration are corrected. These differences can be, for example, systematic errors in calibration, different resolutions (samples per wavenumber), different measurement ranges (e.g., for Raman shift), different file formats of spectra from different equipment manufacturers, etc.

[0016] The standardized spectra of the application ideally appear exactly the same, regardless of the measuring device used to record them. Then, all residual variances originate solely from the application (aside from the residual errors from standardization).

[0017] If the distribution is known, statistical, non-systematic errors can also be considered. For this purpose, a set of normalized spectra is created from a single calibration spectrum. The set of normalized spectra contains the expected variance due to statistical dispersion around the assumed "true" spectrum. This set of normalized spectra is used in model training to make the model robust to the expected dispersion, enabling correct measurements despite the dispersion.

[0018] One embodiment specifies that spectrometer-type specific information includes information explicitly provided or derived from the spectrum, such as reference spectra (e.g., isopropanol), datasheet information, previous measurements, models, configurations, settings, etc.

[0019] A “spectrometer-specific transformation function” is a function that takes into account “spectrometer-type-specific information” to convert the spectrum from the device into a standardized spectrum.

[0020] For unknown devices, spectrometer-type-specific information must be comprehensive enough to create a transformation function. If a sufficiently generalized transformation function is known (e.g., because the spectrometer and application have been used in this combination many times before), it can be configured and applied with less information.

[0021] One embodiment specifies that, in addition to the transformation function, a distribution function is used, which includes the distribution of the residual error of the calibration, wherein the normalized spectrum includes the probability of dispersion around the calibration spectrum. The distribution function describes the distribution of the residual error of the calibration. In this case, the normalized spectrum is not a spectrum, but a probability statistically described by the dispersion around the “true” calibration spectrum. This dispersion can then also be part of the normalized spectrum or an associated set of spectra.

[0022] One embodiment specifies that the transformation function is taken from a transformation function database, wherein the transformation function database is formed by existing transformation functions.

[0023] One embodiment specifies that application-specific data includes information about the application, particularly the expected composition and measurement variables of the medium, temperature, pressure, and / or background matrix of substances that will not be determined. In one embodiment, application-specific data includes a small number of spectra from measurements with known concentrations, which are used for automatic model adaptation (e.g., zero measurement, span 0% and 100%, etc.).

[0024] One embodiment specifies that new or adapted models are stored in a database, particularly for models from standardized, non-standardized and / or calibrated spectra, and / or models are loaded from the database.

[0025] One implementation specifies that a standardized model will be converted into a calibration model, and vice versa.

[0026] One implementation specifies that more than one model may be used. In most cases, only one model will be used. If multiple models are used to compare results as a “joint” analysis, quality metrics may be considered, and the best / most reliable result may be selected, or there may be a combination of different results.

[0027] One embodiment specifies that reliability values ​​will be calculated and displayed for the model. The same or similar metrics used to determine how well the model fits the model are used for this purpose. In one embodiment, concentration can be used at the model level. If the concentration is within an acceptable range, the reliability is good enough for the application. Measurement errors in concentration determination can be used.

[0028] This objective is further achieved by applying a Raman spectrometer to the method described above. Attached Figure Description

[0029] Please refer to the attached diagram for a more detailed explanation.

[0030] Figure 1 The method for which protection is sought is shown.

[0031] Figure 2 Examples of some method steps of the claimed method are shown.

[0032] Figure 3 It symbolically represents standardization. Detailed Implementation

[0033] Figure 1 The method for which protection is sought is illustrated. Based on the application, a spectrometer, such as a Raman spectrometer, is selected. The spectrometer is used for the spectral determination of at least one measured variable in the medium.

[0034] A spectrograph is an optical instrument that disperses light of different wavelengths into its spectrum and records the resulting spectrum using a suitable detector. The instrument used in spectroscopy, and the visual observation of the spectrum, is called a spectroscope. A spectrometer is a device used to represent a spectrum. Unlike a spectroscope, a spectrometer provides the possibility of measuring the spectrum. In the context of this document, "spectrometer" refers to the entire device that ultimately outputs the measured variable as a concentration value from the spectrum via a model.

[0035] Therefore, the spectrometer produces the raw spectrum of the medium. In some cases, the raw spectrum may not be available to the user but is only used for further calculations. It is typically calibrated (see below), and this calibration is only shown to the user.

[0036] There are basically three possibilities, indicated by the figures labeled 1, 2, and 3.

[0037] In all variations, more than one spectrum is typically used. A set of spectra is created through experimental design and execution, representing the best possible trade-off between effort and coverage of all relevant variances from the expected measurement. For simplicity, unless otherwise stated, "spectrum" below always refers to multiple spectra.

[0038] In the left path, indicated by reference numeral 1 in the attached figure, models for the individual spectrometer and the individual application are created based on the spectrum. If this is complete, the model is selected. This produces a model precisely for this one spectrometer and precisely for this one application. The model provides the user with the measured values ​​of the measured variables.

[0039] In the intermediate path indicated by reference numeral 2 in the attached figure, the original spectrum is first calibrated. In one embodiment, this is done via wavelength and intensity. A model suitable for a specific spectrometer type and application is created from the calibrated spectrum. If the cluster already exists, the model is selected. If only minor details differ, the model is adapted. A corresponding database (see below) is also used in this process.

[0040] Model adaptation can be device-dependent, such as replacing components with the same components (e.g., replacing probes with the same probes).

[0041] The model's adaptation can be process-dependent, for example, through very similar applications (e.g., biological processes with essentially the same measurement task or with slightly altered process controls and corresponding variances in the spectrum, whose nutrient solution composition was not previously covered by the model's training dataset).

[0042] The model can be adapted to environmental parameters, such as applications in different factories, experimenters, and with different ambient air pressures / temperatures.

[0043] The adaptation of a model can be determined using measured variables; for example, a model already exists for two measured variables, but the user wants three, two of which are the same as the original model.

[0044] This results in spectrometer-specific and application-specific models that determine the measured variables.

[0045] The third path, 3, is described below. This path allows for the most probable and easiest handling of the spectrometer. The idea is that the user only specifies what will be measured. This method allows measurements to be performed with reduced effort, whether for purely technical reasons and / or for certification / auditing.

[0046] First, the original spectrum is calibrated as described above. Then, the calibrated spectrum is normalized. This is described below.

[0047] In the dashed path from the spectrometer to the normalizer, the raw spectrum without explicit calibration steps meets the stability criteria required for the calibration spectrum as well. In this case, the raw spectrum corresponds to the calibration spectrum and is then normalized.

[0048] Finally, depending on the initial conditions, the model is recreated, selected, or adapted based on the standardized spectrum. The result is a model that applies only to specific spectrometer types, but is independent of the spectrometer type.

[0049] During the “normalization” of the calibration spectrum, any remaining differences (between spectrometers) after calibration are corrected. These differences can be, for example, systematic errors in calibration, different resolutions (samples per wavenumber), different measurement ranges (e.g., for Raman shift), different file formats of spectra from different equipment manufacturers, etc.

[0050] Ideally, the standardized spectra of an application should appear exactly the same, regardless of the type of spectrometer used to record it. Then, all residual variance originates solely from the application, and if standardization is successful, its impact is negligible compared to acceptable measurement errors.

[0051] Figure 2 Examples of various method steps are shown. Figure 2 Based on calibration spectra. It can be seen that there are databases for both calibration and normalization spectra—that is, for path 2 and for path 3—(see...). Figure 1 Each of these is called a "library". Depending on the type of spectrum, there may be one database per application and spectrometer type, or one database per application. If the spectrum-specific transformation function is known, the databases can be converted to each other. If the transformation function is known, a direct conversion from calibration to normalized spectrum is possible. Only when the transformation function is uniquely invertible, considering all the meta-information on the particular spectrum, is a reverse path from normalized spectrum to calibration-only spectrum possible (without loss).

[0052] Meta-information includes, for example, analyte, background matrix, application, exposure time, integration time, temperature, pressure, hardware, and manufacturer information.

[0053] Artificial spectra can also be used in each case. These can be fully synthetically generated spectra or modified actual spectra, so-called enhanced spectra. An example of this is the aforementioned set of normalized spectra (e.g., to account for statistical, non-systematic errors). This set contains the expected variance due to statistical dispersion around the assumed "true" spectrum.

[0054] Data from the database, specifically spectra, can be used in the model. The spectra are selected from the database, allowing for the selection, modification, or creation of sufficiently good models for measurement tasks, with or without additional spectral measurements.

[0055] There are databases for calibrating and standardizing spectra.

[0056] There also exist databases for the models, namely at least one database for models with non-normalized spectra and one database for models with normalized spectra. There are one or more models for path 2 and path 3, see [link to database]. Figure 2 Transformation between models is possible. New models can be loaded into a database, while existing models can be retrieved from it. Without such a database, new training data (spectrums) must be generated for each application. If a model already exists, it can be simply selected or optionally adapted to achieve better performance. If the model does not exist, it is recreated. New models can be created based on existing models that are close to the model to be created. Therefore, spectra can be retrieved from a database to reduce the effort required to provide spectra for model selection / adaptation / creation. Alternatively, one or more models can be retrieved from a database to reduce the effort involved in model adaptation or model creation.

[0057] In one embodiment, a new or adapted model is loaded into the database.

[0058] Models used for non-standardized spectra are often not transferable to other spectrometer types with acceptable accuracy.

[0059] Ideally, model selection is automatic and transparent to the user. This makes it as easy as possible for the user to achieve acceptable measurement results. In addition to the automatic model selection mentioned above, for further simplification, a database of models can also be provided to devices in a way that is retrievable by the device or available in the cloud / at a central location.

[0060] Figure 3 The standardization is symbolically shown. Standardization is performed by the "spectral normalizer".

[0061] With perfect calibration, only normalization is needed to normalize the resolution of the x-axis and / or y-axis. However, in most cases, normalization is additionally required to compensate for residual errors remaining in the calibration spectrum.

[0062] If the calibrated spectrum already meets the requirements of the standardized spectrum, the standardization step can simply consist of transferring the data. Then, only metadata can be added.

[0063] After calibration, the following defects and changes may occur: calibration tolerance, resolution, range, or SNR. However, the normalizer cannot add back lost information. The normalized spectrum cannot be superior to the calibration spectrum (unless additional information about the device exists), and can only make the spectrum equally "bad," not better. Noise can only be removed to a limited extent. The informational content of the spectrum relevant to the calculation of evaluations and measurement results in the model must not be negatively affected by noise suppression. However, if the measurement is not resampled, the resolution can be adjusted. The measurement range can be reduced. Systematic and known calibration errors can be compensated for.

[0064] In one embodiment, a normalizer is supplied with the model. In the best-case scenario, the user does not notice the pre-normalized spectrum of the model, but the model automatically becomes more robust and general.

[0065] Each spectrometer can have one or more normalizers.

[0066] Normalized spectra can be transferred from the first analyzer to the second analyzer and vice versa via "inverse normalization." For this purpose, a normalized spectrum initially measured using the first spectrometer type is converted to a format corresponding to the calibration spectrum of the second spectrometer type using an inverse normalization function of the second spectrometer type. This means that the model initially created for the calibration spectrum of the second spectrometer type can also be used for the spectrum of the first spectrometer type. A reverse route from the second to the first is also possible. In both cases, the invertibility of the transformation function must be provided.

[0067] The normalizer receives the calibration spectrum, spectrometer-type specific information, and a transformation function as input values. The output value is the normalized spectrum.

[0068] Spectrometer type-specific information consists of information about the spectrometer type that is explicitly provided or derived from the spectrum—for example, reference spectra, datasheet information, separate measurements, etc.

[0069] For transformation functions, there can also exist a database (“library”) containing existing transformation functions.

[0070] The transformation function is formed from the calibration spectrum and calibration characteristics. The transformation function is extracted from these characteristics to generate a spectrometer-type-specific transformation function.

Claims

1. A method for determining the spectrum of at least one measured variable in a medium, comprising the following steps: - Define the application; - Provides spectrometers; - Provide the original spectrum, particularly by measuring the medium using the spectrometer; - Apply calibration to the original spectrum to obtain the calibrated spectrum; - The calibration spectrum is standardized to obtain the standardized spectrum. Standardization is performed using spectrometer-type-specific information and transformation functions. The transformation function includes the calibration spectrum and additional measurement information as input variables, and includes spectrometer-specific transformation functions and calibration characteristics; - Select, adapt, or create spectrometer-independent and application-specific models based on the standardized spectra and application-specific data; and -The measurement variable is determined based on the original spectrum using the model.

2. A method for determining the spectrum of at least one measured variable in a medium, comprising the following steps: - Define the application; - Provides spectrometers; - Provide the original spectrum, particularly by measuring the medium using the spectrometer; - Apply calibration to the original spectrum to obtain the calibrated spectrum; - Based on the calibration spectrum, specific data is applied, and at least one database is used to select, adapt, or create a spectrometer-type-specific model, whereby the database includes models for calibration and non-standardized spectra; and -The measurement variable is determined based on the original spectrum using the model.

3. The method according to claim 1 or claim 2, in, The calibration includes wavelength and / or intensity.

4. The method according to any one of the preceding claims, in, The calibration takes into account the measurement setup with the spectrometer and connected additional measurement equipment—particularly measurement probes.

5. The method according to any one of the preceding claims, in, The application considers the qualitative, preferably quantitative, determination of the composition of the medium by one or more substances to be determined—that is, the measurement variables.

6. The method according to any one of the preceding claims, in, In addition to the transformation function, a distribution function is used, which includes the distribution of the residual error of the calibration, wherein the normalized spectrum includes the probability of dispersion around the calibration spectrum.

7. The method according to any one of the preceding claims, in, The transformation function is designed to be configurable, wherein the characteristics of the normalized spectrum are adjustable, particularly the wavenumber range and resolution of the initial spectrum.

8. The method according to any one of the preceding claims, in, The spectrometer-type specific information includes information explicitly provided or derived from the spectrum, such as... -Reference spectrum, such as isopropanol, - Data sheet information, especially the measurement range expressed in wavenumbers. -Previous measurements, -Configuration, -set up, - Known and / or systematic calibration errors, which can subsequently be compensated in the calibration spectrum, particularly due to nonlinearities in wavelength axis or intensity calculations. -Temperature correlation, - Artifacts in the spectrum, particularly those originating from optical elements in the device or the probes or process connections, the electronics in the device, or signal processing. -wait.

9. The method according to any one of the preceding claims, in, The transformation function is taken from a transformation function database, which is formed by existing transformation functions, and / or The modified transformation function is stored in the transformation function database.

10. The method according to any one of the preceding claims, in, The application-specific data includes information about the application, particularly the expected composition of the medium and the measured variables, temperature, and pressure.

11. The method according to any one of the preceding claims, in, New or adapted models are stored in the database, particularly for models derived from normalized, non-normalized, and / or calibrated spectra, and / or The model is loaded from a database.

12. The method according to any one of the preceding claims, in, Reliability values ​​are calculated and displayed for the model.

13. Application of the method according to any one of the preceding claims using a Raman spectrometer.

Images