Spectrometric measurement method

Abstract

A spectrometric measurement method for determining, in measurement situation(s) within a predetermined application, measurement results of a measurand(s) of a medium of an application-specific type, wherein each measurement situation is predetermined by specifying at least one influencing factor that affects measured spectra in the respective measurement situation, comprising: based on universal training data, determining a universal model for determining predictions for each measurand; for each measurement situation, based on the universal model and supplementary training data including at least one supplementary spectrum determined in the respective measurement situation and corresponding supplementary value(s) of each measurand, determining a dedicated model for determining measurement results of each measurand in the respective measurement situation;and based on measured spectra determined in the respective measurement situation and the dedicated model, determining measurement results for each measured quantity of the medium.;

Description

[0001] The present disclosure relates to a spectrometric measurement method in which, in a predetermined application, measurement results of at least one measurand of a medium of an application-specific type are determined using spectrometers of a predetermined type.

[0002] Spectrometers of various types are currently used in a wide variety of applications, including industrial applications such as the chemical, pharmaceutical, and oil and gas industries, as well as in biotechnology and laboratory applications, to determine and provide measurement results for various quantities. For example, Raman spectrometers are used to determine the concentration of at least one component contained in a medium, such as the methane and / or ethane concentration in a liquefied natural gas, and / or at least one other quantity of the medium. Similarly, absorption spectrometers, which measure absorption, are used in various applications to determine the concentration of at least one component contained in a medium.

[0003] Spectrometers typically include a light source that shines light onto a sample of the medium, and a spectrometric unit that receives the measurement light resulting from the interaction of the emitted light with the medium and provides raw spectra of the received measurement light. These raw spectra are usually provided to a signal processor, which determines the spectral values ​​of the raw spectra based on the raw spectra and an algorithm for determining the spectral values ​​of the measured spectra. The measured spectra are then provided, for example, to an evaluation unit, which determines the measurement results of the measured quantity(ies) based on a previously defined model for determining the measurement results based on the spectral values ​​of the measured spectra.

[0004] Models used in spectroscopy to determine measurement results of measurands are generally based on a detailed mathematical analysis of training data, including experimentally determined reference spectra of reference samples of the medium that have known reference values ​​of the measurand. However, determining these models is a laborious and time-consuming process, particularly due to the considerable number of reference spectra required, the complexity of the relationships between the spectral values ​​of the reference spectra and the reference values ​​of the measurand, and / or the influence of application-specific properties that affect the spectral values ​​and / or the spectral distribution of the reference spectra.

[0005] Accordingly, there is a desire to use the same model on multiple spectrometers at multiple measuring points in the same application.

[0006] However, different spectrometers exhibit different measurement characteristics. Although some progress has been made in the calibration of spectrometers, particularly in calibrating their spectral sensitivity, a certain degree of variability remains, linked to the technical characteristics of different spectrometers of the same type. This detrimental effect is particularly pronounced in Raman spectrometers, as variations in their technical characteristics have a significant impact due to the extremely low signal-to-noise ratio inherent in Raman spectroscopic measurements, which is attributable to the notoriously low intensity of Raman scattered light.

[0007] Furthermore, spectrometric measurements at different measurement points within a predetermined application, specified by the quantity(ies) to be measured and the application-specific medium, can be influenced by a variety of other factors. These factors include, for example, factors related to the technical characteristics of the spectrometers, the measurement setup, the processes carried out at the measurement points, the measurement conditions, and / or the properties of the application-specific medium that prevail at the different measurement points within the same application.For example, the measurement setup, measurement conditions and / or composition of a liquefied natural gas in a liquefied natural gas generation plant may differ from the measurement setup and / or composition of a liquefied natural gas in a liquefied natural gas cooling plant.

[0008] Consequently, measured spectra of the medium, determined with different spectrometers at different measuring points within the same application, exhibit variability due to different influencing factors affecting the measurements at the different measuring points.

[0009] Consequently, measurement results determined using the same model based on measured spectra obtained from different spectrometers of the same type at different measurement points within the same application will exhibit a measurement error caused by the variability of the measured spectra.

[0010] This measurement error can be reduced to some extent by determining the model based on training data that covers at least some of the variations of at least some of the known influencing factors that can occur within a given application. However, this leads to a significant increase in the effort required to determine and provide the training data for modeling. Furthermore, in practice, it is neither possible nor economical to determine and provide training data that covers the variations of all possible current and future influencing factors that must be considered for the predetermined application.Even if the model is determined based on training data that takes some of the variations into account, the measurement results determined with this model still have a measurement error due to the remaining variabilities that are not taken into account in the training data.

[0011] Accordingly, there remains a need for further contributions in this technical field.

[0012] For example, there is a need for a spectrometric method that uses spectrometers of a specific type to determine measurement results of at least one quantity of a medium of an application-specific type in several different measurement situations within a specific application, and enables a more efficient and accurate determination of the measurement results in each measurement situation.

[0013] As another example, there is a need for a spectrometric measurement method that is better suited to deal with the variability of measured spectra of the medium determined by spectrometers in different measurement situations within a predetermined application.

[0014] The present disclosure includes a spectrometric method for determining measurement results of at least one measurand of a medium of an application-specific type in one or more different measurement situations within a predetermined application using at least one spectrometer of a predetermined type, wherein each measurement situation is predetermined by specifying at least one influencing factor that affects measured spectra determined by the spectrometer(s) in the respective measurement situation; wherein the method comprises: Performing reference measurements by determining reference spectra of reference samples of the medium using at least one spectrometer, and determining and / or providing a reference value for each measurement parameter of the reference sample for each reference sample; based on universal training data including the reference spectra and the corresponding reference values, determining a universal model to determine predictions for each measurand based on measured spectra of the medium determined by spectrometers in the predetermined application; and Perform the following procedure steps for each measurement situation: in the respective measurement situation with at least one spectrometer, determining and providing measured spectra of the medium; for at least one supplementary spectrum or a limited number of supplementary spectra, wherein each supplementary spectrum is given by one of the measured spectra determined by the spectrometer(s) used in the respective measurement situation, determining and / or providing a supplementary value of each measured quantity of the medium; based on the universal model and supplementary training data including at least one or each supplementary spectrum and the corresponding supplementary value(s), determining a dedicated model for determining measurement results of each measurand based on measured spectra of the medium, determined and provided by the spectrometer(s) used in the respective measurement situation; and based on measured spectra of the medium, which are determined and provided by the spectrometer(s) used in the respective measurement situation, and the dedicated model, determining and providing measurement results for each measured quantity of the medium.

[0015] The method offers the advantage of enabling a more efficient and accurate determination of the measurement results of each measured quantity of the medium in each predetermined measurement situation within the predetermined application.

[0016] The universal model offers the advantage of taking into account the application-specific relationships between the spectral values ​​of the measured spectra and the measurand(s) in the predetermined application. Furthermore, the supplementary spectra and corresponding supplementary values ​​of each measurand, determined for each measurement situation, offer the advantage of reflecting the effect of characteristics of the respective measurement situation on the measured spectra determined in that situation.

[0017] The universal model, which already takes application-specific relationships into account, offers the advantage that determining the dedicated models based on the universal model and the supplementary training data is significantly less complex and demanding than determining the universal model itself. This means that only a small number of supplementary spectra and corresponding supplementary values ​​are needed to determine the dedicated model for each measurement situation, thus adequately considering the impact of characteristics of the respective measurement situation on the measured spectra determined in that situation.

[0018] Determining and subsequently applying the dedicated model for each measurement situation offers the advantage of achieving high measurement accuracy based on the dedicated model, without the need to determine the universal model based on universal training data covering the entire range of variability in the measured spectra that can occur within the predetermined application. This reduces the time and effort required to determine the universal training data for the universal model. Simultaneously, the measurement results determined with the dedicated models, which each take into account the effect of the characteristics of the respective measurement situation on the measured spectra determined in that situation, are significantly more accurate than the predictions for the measured quantity(ies) determined with the universal model in the respective measurement situation.

[0019] In certain embodiments, each dedicated model is determined such that the measurement results of each measurand determined with the dedicated model based on the or each supplementary spectrum correspond to the supplementary value of the respective measurand exhibited by the medium when determining the respective supplementary spectrum.

[0020] According to the first embodiment, the dedicated model is determined for at least one or each measurement situation by retraining the universal model based on the universal model and the supplementary training data, taking into account the supplementary training data, and determining the dedicated model given by the newly trained universal model.

[0021] In certain embodiments of the first embodiment, the retraining of the universal model is carried out in the same way as the training of the universal model and / or is based on the same underlying algorithm that was previously used to determine the universal model, and / or includes an enhancement of the effect of the circumstances prevailing in the respective measurement situation on the newly trained universal model by assigning a greater weight to the supplementary training data than to the universal training data.

[0022] According to a second embodiment, the dedicated model is determined for at least one or each measuring point by adapting the universal model according to the supplementary training data such that the measurement results of each measurand determined with the adapted universal model based on the supplementary spectra correspond to the supplementary values ​​of the respective measurand that the medium exhibits when determining the respective supplementary spectrum, and the dedicated model given by the adapted universal model is determined.

[0023] In certain embodiments of the second embodiment, the fitting of the universal model is carried out by determining the prediction of each measured quantity based on each complementary spectrum with the universal model, determining prediction errors shown by the predictions based on the complementary spectra and the complementary values, and performing at least one of the following: a) based on the prediction errors, adapting the universal model so that prediction errors of predictions determined with the adapted universal model based on the complementary spectra are minimized; b) Performing an iterative process of adapting the universal model based on the prediction errors and repeating the procedural steps of determining the prediction of each measurement based on each complementary spectrum with the adapted universal model, determining prediction errors shown by these predictions, and subsequently adjusting the aforementioned adapted universal model until the prediction errors of predictions determined with the resulting adapted universal model fall below a predetermined threshold; and c) Adapting the universal model by adjusting at least one weighting factor, at least one parameter, one filter, one smoothing algorithm and / or at least one other model component of the universal model, so that deviations between predictions determined with the adapted universal model based on the complementary spectra and the corresponding complementary values ​​are minimized.

[0024] According to a third embodiment, the dedicated model for at least one or each measuring point is determined as a model given by the universal model, which determines predictions for each measurand based on measured spectra determined by the spectrometer(s) used in the respective measurement situation, and a transformer that determines measurement results of the measurand(s) based on the predictions determined by the universal model; and the determination of the dedicated model includes: based on at least one or each complementary spectrum, the universal model determines the corresponding prediction of each measured quantity; and based on the prediction(s) determined on the basis of the supplementary spectrum(s) and the corresponding supplementary value(s), determining the transformer such that the measurement result of each measurand determined with the transformer based on the prediction(s) determined with the universal model based on the supplementary spectrum corresponds to the corresponding supplementary value of the respective measurand that the medium exhibited during the determination of the respective supplementary spectrum.

[0025] In certain embodiments of the third embodiment, the transformer is determined for at least one or each measuring point by determining, for each measured quantity, a transfer function for calculating measurement results of the respective measured quantity as a function, as a linear function, as a polynomial function of a certain order or as another mathematical function for predicting the respective measured quantity and a set of at least one parameter;wherein for each measurand at least one parameter of the transfer function is determined by fitting the respective transfer function to a set of data points, each data point is given by one of the predictions of the respective measurand determined with the universal model based on one of the complementary spectra and the corresponding complementary value, and the set of data points consists of a single data point, two data points, a limited number of data points less than or equal to 20, less than or equal to 10 or even less than or equal to 5 data points, or several data points.

[0026] In a fourth embodiment, which is given by a further embodiment of the third embodiment, the transformer included in the dedicated model is determined for at least one or each measuring point in the form of a transfer model for determining measurement results of the measurand(s) based on predictions determined with the universal model based on measured spectra determined by the spectrometer(s) used in the respective measurement situation, and the determination of the respective dedicated model based on transfer model training data including predictions of the measurands determined with the universal model based on the supplementary spectra and the corresponding supplementary value(s) of each measurand, including the determination and provision of the transfer model.

[0027] In certain embodiments of the fourth embodiment, determining the transfer model for at least one or each measuring point includes performing at least one of the following process steps: a) based on a detailed analysis of the transfer model training data, determining and providing an algorithm to calculate the measurement results of each measurand based on predictions of the measurand(s) determined by the universal model, based on measured spectra determined by the spectrometer(s) used in the respective measurement situation, and b) based on the training data of the transfer model, performing a multivariate analysis, partial least squares regression, support vector regression and / or principal component analysis of the predictions and / or a procedural step of quantitatively assessing dependencies between the predictions of the measured variables and the corresponding complementary values ​​of each measured variable; or Based on the transfer model training data, this includes training a neural network to determine the measurement results of the measured quantity(s) based on predictions determined with the universal model based on measured spectra determined in the respective measurement situation, and determining the transfer model to be provided by the trained neural network.

[0028] In further embodiments of the fourth embodiment, the transfer model is determined for at least one or each measurement situation based on supplementary training data including a limited number of less than or equal to 100, less than or equal to 50 or even less than or equal to 20 supplementary spectra and the corresponding supplementary values.

[0029] In a fifth embodiment, the dedicated model for at least one or each measuring point is determined as a model given by an adapted universal model that determines adapted predictions for each measurand based on measured spectra determined in the respective measurement situation, and an adapted transformer that determines the measurement results of the measurand(s) based on adapted predictions determined by the adapted universal model based on measured spectra determined in the respective measurement situation; and determining the dedicated model includes: based on the universal model and the supplementary training data, retraining the universal model taking into account the supplementary training data and determining the adapted universal model given by the newly trained universal model; based on at least one or each complementary spectrum with the adapted universal model, determining the adapted prediction of each measured quantity; and based on the adapted prediction(s) determined with the adapted universal model based on the or each supplementary spectrum and the corresponding supplementary value(s), determining the adapted transformer such that the measurement result of each measurand determined with the adapted transformer based on the prediction(s) determined with the adapted universal model based on the or each supplementary spectrum corresponds to the supplementary value of the respective measurand exhibited by the medium during the determination of the respective supplementary spectrum.

[0030] In certain embodiments of the fifth embodiment, the determination of the adapted universal model is carried out by retraining the universal model in the same way as the training of the universal model and / or based on the same underlying algorithm that was previously used to determine the universal model, and / or in a way that preserves a global validity of the newly trained universal model by assigning less weight to the supplementary training data than to the universal training data.

[0031] In further embodiments of the fifth embodiment, determining the adapted transformer for each measurand includes determining an adapted transfer function for calculating measurement results of the respective measurand as a function, as a linear function, as a polynomial function of a certain order, or as another mathematical function of the adapted prediction of the respective measurand and a set of at least one parameter, wherein for each measurand the at least one parameter of the adapted transfer function is determined by fitting the respective adapted transfer function to a set of at least one or more data points, wherein each data point is given by one of the adapted predictions of the respective measurand determined with the adapted universal model based on one of the complementary spectra and the corresponding complementary value.

[0032] According to a sixth embodiment, which is given by a further embodiment of the fifth embodiment, for at least one or each measuring point, the adapted transformer is determined in the form of an adapted transfer model for determining measurement results of the measured quantity(s) based on adapted predictions of the measured quantity(s) determined with the adapted universal model based on measured spectra determined in the respective measurement situation, and the adapted transfer model is determined based on transfer model training data including the adapted predictions of the measured quantity(s) determined with the adapted universal model based on the supplementary spectra and the corresponding supplementary values.

[0033] In a certain embodiment of the sixth embodiment, determining the adapted transfer model includes at least one of the following process steps: a) based on a detailed analysis of the transfer model training data, determining and providing an algorithm for calculating measurement results of each measurand based on adapted predictions of the measurand(s) determined with the adapted universal model, based on measured spectra determined in the respective measurement situation, and b) Performing a multivariate analysis, partial least squares regression, support vector regression and / or principal component analysis of the adapted predictions and / or a procedure step of quantitatively assessing dependencies between the adapted predictions of the measured variable(s) and the corresponding complementary values ​​of each measured variable; or Based on the transfer model training data, this includes training a neural network to determine the measurement results of the measured quantity(s) based on the adapted predictions determined with the adapted universal model based on the measured spectra determined in the respective measurement situation, and determining the adapted transfer model to be provided by the trained neural network.

[0034] In further embodiments of the sixth embodiment, specifying the at least one influencing factor for at least one or each measurement situation includes performing at least one of the following: specifying at least one influencing factor associated with a measurement setup; specifying the measurements to be performed on samples of the medium or which are to be in-situ measurements; specifying the measurements to be performed in a flow cell carrying the medium or in a container, bioreactor, or other type of vessel containing the medium; specifying at least one influencing factor associated with the measurement conditions; specifying a parameter range for at least one parameter and / or temperature; specifying at least one influencing factor associated with the application-specific type of medium within the respective measurement situation;Specify at least one property of the medium; specify that the medium is subjected to a specific process, a specific production process and / or a specific processing procedure; specify a facility that contains and / or processes the medium; specify that the medium includes at least one specified component, that it includes several specified components, or that it consists of specified components; specify the specific spectrometer that performs the measurements; and / or specify the specific measuring point where the measurements are performed.

[0035] In certain embodiments, the predetermined application is a biotechnological application wherein mammalian cells producing an active ingredient of a drug are cultivated in a cell culture medium, and at least one measurement situation is predetermined by specifying or by specifying alone the type of cell culture medium and / or the type of mammalian cells.

[0036] In further embodiments, the method further comprises, for at least one or each measurement situation, performing the process steps, based on the supplementary training data and the dedicated model, determining a measurement accuracy of measurement results determined with the dedicated model, and performing at least one of the following: a) Providing measurement accuracy; and b) in the event that the measurement accuracy is lower than a predetermined minimum accuracy, perform at least one of the following: b1) Determining additional supplementary training data, re-determining the dedicated model based on the supplementary training data and the additional supplementary training data, and determining the dedicated model given by the newly determined dedicated model, and b2) Limiting the respective measurement situation, which is given by a limited measurement situation, which is specified by specifying at least one supplementary influencing factor that affects measured spectra determined by the spectrometer(s) in the respective measurement situation, and based on the universal model and the supplementary training data determined in the limited measurement situation, determining the dedicated model for determining measurement results of each measurand based on measured spectra of the medium determined and provided by the spectrometer(s) used in the limited measurement situation.

[0037] In certain embodiments, the method further comprises at least once performing the process steps of updating the universal model based on the universal training data and the supplementary training data previously determined for at least one measurement situation; and subsequently determining the corresponding dedicated model based on the updated universal model for at least one measurement situation.

[0038] The described embodiments and other features, advantages and disclosures contained herein, and the manner in which they are achieved, will become clearer by reference to the following description of various embodiments of the present disclosure in conjunction with the accompanying drawings, and the present disclosure will be more readily understood, wherein: Fig. 1 shows a flowchart of a spectrometric measurement procedure according to the present disclosure; Fig. 2 shows a schematic representation of a spectrometer; Fig. 3 a method for determining a dedicated model by retraining a universal model according to the present disclosure; Fig. 4 shows a method for determining a dedicated model by fitting a universal model according to the present disclosure; Fig. 5 a method for determining a dedicated model including a universal model and a transformer according to the present disclosure; and Fig. 6 shows a method for determining a dedicated model including an adapted universal model and an adapted transformer according to the present disclosure.

[0039] The present disclosure includes a spectrometric measurement method in which, in at least one measurement situation or in several different measurement situations within a predetermined application, measurement results MR of at least one measurand of a medium of an application-specific type are determined using at least one spectrometer of a predetermined type.

[0040] A flowchart of the spectrometric measurement procedure is shown in Fig. 1 shown.

[0041] At the in Fig. The predetermined application of the process shown in Figure 1 is, for example, a specific process in the life science industry, the pharmaceutical industry, biotechnology, the oil and gas industry, the chemical industry, the food and beverage industry, or a process in another field.

[0042] In certain embodiments, the predetermined application is specified, for example, by the quantity(ies) to be measured and the application-specific type of medium. Depending on the predetermined application, the at least one quantity to be measured includes, for example, the concentration of at least one component contained in the medium, the pH value of the medium, the melting index of the medium, the cell motility of the medium, and / or at least one other property of the medium.

[0043] In certain embodiments, the predetermined application is, for example, a liquefied natural gas (LNG) application. In this case, the media of the application-specific type are liquefied natural gases, and the measured variable(s) includes, for example, the concentration of at least one component, e.g., methane (CH4) and / or ethane (C2H6), which is contained in the liquefied natural gas.

[0044] In other embodiments, the predetermined application is, for example, a biotechnological application in which mammalian cells producing an active ingredient of a drug are cultivated in a cell culture medium. In this case, application-specific media are, for example, cell culture media including mammalian cells, and the measured parameter(s) include, for example, a glucose concentration, a lactate concentration, a viable cell density of the mammalian cells contained in the cell culture medium, and / or at least one other property of the medium.

[0045] In the method according to the present disclosure, which is disclosed herein, each measurement situation is predetermined by specifying at least one influencing factor that affects the measured spectra determined by the spectrometer(s) used in the respective measurement situation. The specification of at least one influencing factor includes, for each measurement situation, at least one influencing factor associated with the measurement setup, at least one influencing factor associated with the measurement conditions, and / or at least one influencing factor associated with the medium of the application-specific type within the respective measurement situation.

[0046] As a further example, a specific measurement situation may be predetermined with regard to the measurement setup in such a way that it is limited to measurements on samples of the medium, to in-situ measurements, e.g., in a flow cell carrying the medium or in a container holding the medium, e.g., a bioreactor or another type of vessel. Additionally or alternatively, a specific measurement situation may be predetermined, e.g., in such a way that it is limited to measurements performed with specified equipment, e.g., a particular type of optical system, which is specified for the respective measurement situation.

[0047] As another example, a particular measurement situation may be predetermined with regard to the measurement conditions in such a way that it is limited to measurements that are carried out in a certain parameter range of at least one parameter, e.g. a certain temperature range.

[0048] As a further example, with regard to the measurement conditions and / or the properties of the medium, a particular measurement situation may be predetermined in such a way that it is limited to measurements of the medium undergoing a specific process, e.g., a specific production and / or processing procedure, and / or may be predetermined in such a way that it is limited to measurements of the medium carried out in specific facilities that contain and / or process the medium. For example, in the context of the aforementioned application for liquefied natural gas (LNG), a particular measurement situation may be predetermined such that limited measurements are carried out at LNG production facilities, LNG processing facilities, LNG cooling facilities, or LNG storage facilities.

[0049] As another example, with regard to the properties of the medium, a particular measurement situation may be predetermined in such a way that it is limited to measurements on media of the application-specific type that include at least one specific component, include several specific components, or consist of specific components.

[0050] As an example, in connection with the aforementioned biotechnological application, a particular measurement situation may be predetermined in such a way that it is limited to measurements carried out on media of the application-specific type, which include a specific cell culture medium, e.g. a cell culture medium provided by a specific manufacturer, and / or which include mammalian cells of a specific type, e.g. hamster ovary cells, human embryonic kidney cells, or mammalian cells of another type.

[0051] For each measurement situation, the flexibility provided by the type and / or number of specified influencing factors, as well as the degree of accuracy of their specification, can be used in various ways.

[0052] In certain embodiments, at least one or every measurement situation is predetermined, e.g., in a restrictive manner. This offers the advantage of limiting the variations of measured spectra determined in the restrictively predetermined measurement situation. A relatively restrictive specification of a measurement situation, leading to a correspondingly large limitation of the variations in the measured spectra, is achieved, for example, by specifying the specific spectrometer that determines the measured spectra and / or the specific measurement point where the spectrometric measurements are performed. Accordingly, in certain embodiments, at least one or every measurement situation is predetermined, e.g., by specifying the specific spectrometer that determines the measured spectra and / or the specific measurement point where the spectrometric measurements are performed.

[0053] Additionally or alternatively, in certain embodiments, at least one or each measurement situation is predefined, for example, in a broader manner. This offers the advantage of increasing the number of measurement points at which the respective measurement situation occurs. As an example, in connection with the aforementioned biotechnological application, at least one measurement situation can be predefined more broadly, for example, by specifying only the specific cell culture medium and / or the specific type of mammalian cells enclosed in the cell culture medium. Additionally or alternatively, at least one measurement situation within this biotechnological application can be predefined more restrictively, for example, as outlined above.

[0054] As in Fig. As shown in 1, the method according to the present disclosure includes a process step A1 of performing reference measurements by using at least one spectrometer to take 100 reference spectra I ref,n Several reference samples of the medium of the application-specific type are determined, and a reference value m is established for each reference sample. ref,n Each measured parameter of the reference sample is determined and / or provided.

[0055] The reference measurements are carried out, for example, by using universal training data including the reference spectra I. ref,n and the corresponding reference values ​​m ref,n for each measured quantity the application-specific relationship between the spectral values ​​of the measured spectra I m to reproduce the medium and the measured quantity(s) within a predetermined measuring range.

[0056] For example, the reference samples in connection with the aforementioned liquefied gas application include, for instance, samples covering a range of methane concentrations and / or a range of ethane concentrations. In certain embodiments, these reference samples include, for example, samples covering a range of different compositions of the liquefied natural gas.

[0057] As a further example, the reference samples in connection with the aforementioned biotechnological application include, for example, samples covering a range of glucose concentrations, a range of lactate concentrations, and / or a range of viable cell densities. In certain embodiments, these reference samples include, for example, samples covering a range of different cell culture medium compositions and / or a range of different mammalian cell types.

[0058] The spectrometer(s) 100 used to perform the reference measurements is preferably of the same type as the spectrometer(s) 100 used in the respective measurement situation.

[0059] An exemplary embodiment of a spectrometer 100 of the predetermined type is shown in Fig. 2 shown. The exemplary spectrometer 100 shown includes a light source 1 that sends light S0 into a measurement region 3 configured to receive a sample 5 of the medium, and a spectrometric unit 7 configured to receive measurement light S M receives, which results from an interaction of the transmitted light S0 with the medium, and is configured to receive raw spectra I raw of the received measuring light S M determines and provides.

[0060] In certain embodiments, the spectrometric unit 7 includes, for example, a disperser 9, e.g., a diffractive or holographic grating, which diffuses the incident measuring light S M scatters, and a detector 11 that detects the scattered measuring light S M receives. In certain embodiments, the detector 11 includes, for example, an array of detector elements, such as an array of charge-coupled devices (CCDs) or an array of photodiodes, each of which receives a portion of the scattered light and determines and provides a detector signal corresponding to the intensity of the portion of the scattered light received by the respective detector element. In these embodiments, the raw spectra I raw e.g. provided by the spectrometric unit 7 in the form of the detector signals provided by the individual detector elements.

[0061] Regarding the predetermined spectrometer type 100, which is in the Fig. In the method shown in Figure 1, each spectrometer 100 is, for example, a Raman spectrometer. In this case, the light source 1 of each spectrometer 100 is, for example, a monochromatic light source, such as a laser, configured to transmit excitation light S0 with a predetermined excitation wavelength, such as a wavelength in the visible or near-infrared wavelength range. In certain embodiments, each Raman spectrometer is configured, for example, with a filter 13, such as a notch filter, to receive light S1 emitted from the illuminated sample 5 and measure light S M including Raman scattered light emanating from the illuminated sample 5, to supply the spectrometric unit 7. Additionally or alternatively, the spectrometric unit 7 of each Raman spectrometer is configured, for example, to measure the measured spectra I mto determine and provide intensity spectra representing the spectral intensities of the Raman scattered light emanating from the illuminated sample 5 in a predetermined spectral range, e.g. a wavelength range or a wavenumber range.

[0062] Alternatively, in certain embodiments, the elements in the Fig. The spectrometer 100 used in the method shown in Figure 1 is of a predetermined type, e.g., an absorption spectrometer. In such an embodiment, the light source 1 is, for example, a broadband light source that emits light S0 with a broad spectral range through the sample 5 located in the measurement region 3, e.g., light S0 that includes wavelengths in the visible, ultraviolet, and / or infrared range. In these embodiments, the measured spectra I m of the measuring light emerging from region 3 L Me.g. as absorption spectra, which represent the spectral absorption of the medium as a function of the spectral line, or as intensity spectra, which represent the spectral intensity values ​​of the measurement light S M represent.

[0063] Another alternative is the one in the Fig. In the method shown, 100 spectrometers of the predetermined type were used, for example, a dispersive spectrometer, a tunable diode laser spectrometer, or another type of spectrometer.

[0064] Regardless of the type of spectrometer 100 used, each spectrometer 100 includes, for example, a signal processor 15, e.g., a computer, a microprocessor, or another type of computing unit, which is connected to and / or communicates with and is configured to process the measured spectra I mof the medium based on the raw spectra provided by spectrometric unit 7 I raw to determine and provide. Determining the measured spectra I m This is done, for example, by the signal processor 15 based on an algorithm ALG for determining the spectral values ​​of the measured spectra I m based on the spectral values ​​of the raw spectra provided by spectrometric unit 7 I raw The ALG algorithm is implemented by the manufacturer in every spectrometer of the respective type, for example.

[0065] The in the in Fig. The spectrometers 100 used in the method shown in the present disclosure are calibrated before being put into operation. For this purpose, for example, calibration methods known from the prior art, including methods for calibrating a spectral axis and / or a spectral sensitivity, are used, and the ALG algorithm is used to determine the spectral values ​​of the measured spectra. m adjusted and / or modified accordingly.

[0066] Following the reference measurements, the procedure further includes a process step A2, based on universal training data including the reference spectra I. ref,n and the corresponding reference values ​​m ref,n , of determining an application-specific universal model GM for determining predictions MP for each measurand based on measured spectra I mof the medium, which were determined by spectrometers of 100 of the predetermined type in the predetermined application.

[0067] With regard to determining the universal model GM, prior art methods can be used to determine models based on reference spectra and corresponding reference values ​​of the measured quantity(s). As an example, in certain embodiments the universal model GM is determined based on the universal training data including the reference spectra I. ref,n and the corresponding reference values ​​m ref,n The training of the universal model GM is carried out, for example, by developing an algorithm for calculating predictions MP for each measured quantity based on spectral values ​​of measured spectra I, based on a detailed analysis of the universal training data. mof the medium is determined and provided. In certain embodiments, the analysis of the universal training data and / or the determination of the universal model GM includes, for example, performing a multivariate analysis, a partial least squares regression, a support vector regression, and / or a principal component analysis of the universal training data, and / or a procedure step for quantitatively evaluating the dependencies between the spectral values ​​of the reference spectra I. ref,n and the corresponding reference values ref,n one for each measured quantity.

[0068] As an alternative, in certain embodiments, determining the universal model GM can include, for example, training a neural network to make predictions MP for each measurement based on the measured spectra I. mto determine the medium. In these embodiments, the trained neural network is then used as the universal model GM.

[0069] Following the preparatory process steps A1 and A2, the procedure further includes a sequence of process steps B1, B2, B3 and B4, which are carried out for each measurement situation.

[0070] The sequence includes a procedure step B1, in the respective measurement situation with at least one spectrometer used in the respective measurement situation, 100 measured spectra I m to determine and provide the medium. Each spectrometer 100 is used to determine the measured spectra I. m performs, of the same predetermined type as the spectrometer(s) 100 which performs / performs the reference measurements, and preferably also calibrated as described above.

[0071] For each measurement situation, the determination of the measured spectra I m e.g., performed at one or more measuring points that meet the specifications by which the respective measurement situation was predetermined. For example, for each measurement situation predetermined by specifying the specific spectrometer 100 and the specific measuring point where this specific spectrometer 100 is installed, each measured spectrum I m with this specific spectrometer 100 at this specific measuring point. As a further example, within the framework of the aforementioned biotechnological application, for the measurement situation or situations that are determined solely by specifying the specific cell culture medium and / or the specific type of mammalian cells contained in the cell culture medium, the measured spectra I me.g. determined with at least one or more spectrometer(s) 100 at one or more measuring point(s), wherein the medium of the application-specific type includes the specified cell culture medium and / or mammalian cells of the specified type.

[0072] The sequence also includes a process step B2, for at least one supplementary spectrum I k , each of which is determined by one of the spectra I measured by the spectrometer(s) 100 in the respective measurement situation m is given a supplementary value m k to determine and / or provide each measured quantity that the medium in the respective measurement situation uses to determine the respective complementary spectrum I k exhibits.

[0073] In certain embodiments, at least one or every supplementary value m kdetermined by at least one or each measured quantity, e.g., by measurement. In this case, the measurement of at least one or each supplementary value m is carried out. k e.g. on a sample of the medium which, at the time of determination of the respective supplementary spectrum I k was taken at the measuring point. In this respect, for example, highly accurate measuring devices and / or laboratory equipment are used to measure the supplementary values. k Each measured parameter of the samples is used. Additionally or alternatively, at least one or each supplementary value m is used. k e.g. measured and provided by a measuring device, e.g. an in-line measuring device, which measures the respective measured quantity at the measuring point.

[0074] At certain measuring points, the values ​​of at least one or every measured quantity may be known during specific time periods, e.g., during certain operating phases and / or during certain stages of a process carried out at the measuring point. In this case, the supplementary spectra I k at least one measured spectrum I m one that was determined during one of the specified time periods, and the corresponding supplementary value(s) m k are given by the corresponding known value of the respective measured quantity.

[0075] Regardless of how the supplementary values ​​m k The measurement parameter(s) determined and / or provided reflect the supplementary training data thus determined, including at least one supplementary spectrum I. k and the corresponding supplementary value(s) keach measured quantity, the relationship between the measured spectra determined by the spectrometer(s) 100 in the respective measurement situation I m and the measured variable(s) under the prevailing conditions of the respective measurement situation. Accordingly, the supplementary training data for each measurement situation reflect the overall influence of the specified influencing factor(s) on the measured spectra I determined in the respective measurement situation. mDepending on how restrictively the respective measurement situation was predefined, the circumstances considered in the supplementary training data include, for example, features associated with the measurement setup, the measurement conditions, and / or the medium of the application-specific type within the respective measurement situation. For each measurement situation predefined by specifying the specific spectrometer 100 and the specific measurement location, the supplementary training data reflect the influence of the technical properties of the specific spectrometer 100, which define the supplementary spectra I. k determine the measurement setup used at the specified measuring point, the measurement conditions prevailing at the specified measuring point, and the properties of the medium prevailing at the respective measuring point.

[0076] Following the determination of the supplementary training data, the sequence further includes a procedure step B3, in which, based on the application-specific universal model GM and the supplementary training data including at least one supplementary spectrum I k and the corresponding supplementary value(s) m k Each measured quantity has a dedicated model LM for determining the measurement results MR of each measured quantity based on the measured spectra I m of the medium, which are determined and provided by the spectrometer(s) used in the respective measurement situation.

[0077] For each measurement situation, the dedicated model LM is determined such that the measurement results MR of each measurand, which are based on the supplementary spectra I, are determined using the dedicated model LM. k were determined, the supplementary values ​​m kcorrespond to the respective measured quantity, which the medium exhibits when determining the respective supplementary spectrum I k exhibits this. This can be achieved in various ways.

[0078] As an example of certain embodiments that determine at least one or every measurement situation, the corresponding dedicated model LM, based on the universal model GM and the supplementary training data, includes adapting the universal model GM to the prevailing circumstances in the respective measurement situation. Exemplary embodiments of determining the dedicated model LM by adapting the universal model GM are described in Fig. 3 and Fig. 4 shown.

[0079] In the Fig. In the embodiment shown in Figure 3, the universal model GM is adapted to the prevailing circumstances in the respective measurement situation by performing a procedure step B3.1a of retraining the universal model GM, taking into account the supplementary training data, based on the universal model GM and the supplementary training data, and then a procedure step B3.2a of determining the dedicated model LM, which is to be given by the newly trained universal model GMRL.

[0080] In certain embodiments, retraining the universal model GM to account for the prevailing circumstances of the respective measurement situation is performed, for example, in the same way as the initial training of the universal model GM and / or is based on the same underlying algorithm that was previously used to determine the universal model GM. In this case, the initial training of the universal model GM and the retraining of the universal model GM differ in that the initial training is based solely on the universal training data, while the retraining is based on the universal training data and the supplementary training data.

[0081] Retraining the universal model GM, taking into account the supplementary training data, offers the advantage that different weights can be assigned to the universal training data and the supplementary training data. For example, in certain embodiments, the retraining of the universal model GM is performed in a way that minimizes the influence of the prevailing circumstances in the respective measurement situation on the newly trained universal model GM. RL this is enhanced by the fact that the supplementary training data, including the supplementary spectra I k and the corresponding supplementary values ​​m k , is assigned a greater weight than the universal training data including the reference spectra I ref,n and the corresponding reference values ​​M ref,n .

[0082] In the Fig. In the embodiment shown in Figure 4, the universal model GM is adapted to the prevailing circumstances in the respective measurement situation by adjusting the universal model GM based on the universal model GM and the supplementary training data so that the universal model GM is used to achieve the desired results. A based on the supplementary spectra I k specific measurement results MR of each measured quantity the supplementary values ​​m k corresponding to the respective measured quantity that the medium uses to determine the respective complementary spectrum I k exhibits.

[0083] In the Fig. In the exemplary embodiment shown in section 4, determining the dedicated model LM includes, for example, a process step B3.1b in which, based on each supplementary spectrum I kThe universal model GM determines the prediction MP of each measured quantity; a procedure step B3.2b, based on the predictions MP, which are based on the complementary spectra I k and the corresponding supplementary values ​​m k were determined, of determining prediction errors E which the predictions MP exhibit; and a procedure step B3.3b, based on the prediction errors E, of fitting the universal model GM such that prediction errors E of predictions made with the adapted universal model GM A based on the supplementary spectra I k have been determined, will be minimized.

[0084] As through the in Fig. As indicated by the dashed arrow P shown in Figure 4, in certain embodiments the setting of the universal model GM includes, for example, an iterative process in which, based on the prediction errors E determined in process step B3.2b, the universal model GM is set in process step B3.3b and process steps B3.1b, B3.2b and B3.3b are carried out based on the previously set universal model by determining predictions for each measured quantity using the previously set universal model, based on these predictions and the corresponding supplementary values ​​m k The prediction errors of these predictions are determined, and then the previously set universal model is adjusted until the prediction errors of the predictions determined with the resulting set universal model GMA fall below a predetermined threshold.

[0085] Additionally or alternatively, in certain embodiments, the setting of the universal model GM includes, for example, the setting of at least one weighting factor, at least one parameter, one filter, one smoothing algorithm and / or at least one other from / of model component of the universal model GM, such that deviations between the predictions MP made with the set universal model GM A based on the supplementary spectra I k were determined, and the corresponding supplementary values ​​m k be minimized.

[0086] Following the adaptation of the universal GM model, the in Fig. 4. The embodiment shown further includes a process step B3.4b of determining and providing the dedicated model LM, which is derived from the adapted universal model GM. A is given, one.

[0087] Additionally or alternatively, in certain embodiments, the dedicated model LM is determined for at least one or every measurement situation, e.g., as a model given by the universal model GM, which makes predictions MP for each measurand based on measured spectra I. m The process step B3 determines the spectrometer(s) 100 used in the respective measurement situation and a transformer T that determines the measurement results MR of the measurand(s) based on the predictions MP determined with the universal model GM in the respective measurement situation. In these embodiments, process step B3 includes determining the dedicated model LM based on the universal model GM and the supplementary training data, including at least one supplementary spectrum I. k and the corresponding supplementary value m k Each measured quantity, the determination and provision of the transformer T.

[0088] An exemplary embodiment of process step B3, including the determination of the transformer T, is shown in Fig. 5 shown. In this embodiment, determining the dedicated model LM includes a process step B3.1c in which, based on at least one or each of the complementary spectrum I k The universal model GM determines the corresponding prediction MP of each measured quantity, and a procedure step B3.2c is entered, based on the prediction(s) MP, which is based on the or any complementary spectrum I. k and the corresponding supplementary value(s) m k the measured quantity(s) is / are determined, the transformer T is determined such that the measurement result MR of each measured quantity, which is determined with the transformer T based on the prediction(s) MP, which is determined with the universal model GM based on the or any supplementary spectrum I kis / are determined, the corresponding supplementary value m k the respective measured quantity that the medium is subject to during the determination of the respective supplementary spectrum I k exhibits.

[0089] In certain embodiments, determining the transformer T for at least one or every measurement situation, e.g., for each measured quantity, includes determining a transfer function f. T to calculate measurement results MR of the respective measured quantity as a function of the prediction MP of the respective measured quantity and a set of at least one parameter. In this embodiment, the transfer function f TFor each measurand, e.g., as a linear function, as a polynomial function of a specific order, or as another mathematical function for calculating the measurement results MR of the respective measurand as a function of the prediction MP of the respective measurand and a set of at least one parameter, the transfer function f is determined. T e.g. by fitting the respective transfer function f T determined by a set of at least one or more data points, where each data point is given by one of the predictions MP of the respective measured quantity, which is based on the universal model GM on one of the complementary spectra I k and the corresponding supplementary value m k was determined. In this respect, for example, regression methods and / or polynomial fitting methods known from the prior art are used.

[0090] Determining the transformer T by determining the transfer function f T For each measured quantity, the advantage is that only minimal computing power is required and very little supplementary training data is needed. As an example, the measurement results MR determined with the dedicated model LM, including the transformer T, are already noticeably more accurate than the predictions MP determined with the universal model GM, if each transfer function f T of the transformer T is determined based on only a single data point, and a significant increase in measurement accuracy is already achieved if each transfer function f T The transformer T is determined based on two data points or a limited number of data points. Accordingly, the transfer function f is... Tfor each measured quantity, e.g., based on a set of data points consisting of a single data point, two data points, a limited number of data points, e.g., a limited number less than or equal to 20, less than or equal to 10, or even less than or equal to 5 data points, or multiple data points.

[0091] Additionally or alternatively, in certain embodiments, the transformer T, e.g., in the form of a transfer model TM, is used for at least one or every measurement situation to determine the measurement results MR based on the predictions MP determined with the universal model GM based on measured spectra I. m, which were determined by the spectrometer(s) 100 used in the respective measurement situation. This approach is particularly advantageous in embodiments of the method according to the present disclosure in which two or more measurands of the medium are to be measured and / or the supplementary training data comprise several supplementary spectra I k and the corresponding supplementary values ​​m k Include every measured quantity.

[0092] In these embodiments, the transfer model TM is determined based on transfer model training data, including the predictions MP of the measured quantities determined with the universal model GM based on the complementary spectra I. k and the corresponding supplementary values ​​m k every measured quantity. This can be achieved in various ways.

[0093] As an example, in certain embodiments, for at least one or every measurement situation, the transfer model TM is, for example, based on a detailed analysis of the transfer model training data, determined and provided with an algorithm for calculating the measurement results MR of each measurand based on the predictions MP of the measurand(s) determined with the universal model GM, based on measured spectra I. m , which were determined by the spectrometer(s) used in the respective measurement situation.

[0094] In certain embodiments, the analysis of the transfer model training data and / or the determination of the transfer model TM includes, for example, based on the transfer model training data, performing a multivariate analysis, a partial least squares regression, a support vector regression and / or a principal component analysis of the predictions MP and / or a procedure step for quantitatively evaluating the dependencies between the predictions MP of the measured variable(s) and the corresponding supplementary values ​​m. k one for each measured quantity.

[0095] In other embodiments, determining the transfer model TM, e.g., based on the transfer model training data, includes training a neural network to determine the measurement results MR of the measured quantity(s) based on predictions MP made with the universal model GM based on measured spectra I mwere determined in the respective measurement situation, and the determination of the transfer model TM that is to be given by the trained neural network.

[0096] Considering that the predictions MP determined with the universal model GM are considerably less complex than the spectra on which they are based, and that the application-specific relationships between the spectral values ​​of the measured spectra I m Since the corresponding values ​​of the measured quantity(s) are already taken into account by the universal model GM, the design and structure of the transfer model TM, which considers the circumstances of the respective measurement situation, are much simpler than the design and structure of the universal model GM. This offers the advantage that correspondingly little time, effort, and / or computing power is required to determine the transfer model TM for the respective measurement situation.

[0097] The simplicity of the transfer model TM also offers the advantage that little supplementary training data is required to determine the transfer model TM, thus achieving high measurement accuracy with the dedicated model LM, including the universal model GM and the transfer model TM. For example, in certain embodiments, the transfer model TM is determined based on a limited number of supplementary spectra I. k , e.g. a limited number of less than or equal to 100, less than or equal to 50 or even less than or equal to 20 supplementary spectra I k , and the corresponding supplementary values ​​m k each measured quantity is determined.

[0098] Additionally or alternatively, in certain embodiments, the dedicated model LM is used for at least one or every measurement situation, e.g., based on a combination of the above in connection with Fig. 3 and Fig. The method described in section 5 determines the dedicated model. In these embodiments, the dedicated model LM is determined, for example, as a model derived from an adapted universal model GM. RG Given that the adapted predictions MP RG each measurement based on measured spectra I m determined by the spectrometer(s) used in the respective measurement situation, and an adapted transformer T A , which calculates the measurement results MR of the measured quantity(s) based on the adapted predictions MP, which are derived from the adapted universal model GM RG were determined in the respective measurement situation.

[0099] A corresponding embodiment of process step B3 for determining the dedicated model LM is described in Fig. 6 shown. As in Fig. As shown in Figure 6, determining the dedicated model LM can be a procedure step B3.1d for determining the adapted universal model GM. RG This is achieved by retraining the universal model GM based on the universal model GM and the supplementary training data, taking the supplementary training data into account. Following retraining, the adapted universal model GM is then used. RG determined, which is given by the newly trained universal model.

[0100] In analogy to the newly trained procedure, which is related to Fig. As described above, in certain embodiments the retraining of the universal model GM to take into account the prevailing circumstances in the respective measurement situation is carried out, e.g., in a process step B3.1d, which is described in Fig. Figure 6 shows that the training of the universal model GM is carried out in the same way as the training of the universal model GM based on the same underlying algorithm that was previously used to determine the universal model GM.

[0101] As described above, retraining the universal model GM offers the advantage that different weights can be assigned to the universal training data and the supplementary training data. In the Fig. The 3 methods shown will be used to train the newly trained universal model GM. RL subsequently used as a dedicated model LM. Accordingly, within the framework of the in Fig. The method shown in section 3 utilizes the flexibility provided by assigning different weights, preferably as outlined above, to enhance the influence of the prevailing circumstances in the respective measurement situation on the newly trained universal model GMRL. In contrast to this method, the method described in Fig. The 6 methods shown show the newly trained universal model GM subsequently as an adapted universal model GM. RG used, which represents only a part of the dedicated model LM and additionally the adapted transformer T A includes. Accordingly, within the framework of the in Fig. In the method shown in Section 6, the flexibility afforded by assigning different weights is preferably used in such a way that the global validity of the resulting newly trained universal model is maintained. As an example, in certain embodiments of the method shown in Section 6, the following applies: Fig. 6. Procedure step B3.1d shown for retraining the universal model GM is carried out in such a way that the global validity of the resulting newly trained universal model is maintained by assigning a lower weight to the supplementary training data than to the universal training data.

[0102] Following the determination of the adapted universal model GM RG includes the determination of the in Fig. 6. Dedicated model LM shown further determines the adapted transformer T. A one. Determining the adapted transformer T A includes, for example, a process step B3.2d in which, based on at least one or each of the complementary spectrum I k with the adapted universal model GM RG the corresponding adapted prediction MP RGeach measured quantity is determined, and a process step B3.3d in which, based on the adapted universal model GM, RG certain adapted predictions MP RG based on the or any complementary spectrum I k and the corresponding supplementary value(s) m k the measured quantity(s) of the adapted transformer T A is determined.

[0103] In analogy to the determination of the transformer T shown in Fig. 5 in process step B3.2c, the adapted transformer T A e.g. determined in such a way that the adapted transformer T A based on the adapted universal model GM RG based on the or any complementary spectrum I k specific adapted prediction MP RG specific measurement result MR of each measured quantity, which is used in the determination of the respective supplementary spectrum I ksupplementary value shown by the medium m k corresponds to the respective measured quantity.

[0104] In further analogy to the one in Fig. The determination of transformer T shown in step B3.2c of Figure 5 includes, in certain embodiments, the determination of the adapted transformer T. A for at least one or every measurement situation, e.g., for each measured quantity, the determination of an adapted transfer function f. TA to calculate measurement results MR of the respective measured quantity depending on the adapted prediction MP RG the respective measured quantity and a set of at least one parameter. In this embodiment, the adapted transfer function f TA for each measured quantity, e.g. as a linear function, as a polynomial function of a certain order, or as another mathematical function to calculate the measurement results MR of the respective measured quantity as a function of the adapted prediction MPRG the respective measured quantity and a set of at least one parameter are determined. For each measured quantity, the at least one parameter of the adapted transfer function f is determined. TA e.g. by fitting the respective adapted transfer function f TA determined by a set of at least one or more data points, where each data point is determined by one of the adapted predictions MP. RG the respective measured quantity is given, which is used with the adapted universal model GM RG based on one of the supplementary spectra I k and the corresponding supplementary value m k was determined. In this respect, for example, regression methods and / or polynomial fitting methods known from the prior art are used.

[0105] Additionally or alternatively, in certain embodiments the adapted transformer T is used for at least one or each measuring point. A, e.g. in the form of an adapted transfer model TM A to determine the measurement results MR based on the adapted universal model GM RG certain adapted predictions MP RG based on measured spectra I m , which were determined by the spectrometer(s) 100 used in the respective measurement situation. In these embodiments, the adapted transfer model TM is used. A preferably determined based on transfer model training data, including that with the adapted universal model GM RG certain adapted predictions MP RG the measured quantity(s) based on the supplementary spectra I k and the corresponding supplementary values ​​mk.

[0106] As an example, the adapted transfer model TM is used in certain embodiments. Ae.g., based on a detailed analysis of the transfer model training data, by determining and providing an algorithm to calculate the measurement results MR of each measured quantity based on the adapted predictions MP. RG the measured quantity(s) that are used with the adapted universal model GM RG were determined based on measured spectra I m , which were determined by the spectrometer(s) used in the respective measurement situation.

[0107] In certain embodiments, the analysis of the transfer model training data and / or the determination of the adapted transfer model TM includes A e.g. performing a multivariate analysis, a partial least squares regression, a support vector regression and / or a principal component analysis of the adapted predictions MP RGand / or a procedural step for quantitatively evaluating the dependencies between the adapted predictions MP RG the measured quantities and the corresponding supplementary values ​​m k one for each measured quantity.

[0108] In other embodiments, determining the adapted transfer model TM includes A , e.g. based on the transfer model training data, training a neural network to adapt the measurement results MR of the measured quantity(s) based on adapted predictions MP RG to determine which are compatible with the adapted universal model GM RG based on measured spectra I m were determined in the respective measurement situation, and the determination of the adapted transfer model TM A , which is to be provided by the trained neural network.

[0109] Regardless of the method used to determine the dedicated model LM, in certain embodiments the method may further include, for at least one or each measurement situation, an optional method step C1 in which, based on the supplementary training data determined in the respective measurement situation and the dedicated model LM determined for the respective measurement situation, a measurement accuracy ΔMR of the measurement results MR of each measurand determined with the dedicated model LM in the respective measurement situation is determined.

[0110] As in Fig. As shown in Figure 1, the optional process step C1 in certain embodiments includes, for example, based on at least one or each supplementary spectrum I. k with the dedicated model LM, the determination of a measurement result MR of each measured quantity and, based on the supplementary spectra I kspecific measurement result(s) MR and the corresponding supplementary values ​​m k , the determination of the measurement accuracy ΔMR of the measurement results MR obtained with the dedicated model LM.

[0111] The measurement accuracy ΔMR determined in this way can be used in various ways. For example, in certain embodiments of the method, an optional process step C2 is included in which the measurement accuracy ΔMR is provided.

[0112] Additionally or alternatively, the measurement accuracy ΔMR can be used to ensure a predetermined minimum accuracy K, e.g., a minimum accuracy predetermined for at least one measurement point within the respective measurement situation where spectrometric measurements are to be performed. In these embodiments, the previously determined model LM is preferably used only if the measurement accuracy ΔMR is greater than or equal to the predetermined minimum accuracy K, and an alternative solution is preferably used if the measurement accuracy ΔMR is less than the predetermined minimum accuracy K. This alternative solution can be achieved in various ways.

[0113] As an example, in certain embodiments, the measurement accuracy ΔMR achieved with the dedicated model LM can be improved for at least one measurement situation by performing the procedure steps of determining additional supplementary training data and recalculating the dedicated model LM based on the supplementary training data. In this context, as indicated by the dashed arrows P1 and P2 in Fig. As specified in section 1, the determination of the additional supplementary training data is carried out, for example, as described above in connection with procedure steps B1 and / or B2. The recalculation of the dedicated model LM is performed, for example, by repeating procedure step B3, which involves determining the dedicated model LM based on the supplementary training data and the additional supplementary training data. Subsequently, the dedicated model LM for the respective measurement situation is determined by the newly determined dedicated model.

[0114] Additionally or alternatively, in certain embodiments, the measurement accuracy ΔMR achieved with the dedicated model LM can be improved for at least one measurement situation by performing the procedure steps of limiting the respective measurement situation by a limited measurement situation, which is determined by specifying at least one additional influencing factor that affects the measured spectra Im which is determined by the spectrometer(s) in the respective measurement situation. In this embodiment, the method further includes, for example, determining the dedicated model for determining the measurement results of each measurand based on the universal model GM and the supplementary training data determined in the limited measurement situation, which is based on the measured spectra I m of the medium, which are determined and provided by the spectrometer(s) 100 used in the limited measurement situation. In this respect, the determination of the supplementary training data and the determination of the dedicated model for the limited measurement situation are carried out, e.g., as described above in connection with procedure steps B1, B2 and B3.

[0115] Regardless of the method used to determine the dedicated model LM, the sequence of procedural steps carried out after the determination of the dedicated model LM for each measurement situation according to the present disclosure further includes a procedural step B4 in which, based on measured spectra I m of the medium, which are determined and provided by the spectrometer(s) used in the respective measurement situation, and the dedicated model LM measurement results MR of the or each measured quantity of the medium are determined and provided.

[0116] In every measurement situation, the measurement results MR determined with the corresponding dedicated model LM offer the advantage of being more accurate than the predictions MP determined with the universal model GM. In this respect, the highest measurement accuracy is achieved with dedicated models LM that have been determined restrictively for predetermined measurement situations.

[0117] As in Fig. As shown in Figure 2, in certain embodiments the measurement results MR are determined, for example, by an evaluation unit 17, which is connected to the signal processor 15, which processes the measured spectra I m provides, is connected to, or communicates with. The evaluation unit 17 is, for example, a computer, a microprocessor, or another type of computing unit that, based on the measured spectra I m and the dedicated model LM for determining the measurement result MR, which was previously determined, determines and provides the measurement results MR. The in Fig. The exemplary evaluation unit 17 shown in section 2 is, for example, a component of a spectrometer 100 that performs the spectroscopic measurements, or a component of a measurement system including the at least one or each of the spectrometers 100 used in the measurement situation(s) and the evaluation unit 17 connected or communicating with the spectrometer(s) 100.

[0118] In another embodiment, if the in Fig. In the embodiment shown in Figure 1, the method is carried out for at least two or more different measurement situations. Supplementary training data, determined above in procedure step B2 for at least one measurement situation, are used to increase the global validity of the universal model GM. In such an embodiment, the method further includes at least once the execution of a procedure step indicated by arrow P3 in Figure 1. Fig.As specified in section 1, the update of the universal model GM is performed based on the universal training data and the supplementary training data previously determined in procedure step B2 for at least one measurement situation. The update of the universal model GM is performed, for example, in the same way as the initial training of the universal model GM and / or is based on the same underlying algorithm previously used to determine the universal model GM. In this case, the initial training of the universal model GM and the update of the universal model GM differ in that the initial training is based solely on the universal training data, while the update is based on the universal training data and previously determined supplementary training data.Updating the universal model GM is preferably performed in such a way as to increase the global validity of the updated universal model GM. This is achieved, for example, by treating the supplementary training data as additional universal training data, which are used in addition to and in the same way as the original universal training data.

[0119] After updating, the updated universal model is then used as the universal model GM. In this respect, the method according to the present disclosure further includes determining the corresponding dedicated model LM for at least one further measurement situation, as described above, based on the updated universal model GM.

[0120] The signal processor 13 and the evaluation unit 17, as well as other units according to the present disclosure, can each be a section of a processing subsystem that includes one or more computing devices with storage, processing, and / or communication hardware. Each can be a single or distributed device, and the functions of each device can be performed by hardware and / or software. Each can include one or more arithmetic logic units (ALUs), central processing units (CPUs), memory, limiters, conditioners, filters, format converters, or the like, which are not shown for clarity. In at least one embodiment, the signal processor 13 and the evaluation unit 17 are programmable to execute algorithms and process data according to an operating logic defined by programming instructions, such as software or firmware.Alternatively or additionally, the operating logic for the units can be defined, at least partially, by hard-wired logic or other hardware, for example, by an application-specific integrated circuit (ASIC) of any suitable type. Each can be used exclusively for the functions described herein or furthermore for the regulation, control, and activation of one or more other subsystems or aspects of the analyzers and spectrometers of this disclosure.

[0121] While various embodiments of an analyzer and methods for their use and construction have been described in great detail herein, these embodiments are offered merely as non-limiting examples of the disclosure described herein. It is therefore understood that various changes and modifications may be made and elements thereof may be replaced by equivalents without deviating from the scope of protection of the disclosure. The present disclosure is not intended to be exhaustive and does not limit the scope of protection of the subject matter of the disclosure.

[0122] Furthermore, the disclosure may have presented a method and / or process as a specific sequence of steps when describing representative embodiments. However, to the extent that the method or process is not based on the specific sequence of steps set forth herein, it should not be limited to the specific sequence of steps described. Other sequences of steps may therefore be possible and remain within the scope of protection of this disclosure.

Claims

[1] Spectrometric method for determining, in a measurement situation or in several different measurement situations within a predetermined application with at least one spectrometer (100) of a predetermined type, measurement results (MR) of at least one measurand of a medium of an application-specific type, wherein each measurement situation is predetermined by specifying at least one influencing factor relating to measured spectra (I m ) effects which are determined by the spectrometer(s) (100) in the respective measurement situation, the procedure encompassing: Performing reference measurements by using at least one spectrometer (100) reference spectra (I ref,n ) are determined from reference samples of the medium, and for each reference sample a reference value (m ref,n ) each measurement parameter of the reference sample is determined and / or provided; based on universal training data including the reference spectra (Iref,n ) and the corresponding reference values ​​(m ref,n ), Determining a universal model (GM) for determining predictions of any measured quantity based on measured spectra (I m ) of the medium, which were determined by spectrometers (100) in the predetermined application; and for every measurement situation: in the respective measurement situation with at least one spectrometer (100), determining and providing measured spectra (I m ) of the medium; for at least one complementary spectrum (I k ) or a limited number of complementary spectra (I k ), where each complementary spectrum (I k ) determined by one of the spectrometer(s) (100) used in the respective measurement situation (I k ) is given, Determine and / or provide a supplementary value (m k ) each measured quantity of the medium; based on the universal model (GM) and supplementary training data including at least one or each supplementary spectrum (I k ) and the corresponding supplementary value(s) (m k ), Determining a dedicated model (LM) for determining measurement results (MR) of each measurand based on measured spectra (I m ) of the medium, which are determined and provided by the spectrometer(s) (100) used in the respective measurement situation; and based on measured spectra (I m ) of the medium, which are determined and provided by the spectrometer(s) (100) used in the respective measurement situation, and the dedicated model (LM), determining and providing measurement results (MR) of each measured quantity of the medium. [2] Method according to claim 1, wherein each dedicated model (LM) is determined such that the measurement results (MR) of each measurand that are based on the dedicated model (LM) or any complementary spectrum ( k The values ​​supplementing the values ​​(m) were determined. k ) correspond to the respective measured quantity, which the medium exhibits when determining the respective complementary spectrum (I k ) exhibits. [3] Method according to claims 1 to 2, wherein determining the dedicated model (LM) for at least one or each measurement situation comprises: based on the universal model (GM) and the supplementary training data, retraining the universal model (GM) taking the supplementary training data into account; and Determining the dedicated model (LM) that will be replaced by the newly trained universal model (GM) RL ) should be given. [4] Method according to claim 3, wherein the retraining of the universal model (GM): is performed in the same way as the training of the universal model (GM) and / or is based on the same underlying algorithm that was previously used to determine the universal model (GM); and / or an increase in the influence of the circumstances prevailing in the respective measurement situation on the newly trained universal model (GM) RL ) by assigning greater weight to the supplementary training data than to the universal training data. [5] Method according to claims 1 to 4, wherein determining the dedicated model (LM) for at least one or each measuring point comprises: Based on the supplementary training data, the universal model (GM) is adapted so that the data is matched to the adapted universal model (GM). A ) based on the complementary spectra (I k) certain measurement results (MR) of each measured quantity the supplementary values ​​(m k ) correspond to the respective measured quantity that the medium uses to determine the respective complementary spectrum (I k ) exhibits; and Determining the dedicated model (LM) that is replaced by the adapted universal model (GM) A ) should be given. [6] Method according to claim 5, wherein the adaptation of the universal model (GM) comprises: based on each complementary spectrum (I k ), Determining the prediction (MP) of each measured quantity using the universal model (GM); based on the predictions (MP) that are based on the complementary spectra (I k ) and the corresponding supplementary values ​​(m k ) were determined, determining prediction errors (E) exhibited by the predictions (MP); and at least one of: based on the prediction errors (E), adapting the universal model (GM) so that prediction errors (E) of predictions, with the adapted universal model (GM) A ) based on the complementary spectra (I k ) determined, be minimized; based on the prediction errors (E), iterative adjustment of the universal model (GM) and repetition of the procedural steps of determining the prediction of each measured quantity based on each complementary spectrum (I) k ) with the adapted universal model, determining prediction errors shown by these predictions, and subsequently adjusting the previously adapted universal model until the prediction errors of predictions made with the resulting adapted universal model (GM) A ) are determined to fall below a predetermined threshold; and Adapting the universal model (GM) by adjusting at least one weighting factor, at least one parameter, one filter, one smoothing algorithm and / or at least one other model component of the universal model (GM) such that deviations between predictions (MP) made with the adapted universal model (GM) A ) based on the complementary spectra (I k ) were determined, and the corresponding supplementary values ​​(m K ) be minimized. [7] Method according to claims 1 to 6, wherein for at least one or each measuring point: the dedicated model (LM) is determined as a model given by the universal model (GM), which makes predictions (MP) of each measurand based on measured spectra (I) m) determined by the spectrometer(s) (100) used in the respective measurement situation, and a transformer (T) that determines measurement results (MR) of the measured quantity(s) based on the predictions (MP) determined with the universal model (GM); and includes determining the dedicated model (LM): based on at least one or each complementary spectrum (I k ), using the universal model (GM), determining the corresponding prediction (MP) of each measured quantity; and based on the prediction(s) (MP) that is based on the or any complementary spectrum (I k ) and the corresponding supplementary value(s) (m k) was determined(s), determining the transformer (T) such that the measurement result (MR) of each measured quantity determined with the transformer (T) based on the prediction(s) (MP) made with the universal model (GM) based on the or any complementary spectrum (I) k ) was determined, the corresponding supplementary value (m k ) corresponds to the respective measured quantity, which the medium represents during the determination of the respective complementary spectrum (I k ) showed. [8] Method according to claim 7, wherein: for at least one or each measuring point the transformer (T) is determined by defining a transfer function (f) for each measured quantity. T) for calculating measurement results (MR) of the respective measured quantity as a function, as a linear function, as a polynomial function of a certain order or as another mathematical function of prediction (MP) of the respective measured quantity and a set of at least one parameter, wherein: for each measured quantity, at least one parameter of the transfer function (f r ) by fitting the respective transfer function (f r ) is determined by a set of data points; each data point is given by one of the predictions (MP) of the respective measured quantity, which is based on the universal model (GM) using one of the complementary spectra (I). k ) and the corresponding supplementary value (m k ) is determined; and the set of data points consists of a single data point, two data points, a limited number of data points less than or equal to 20, less than or equal to 10, or even less than or equal to 5 data points, or consists of several data points; and / or for at least one or each measuring point: the transformer (T) included in the dedicated model (LM) is determined in the form of a transfer model (TM) for determining measurement results (MR) of the measured quantity(s) based on predictions (MP) made with the universal model (GM) based on measured spectra (I) m ) are determined by the spectrometer(s) (100) used in the respective measurement situation; and Determining the respective dedicated model (LM) based on transfer model training data including predictions (MP) of the measured variables, which are based on the universal model (GM) based on the complementary spectra (Ik ) and the corresponding supplementary value(s) (m k ) of each measured quantity, including the determination and provision of the transfer model (TM). [9] Method according to claim 8, wherein: For at least one or each measuring point, determining the transfer model (TM) includes at least one of the following: a) based on a detailed analysis of the transfer model training data, determining and providing an algorithm to calculate the measurement results (MR) of each measurand based on predictions (MP) of the measurand(s) determined with the universal model (GM), based on measured spectra (I m ), which were determined by the spectrometer(s) (100) used in the respective measurement situation; and b) based on the transfer model training data, perform a multivariate analysis, partial least squares regression, support vector regression and / or principal component analysis of the predictions (MP) and / or quantitatively assess dependencies between the predictions (MP) of the measured variables and the corresponding complementary values ​​(m) k ) of any measured quantity; or based on the transfer model training data, training a neural network to determine the measurement results (MR) of the measured quantity(s) based on predictions (MP) made with the universal model (GM) based on measured spectra (I) m ) were determined, which were determined in the respective measurement situation, and determining the transfer model (TM) to be provided by the trained neural network, and / or For at least one or every measurement situation, the transfer model (TM) is based on supplementary training data including a limited number of less than or equal to 100, less than or equal to 50, or even less than or equal to 20 supplementary spectra (I). k ) and the corresponding supplementary values ​​(m k ) certainly. [10] Method according to claims 1 to 9, wherein for one or each measuring point: the dedicated model (LM) is defined as a model that is adapted by a universal model (GM). RG ) is given that adapted predictions (MP RG ) of each measured quantity based on measured spectra determined in the respective measurement situation (I m ) determined, and an adapted transformer (T A ) the measurement results (MR) of the measured quantity(s) based on adapted predictions (MP) RG ) determined, which is determined with the adapted universal model (GM) RG) based on measured spectra determined in the respective measurement situation (I m ) be determined; and includes determining the dedicated model (LM): based on the universal model (GM) and the supplementary training data, retraining the universal model (GM) taking into account the supplementary training data and determining the adapted universal model (GM). RG ), which is given by the newly trained universal model; based on at least one or each complementary spectrum (I k ), with the adapted universal model (GM RG ) determining the adapted prediction (MP) RG ) of each measured quantity; and based on the prediction(s) (MP RG ), which are based on the adapted universal model (GM) RG ) based on the or any complementary spectrum (I k ) and the corresponding supplementary value(s) (m k) was determined(s), Determining the adapted transformer (T A ), so that the measurement result (MR) of each measured quantity, which is obtained with the adapted transformer (T A ) based on the adapted prediction(s) (MP RG ) was determined, which is based on the adapted universal model (GM) RG ) based on the or any complementary spectrum (I k ) was determined, the supplementary value (m k ) corresponds to the respective measured quantity, which the medium represents during the determination of the respective complementary spectrum (I k ) showed. [11] Method according to claim 10, wherein: Determining the adapted universal model (GM) RG) retraining the universal model (GM) in the same way as training the universal model (GM) and / or based on the same underlying algorithm previously used to determine the universal model (GM), and / or in a way that preserves global validity of the retrained universal model by assigning less weight to the supplementary training data than to the universal training data; Determining the adapted transformer (T) for at least one or each measuring point A ) for each measured quantity, determining an adapted transfer function (f TA ) to calculate measurement results (MR) of the respective measured quantity as a function, as a linear function, as a polynomial function of a certain order or as another mathematical function of the adapted prediction (MP) RG) the respective measured quantity and a set of at least one parameter, wherein for each measured quantity the at least one parameter of the adapted transfer function (f TA ) by fitting the respective adapted transfer function (f TA ) is determined on a set of at least one or more data points, where each data point is determined by one of the adapted predictions (MP) RG ) of the respective measured quantity, which is given with the adapted universal model (GM) RG ) based on one of the complementary spectra (I k ) and the corresponding supplementary value was determined (m k ); and / or for at least one or each measuring point the adapted transformer (T A ) in the form of an adapted transfer model (TM) A ) to determine measurement results (MR) of the measured quantity(s) based on adapted predictions (MP) RG ) of the measured quantity(s) that are associated with the adapted universal model (GM)RG ) based on spectra determined in the respective measurement situation (I m were determined, determined, and the adapted transfer model (TM) A ) based on transfer model training data including adapted predictions (MP) RG ) of the measured quantity(s) that are associated with the adapted universal model (GM) RG ) based on the complementary spectra (I k ) and the corresponding supplementary values ​​(m k ) were determined, will be determined. [12] Method according to claim 11, wherein determining the adapted transfer model (TM) A ) includes at least one of the following: a) based on a detailed analysis of the transfer model training data, determining and providing an algorithm for calculating measurement results (MR) of each measurement quantity based on adapted predictions (MP) RG) of the measured quantity(s) that are associated with the adapted universal model (GM) RG ) were determined based on measured spectra (I m ), which were determined in the respective measurement situation; and b) Performing a multivariate analysis, partial least squares regression, support vector regression and / or principal component analysis of the adapted predictions (MP) RG ) and / or a procedural step of quantitatively assessing dependencies between the adapted predictions (MP) RG ) the measured quantity(s) and the corresponding supplementary values ​​(m k ) of any measured quantity; or based on the transfer model training data, training a neural network to predict the measurement results (MR) of the measured quantity(s) based on the adapted predictions (MP). RG ) to determine which are compatible with the adapted universal model (GM) RG ) based on the measured spectra (I m) were determined, which were determined in the respective measurement situation, and the determination of the adapted transfer model (TM) A ), which is to be provided by the trained neural network. [13] Method according to any one of claims 1 to 12, wherein specifying the at least one influencing factor for at least one or each measurement situation includes at least one of the following: Specify at least one influencing factor associated with a measurement setup; Specifying the measurements to be performed on samples of the medium, or which are to be in-situ measurements; Specifying the measurements to be performed in a flow cell carrying the medium, or in a container, bioreactor, or other type of vessel containing the medium; Specify at least one influencing factor linked to the measurement conditions; Specifying a parameter range for at least one parameter and / or temperature; Specify at least one influencing factor associated with the medium of the application-specific type within the respective measurement situation; Specify at least one property of the medium; Specify that the medium is subjected to a specific process, a specific production process and / or a specific processing procedure; Specifying a facility that contains and / or processes the medium; Specify that the medium includes at least one specified component, that it includes several specified components, or that it consists of specified components; Specifying the specific spectrometer that performs the measurements; and Specify the specific measuring point where the measurements will be carried out. [14] Method according to claims 1 to 13, wherein: the predetermined application is a biotechnological application in which mammalian cells producing an active ingredient of a drug are cultivated in a cell culture medium; and at least one measurement situation is predetermined by specifying or by specifying alone the type of cell culture medium and / or the type of mammalian cells. [15] Method according to claims 1 to 14, further comprising for at least one or each measurement situation: based on the supplementary training data and the dedicated model (LM), determine a measurement accuracy (ΔMR) of the measurement results (MR) determined with the dedicated model (LM); and Performing at least one of: a) Providing measurement accuracy (ΔMR); and b) if the measurement accuracy (ΔMR) is lower than a predetermined minimum accuracy (K): Determine additional supplementary training data, redetermine the dedicated model (LM) based on the supplementary training data and the additional supplementary training data, and determine the dedicated model (LM) given by the newly determined dedicated model; and / or Limiting the respective measurement situation, which is given by a limited measurement situation, which is determined by specifying at least one additional influencing factor, the measured spectra (I m ) influenced by the spectrometer(s) (100) determined in the respective measurement situation, is specified, and based on the universal model (GM) and the supplementary training data determined in the limited measurement situation, determine the dedicated model (LM) to determine measurement results (MR) of each measurand based on measured spectra (I m) of the medium, which are determined and provided by the spectrometer(s) used in the limited measurement situation (100). [16] Method according to claims 1 to 15, further comprising at least one: Updating the universal model (GM) based on the universal training data and supplementary training data previously determined for at least one measurement situation; and Subsequently, for at least one further measurement situation, the determination of the corresponding dedicated model (LM) based on the updated universal model (GM).

Images