Data extraction system, data extraction method, and data extraction program

Abstract

A data extraction system is provided with: a setting unit that sets an input reference indicating performance required for data analysis; an input unit that acquires an input signal indicating a measurement result of an object to be measured during biological production; a pre-processing unit that generates a pre-processed signal by performing a pre-processing operation for comparing the input signal with an input reference; a post-processing unit that generates a post-processed signal by performing a post-processing operation using another signal obtained by measuring the object to be measured on the pre-processed signal when the pre-processed signal does not satisfy the input reference; and an extraction unit that extracts the pre-processed signal as extraction data when the pre-processed signal satisfies the input reference, and extracts the post-processed signal as extraction data when the post-processed signal satisfies the input reference.

Description

Technical Field

[0001] This disclosure relates to a data extraction system, a data extraction method, and a data extraction program. Background Technology

[0002] A technique for measuring the time-dependent changes of substances using sensors is known. Patent Document 1 discloses a technique for sensing the presence or absence of anomalies in a culture medium. In the technique described in Patent Document 1, the distribution of characteristic values ​​of the culture medium is determined, and an anomaly determination threshold is set based on the distribution. Patent Document 2 discloses a technique for analyzing the deterioration of plastic materials, etc. In the technique described in Patent Document 2, second-order differential processing is performed on near-infrared spectral data of plastic materials, etc., obtained by near-infrared spectroscopy, to obtain corrected spectral data.

[0003] Existing technical documents Patent documents Patent Document 1: Japanese Patent No. 6977977 Patent Document 2: Japanese Patent Application Publication No. 2022-7236 Summary of the Invention

[0004] The problem that the invention aims to solve The goal is to improve the quality of the data used for analysis.

[0005] Solution for solving the problem One aspect of the data extraction system disclosed herein includes: a setting unit for setting an input benchmark representing the performance required for data analysis; an input unit for acquiring an input signal representing the measurement result of a analyte in biological production; a preprocessing unit for performing preprocessing operations on the input signal for comparison with the input benchmark to generate a preprocessed signal; a postprocessing unit for performing postprocessing operations on the preprocessed signal using other signals representing the measurement result of the analyte when the preprocessed signal does not meet the input benchmark to generate a postprocessed signal; and an extraction unit for extracting the preprocessed signal as extracted data when the preprocessed signal meets the input benchmark, and extracting the postprocessed signal as extracted data when the postprocessed signal meets the input benchmark.

[0006] In one aspect of the data extraction system disclosed herein, a preprocessing operation is performed on the input signal to generate a preprocessed signal. Then, the preprocessed signal is compared with an input benchmark representing the performance required for data analysis. If the preprocessed signal meets the input benchmark, the preprocessed signal is extracted as extracted data. Conversely, if the preprocessed signal does not meet the input benchmark, a postprocessing operation is performed on the preprocessed signal to generate a postprocessed signal. Then, the postprocessed signal is compared with the input benchmark. If the postprocessed signal meets the input benchmark, the postprocessed signal is extracted as extracted data. Thus, extracted data with guaranteed quality can be extracted. As a result, the quality of the data used for analysis can be improved.

[0007] Invention Effects According to this disclosure, a technique can be provided that can improve the quality of data used for analysis. Attached Figure Description

[0008] Figure 1 This is a diagram that schematically illustrates the configuration of the data extraction system according to the first embodiment.

[0009] Figure 2 This is a schematic diagram illustrating an example of the actions of a data extraction system.

[0010] Figure 3 This is a diagram illustrating an example of spectroscopic measurement in biological production.

[0011] Figure 4 (a) ~ Figure 4 (c) is an example of an extraction graph illustrating the changes in the concentration of organic molecules in near-infrared spectroscopy.

[0012] Figure 5 This is a schematic diagram illustrating another example of the actions of a data extraction system.

[0013] Figure 6 This is a summary diagram illustrating yet another example of the actions of a data extraction system.

[0014] Figure 7 (a) ~ Figure 7 (d) is a diagram illustrating an example of the effect of preprocessing operations and input baseline settings.

[0015] Figure 8 (a) ~ Figure 8 (d) is a diagram that illustrates another example of how the effects of preprocessing operations and input baseline settings can be used to explain the results.

[0016] Figure 9 This is an overall diagram illustrating an example of the application of the data extraction system in the embodiment.

[0017] Figure 10 This is a diagram illustrating an example of the overall operation of a biological production system, including a data extraction system.

[0018] Figure 11 This is a diagram illustrating an example of the actions of a data extraction system.

[0019] Figure 12 This is a diagram illustrating an example of the hardware configuration associated with a data extraction system. Detailed Implementation

[0020] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In the description of the drawings, the same elements are labeled with the same reference numerals, and repeated descriptions are omitted.

[0021] [summary] The data extraction system disclosed herein extracts data measured by sensors for data analysis. Hereinafter, the data used for data analysis will be referred to as "extracted data." In data analysis, a minimum performance requirement may be placed on the extracted data. Hereinafter, this minimum performance requirement will be referred to as "evaluation metrics." The data extraction system can be applied, for example, to data analysis in bioproduction or pharmaceutical manufacturing, but is not limited thereto. In this disclosure, examples of the data extraction system being applied to data analysis in bioproduction will be described.

[0022] The data extraction system acquires signals measured by sensors as input signals. It sets an index for the input signal corresponding to the evaluation index as the input benchmark. The system performs preprocessing operations on the input signal for comparison with the input benchmark, generating a preprocessed signal. Finally, the system compares the preprocessed signal with the input benchmark to determine whether it meets the benchmark's requirements.

[0023] The data extraction system performs post-processing operations on pre-processed signals that do not meet the input benchmark, using other signals measured by sensors to generate a post-processed signal. These other signals can be, for example, signals preceding or following the input signal in a time series. The data extraction system compares the post-processed signal with the input benchmark to determine whether it meets the benchmark. Then, the system extracts either the pre-processed or post-processed signal that meets the input benchmark as the extracted data. Alternatively, the system can remove post-processed signals that do not meet the input benchmark. The system can also extract data obtained by performing output operations on the pre-processed or post-processed signals as extracted data. In this way, the data extraction system extracts data that meets the evaluation criteria.

[0024] [Implementation Method] Figure 1This is a diagram that schematically illustrates the configuration of the data extraction system 1 in an embodiment. The data extraction system 1 is connected to the sensor 30 and the database 40 in a communicative manner. For example, the data extraction system 1 acquires an input signal from the sensor 30 and extracts multiple data points. The data extraction system 1 stores the multiple data points in the database 40.

[0025] Sensor 30 is a measuring device that outputs a signal representing the measurement result of the object being measured in biological production. The number of sensors 30 is not limited. Sensors 30 may be, for example, a hydrogen ion concentration (pH: Potential Hydrogen) sensor, a dissolved oxygen (DO: Dissolved Oxygen) sensor, a temperature sensor, a near-infrared sensor, an RGB sensor, a potential sensor, and a torque sensor, but are not limited to these. Sensor 30 is disposed in culture tank 20. Culture tank 20 is a container used for cell culture, etc. Culture tank 20 contains culture medium. Sensor 30 outputs a signal representing the result of measuring the state of culture tank 20 to data extraction system 1.

[0026] Database 40 is a non-temporary storage medium or device for storing information. Database 40 may store evaluation indicators and extracted data, for example. Database 40 may also store signals acquired by sensor 30 as data. Database 40 can be constructed as a single database or as a collection of multiple databases. The location of database 40 is not limited. For example, database 40 may be located within a different computer system than data extraction system 1.

[0027] The data extraction system 1 includes an input unit 2, a preprocessing unit 3, a postprocessing unit 4, an extraction unit 5, and a control block 10 as its functional elements. The control block 10 includes a control unit 6, a setting unit 7, a comparison unit 8, and an anomaly determination unit 9 as its functional elements.

[0028] Input unit 2 acquires signals output from sensor 30 as input signals. For example, input unit 2 acquires M input signals (M > 0) from M sensors 30. Examples of input signals include pH, dissolved oxygen, temperature, spectrum, RGB value, potential (electrode potential), and stirrer speed (Agit), but are not limited to these. Input unit 2 may also acquire signals stored in database 40 as input signals.

[0029] The control unit 6 determines which sensor 30 detected the input signal. For example, the control unit 6 can distinguish the sensor 30 by reading the sensor ID attached to the input signal. The control unit 6 sets the sensor 30 in the optimal manner for the input signal throughout the entire cultivation period. For example, the control unit 6 sets the range of the sensor 30 outputting the spectrum according to the cultivation stage. In the early stages of cultivation, the transmittance of the culture medium is high, so the control unit 6 reduces the range (shortens the exposure time). As cultivation progresses, the transmittance of the culture medium decreases, so the control unit expands the range (extends the exposure time).

[0030] The setting unit 7 sets the input benchmark representing the performance requirements of the data analysis. The setting unit 7 sets the index of the input signal corresponding to the evaluation index as the input benchmark. The setting unit 7 can obtain the evaluation index from the database 40 or an external device, or it can accept the input of the evaluation index. In one example, the input benchmark can be the signal-to-noise ratio (SNR).

[0031] The preprocessing unit 3 performs preprocessing operations on the input signal for comparison with an input reference, generating a preprocessed signal. That is, the preprocessing unit 3 processes the input signal to correspond to the form of the input reference. For example, as preprocessing operations, the preprocessing unit 3 can perform gain adjustment and calibration. As preprocessing operations, the preprocessing unit 3 can also perform operations using multiple input signals. As preprocessing operations, the preprocessing unit 3 can also calculate the signal-to-noise ratio (SNR).

[0032] The comparison unit 8 compares the signal with the input reference corresponding to the signal and generates a comparison result. The comparison result indicates whether the signal meets the input reference. The comparison unit 8 can compare the pre-processed signal with the input reference corresponding to the pre-processed signal and generate a comparison result. The comparison unit 8 can also compare the post-processed signal with the input reference corresponding to the post-processed signal and generate a comparison result.

[0033] If the preprocessing signal does not meet the input reference, the postprocessing unit 4 performs postprocessing operations on the preprocessing signal using other signals obtained from measuring the object being measured, and generates a postprocessing signal. For example, the postprocessing operation can be the acquisition of a summation average or a moving average of the signals. If the preprocessing signal meets the input reference, the postprocessing unit 4 does not perform postprocessing operations. That is, if the preprocessing signal meets the input reference, no postprocessing signal is generated.

[0034] The extraction unit 5 extracts data based on the comparison result obtained by the comparison unit 8. If the preprocessing signal meets the input reference, the extraction unit 5 extracts the preprocessing signal as the extracted data. If the postprocessing signal meets the input reference, the extraction unit 5 extracts the postprocessing signal as the extracted data. The extraction unit 5 may also extract data obtained by performing output operations on the preprocessing or postprocessing signals as the extracted data. Alternatively, if the postprocessing signal does not meet the input reference, the extraction unit 5 may remove the postprocessing signal. In other words, the extraction unit 5 performs data conversion and selection.

[0035] The anomaly detection unit 9 determines anomalies based on input signals, preprocessing signals, or postprocessing signals. For example, the anomaly detection unit 9 may determine anomalies based on changes in input signals or the correlation between input signals. In one example, the anomaly detection unit 9 may sense anomalies based on changes in dissolved oxygen levels. In another example, the anomaly detection unit 9 may sense anomalies based on the correlation between dissolved oxygen levels or pH and electrode potential. As a handling mechanism for anomalies, the anomaly detection unit 9 may invalidate the extracted data. "Invalidating the extracted data" may mean replacing the preprocessing signal, postprocessing signal, or extracted data with null (None) data.

[0036] The evaluation metrics and input benchmarks may vary depending on the content of the input signal. (See reference...) Figure 2 Examples of the three scenarios (Scenario 1, Scenario 2, and Scenario 3) for the evaluation indicators are provided. Figure 2 This is a schematic diagram illustrating an example of the actions of data extraction system 1. Figure 2 In the example shown, three data points are extracted from six input signals.

[0037] In the first case, an evaluation index e1 for evaluating a signal is used. The setting unit 7 converts the evaluation index e1 into an input reference c1. The evaluation index e1 may also be the same index as the input reference c1.

[0038] In one example, when the concentration of an aqueous solution changes by 50 mM (mmol / L), the SNR corresponding to the rate of change of absorbance required to determine this change can be used as an evaluation index e1. Assuming the evaluation index e1 is 20 dB, the setting unit 7 calculates the required SNR for transmitted light (input reference c1) using, for example, the following formula.

[0039] ΔA (SNR) = 20 dB 20×LOG(ΔS / N)-6(dB)=20dB Here, ΔA is the difference in absorbance. ΔS is the difference in transmitted light intensity. N is the floor noise of the photosensor. 6dB is the SNR that degrades up to ΔA calculated from the transmitted light S. When ΔS is set to S × 1%, the required SNR for transmitted light (S) is calculated to be 66dB (= input reference c1). The setting unit 7 converts the evaluation index e1 into the input reference c1 by performing such calculation. The setting unit 7 can also always calculate the absorbance during pretreatment and make the evaluation index e1 consistent with the input reference c1. Such selection can be made depending on the type of culture and the elements to be monitored.

[0040] In one example, there is a case where absorbance changes are monitored based on the absorbance at the initial stage of cultivation, and the changes are small in the early stages but large in the middle and later stages. In this case, even if the evaluation criterion corresponding to the absorbance change ΔA remains the same, the setting unit 7 can separate the input criterion for the initial transmitted light value and the input criterion for the transmitted light after a certain period of time. For example, in the early stages of cultivation where changes are small, the setting unit 7 can extract data with a high SNR input criterion, and in the middle and later stages of cultivation where changes are large, the setting unit 7 can lower the SNR corresponding to the transmitted light. In another example, there is a case where the absorbance change (ΔA) is monitored at regular intervals. In this case, the setting unit 7 can make the evaluation index e1 consistent with the input criterion c1. In this way, by setting the input criterion corresponding to each case, optimal data extraction can be achieved.

[0041] Input unit 2 acquires an input signal x1. The input signal x1 is, for example, a spectrum. Preprocessing unit 3 performs preprocessing operations such as gain adjustment and calibration on the input signal x1 to generate a preprocessed signal f1. For example, gain adjustment of the spectrum absorbs the difference in exposure time. Calibration of the spectrum is dark correction. In another example, at least one of gain adjustment and calibration may be performed corresponding to the type of input signal x1 or the type of sensor 30. Comparison unit 8 compares the preprocessed signal f1 with an input reference c1 to generate a comparison result.

[0042] If the preprocessed signal f1 does not meet the input reference c1, the postprocessing unit 4 performs postprocessing operations such as summation averaging or moving average on the preprocessed signal f1 to generate a postprocessed signal p1. The comparison unit 8 compares the postprocessed signal p1 with the input reference c1 to generate a comparison result.

[0043] If the preprocessing signal f1 satisfies the input reference c1, the extraction unit 5 extracts the preprocessing signal f1 as extracted data d1. If the postprocessing signal p1 satisfies the input reference c1, the extraction unit 5 extracts the postprocessing signal p1 as extracted data d1. If the postprocessing signal p1 does not satisfy the input reference c1, the extraction unit 5 removes the postprocessing signal p1. For example, the extraction unit 5 can remove the postprocessing signal p1 by replacing it with null data. The extraction unit 5 can also extract null data as extracted data d1. The database 40 stores the extracted data d1.

[0044] In the second scenario, an evaluation index e2 is used to evaluate the results obtained from operations on multiple signals. The setting unit 7 converts the evaluation index e2 into an input reference c2.

[0045] Input unit 2 acquires input signal x2 and input signal x3. Input signal x2 is, for example, the natural potential of the platinum electrode relative to GND. GND is the potential that serves as the reference for the measurement system. Natural potential is the potential generated by the redox equilibrium state of a metal electrode immersed in an aqueous solution. Input signal x3 is, for example, the reference electrode potential also used in pH measurement. Regarding this input signal x3, silver / silver chloride (an electrode whose surface has been oxidized by silver chloride) is immersed in KCl (potassium chloride solution) to keep the concentrations of Ag ions and Cl ions constant, thereby obtaining a stable equilibrium potential "Ag(+) + Cl(-) = AgCl". This input signal x3 serves as the reference potential for potential measurement.

[0046] When ΔE is defined as how the platinum electrode potential (E) relative to the reference electrode changes from the initial stage of culture, the SNR corresponding to the rate of change of potential required to know the change in the culture medium can be used as an evaluation index e2. Assuming the evaluation index e2 is 20dB, the setting unit 7 calculates the potential-related SNR (input reference c2) using, for example, the following formula.

[0047] ΔE(SNR) = 20dB 20×LOG(ΔE / N)-3(dB)=20dB Here, ΔE is the potential difference. N is the floor noise of the potential sensor. 3dB is the SNR degradation caused by the differential operation with the reference electrode. When ΔE is set to E × 10%, the required SNR for transmitted light (S) is calculated to be 43dB (=c1). The setting unit 7 converts the evaluation index e2 into the input reference c2 by performing such calculation on the evaluation index e2.

[0048] The preprocessing unit 3 subtracts the input signal x3 from the input signal x2 to generate a preprocessing signal f2, wherein the preprocessing signal f2 is the potential of the input signal x2 (platinum electrode) relative to the reference potential of the input signal x3. A plurality of metal electrodes are also disposed thereon, thereby multi-signaling the input signal x2 and the extracted data d2. The comparison unit 8 compares the preprocessing signal f2 with the input reference c2 to generate a comparison result.

[0049] If the preprocessing signal f2 does not meet the input reference c2, the postprocessing unit 4 performs postprocessing operations such as summation averaging or moving average on the preprocessing signal f2 to generate a postprocessing signal p2. The comparison unit 8 compares the postprocessing signal p2 with the input reference c2 to generate a comparison result.

[0050] If the preprocessing signal f2 satisfies the input reference c2, the extraction unit 5 extracts the preprocessing signal f2 as extracted data d2. If the postprocessing signal p2 satisfies the input reference c2, the extraction unit 5 extracts the postprocessing signal p2 as extracted data d2. If the postprocessing signal p2 does not satisfy the input reference c2, the extraction unit 5 removes the postprocessing signal p2. For example, the extraction unit 5 can remove the postprocessing signal p2 by replacing it with null data. The extraction unit 5 can also extract null data as extracted data d2. The database 40 stores the extracted data d2.

[0051] In the third scenario, evaluation index e3 is used to evaluate multiple signals separately. Input unit 2 acquires input signals x4 to x6. Input signals x4 to x6 are RGB signals (e.g., the brightness values ​​of R, G, and B in the RGB color model). Alternatively, when RGB signals are input, the final signal can be extracted in the form of ΔE*ab in the L*a*b* color space. In this case, the SNR required to set the value of ΔE*ab is used as evaluation index e3. Setting unit 7 sets the SNR required for each RGB signal as input references c31 to c33 (input reference c3).

[0052] The preprocessing unit 3 performs preprocessing operations such as gain adjustment and white balance adjustment on the input signals x4 to x6 respectively, generating preprocessing signals f3 to f5. For example, the white balance adjustment performed on the luminance values ​​of R, G, and B is a calibration for individual unevenness. Preprocessing signal f3 corresponds to input reference c31. Preprocessing signal f4 corresponds to input reference c32. Preprocessing signal f5 corresponds to input reference c33. The comparison unit 8 compares the preprocessing signal f3 with input reference c31, compares the preprocessing signal f4 with input reference c32, and compares the preprocessing signal f5 with input reference c33, generating a comparison result.

[0053] If at least one of the preprocessing signals f3 to f5 does not satisfy its corresponding input references c31 to c33, the postprocessing unit 4 performs postprocessing operations such as summation averaging or moving average on the preprocessing signals f3 to f5 to generate postprocessing signals p3 to p5. Postprocessing signal p1 corresponds to input reference c31. Postprocessing signal p2 corresponds to input reference c32. Postprocessing signal p3 corresponds to input reference c33. The comparison unit 8 compares postprocessing signal p3 with input reference c31, compares postprocessing signal p4 with input reference c32, and compares postprocessing signal p5 with input reference c33, generating a comparison result.

[0054] When the preprocessing signals f3 to f5 satisfy their respective input references c31 to c33, the extraction unit 5 extracts the preprocessing signals f3 to f5 as extracted data d3. When the postprocessing signals p3 to p5 satisfy their respective input references c31 to c33, the extraction unit 5 extracts the postprocessing signals p3 to p5 as extracted data d3. The extraction unit 5 may also extract data obtained by performing output operations on the preprocessing signals f3 to f5 or the postprocessing signals p3 to p5 as extracted data d3. The extraction unit 5 can use the preprocessing signals f3 to f5 or the postprocessing signals p3 to p5 to calculate the value of ΔE*ab in the L*a*b* color space and extract the value of ΔE*ab as extracted data d3. If at least one of the post-processing signals p3 to p5 does not satisfy its corresponding input references c31 to c33, the extraction unit 5 removes the post-processing signals p3 to p5. For example, the extraction unit 5 can remove the post-processing signals p3 to p5 by replacing them with null data. The extraction unit 5 can also extract null data as extracted data d3. The database 40 stores the extracted data d3.

[0055] Figure 3 This is a diagram illustrating an example of spectroscopic measurements in biological production. In Figure 3 In the example shown, a spectral measurement was performed using a culture tank 20, a light source 21, a flow cell 22, and a spectrometer 23. The light source 21 illuminates the flow cell 22. The flow cell 22 draws out the culture medium from the culture tank 20 and circulates it. The transmitted light from the flow cell 22 enters the spectrometer 23. The spectrometer 23 outputs the spectrum.

[0056] Figure 3Figure 24 shows a schematic representation of the output of spectrometer 23. Figure 24 illustrates the signal level group L1 under conditions of multiple spectral measurements and the signal level L2 in darkness. Even when the state of the culture medium flowing in the flow cell 22 is constant, the signal level (signal output) will vary. As a result, there may be a variation F in the signal level group L1. This variation F may be caused, for example, by the setup conditions of hardware such as optical fibers related to the spectral measurement. Figure 3 In the example shown, dark correction and relative intensity correction are performed as preprocessing operations.

[0057] As a preprocessing operation, the preprocessing unit 3 performs dark correction on the spectrum. For example, the preprocessing unit 3 subtracts the signal level L2 under dark conditions from the signal level of the spectrum. As a result, the signal level that varies according to the state of the culture medium flowing in the flow cell 22 is derived.

[0058] As a preprocessing operation, the preprocessing unit 3 can further perform relative intensity correction on the dark-corrected signal. For example, the preprocessing unit 3 uses the signal in a wavelength region R that is not responsive to changes in the culture medium and has a non-zero transmittance (signal level is not zero) to calculate the overall relative intensity. "Not responsive to changes in the culture medium" means that the signal level does not change even if there is a change in the culture medium. For example, the absorption of light varies at different wavelengths until light passes through an object. This is because the absorption wavelengths differ depending on the molecular structure of the object. At wavelengths that do not react to any molecular structure, light becomes a wavelength that is "not responsive to changes in the culture medium," that is, "the signal level does not change even if there is a change in the culture medium." For example, a wavelength region around 1300 nm is an example of a wavelength region R that is not responsive to changes in the culture medium and has a non-zero transmittance. The preprocessing unit 3 uses the dark-corrected signal to calculate the relative intensity relative to the signal level in this wavelength region R. That is, the preprocessing unit 3 generates a preprocessing signal by calculating the relative intensity of the signal level in a wavelength region that is not responsive to changes in the culture medium and has a non-zero transmittance.

[0059] Spectroscopic measurements can be performed using near-infrared spectroscopy. In near-infrared spectroscopy, there exists a wavelength region (e.g., 1250 nm to 1350 nm) where the signal level does not respond to changes in organic molecules and where water absorption is not excessive. "Not responding to changes in organic molecules" means that even with changes in organic molecules, the signal level does not change or is not affected by the changes in organic molecules. "Water absorption is not excessive" means that light is transmitted at a certain intensity, allowing for stable observation of the light intensity by the spectrometer 23. In other words, it means that the light intensity (signal level) required for near-infrared spectroscopy is not affected by water absorption. Here, the appropriate absorbance range depends on the object being measured, for example, 0.05 to 1.5. In the wavelength region where water absorption is not excessive, for example, the absorbance can be 1.0 or less (absorbance rate 90%). The wavelength region where the signal level does not respond to changes in organic molecules and where water absorption is not excessive is an example of wavelength region R. The preprocessing unit 3 uses a dark-corrected signal to calculate the relative intensity of the signal level relative to this wavelength region. That is, the pretreatment unit 3 generates a pretreatment signal by calculating the relative intensity of the signal level in a wavelength region R that is not responsive to changes in organic molecules and where water absorption is not excessive.

[0060] exist Figure 3 In the example shown, input unit 2 and preprocessing unit 3 function as part of a data extraction system or sensing system. The data extraction system acquires the spectrum measured by spectral analysis in biological production. The data extraction system outputs a signal obtained by calculating the relative intensity of the spectrum's signal level with respect to the wavelength region where the transmittance is not zero and does not respond to changes in the culture medium.

[0061] Reference Figure 4 (a) ~ Figure 4 (c) explains the relative intensity correction. Figure 4 (a) ~ Figure 4 (c) is an example graph illustrating the changes in organic molecule concentrations during near-infrared spectroscopy measurements. Figure 4 (a) ~ Figure 4 In (c), the vertical axis represents the signal level, and the horizontal axis represents the wavelength. Figure 4 (a) ~ Figure 4 (c) shows the wavelength range (e.g., 1250 nm–1350 nm) that does not respond to changes in organic molecules and where water absorption is not excessive, in the form of "wavelength regions that do not change". Furthermore, Figure 4 (a) ~ Figure 4 (c) shows the calibration curve of glucose (1600 nm) in the form of “wavelength of the target substance”.

[0062] Figure 4(a) is a graph representing the signal level obtained by circulating tap water instead of culture medium in flow cell 22 and performing near-infrared spectroscopy measurements for several hours. Figure 4 (a) indicates that the signal level of the near-infrared spectrum has changed.

[0063] Figure 4 (b) indicates that... Figure 4 The signal level graph shown in (a) is obtained by performing dark correction on the signal level. That is, Figure 4 (b) is from Figure 4 Example (a) shows the signal level minus the signal level in darkness. When... Figure 4 (b) and Figure 4 When comparing (a), it can be said that the offset is eliminated through dark correction.

[0064] Figure 4 (c) indicates the use of Figure 4 The graph shown in (b) is a graph of signal levels obtained by converting the signal level near 1300 nm (e.g., 1250 nm to 1350 nm) into relative intensity. Figure 4 The signal levels shown in (c) all exhibit the same shape. That is, it can be said that variations in the signal level are eliminated through relative intensity correction. Therefore, the signal level obtained by converting to relative intensity easily captures changes in the wavelength region of the target substance.

[0065] Figure 5 This is a schematic diagram illustrating another example of the actions of a data extraction system. Figure 5 It shows the use of Figure 3 The example shown illustrates a configuration used to acquire an input signal. Figure 5 In this process, evaluation metrics e4 and e5 are used to evaluate SNR. Setting unit 7 sets the input reference c4 corresponding to evaluation metric e4. Setting unit 7 sets the input reference c5 corresponding to evaluation metric e5. Input unit 2 acquires the input signal x7 from spectrometer 23. Input signal x7 is a spectrum. Similar to the calculation method of input reference c1 based on evaluation metric e1, setting unit 7 converts evaluation metrics e4 and e5 into input references c4 and c5.

[0066] In one example, the SNR corresponding to the rate of change of absorbance required to determine the change in the turbidity of a certain aqueous solution can be used as the evaluation index e4. Assuming the evaluation index e4 is 10 dB, the setting unit 7 calculates the required SNR for transmitted light (input reference c4) for example using the following formula.

[0067] ΔA (SNR) = 10 dB 20×LOG(ΔS / N)-6(dB)=10dB Here, ΔA is the difference in absorbance. ΔS is the difference in transmitted light intensity. N is the floor noise of the optical sensor. 6dB is the SNR that degrades up to ΔA calculated from the transmitted light S. When ΔS is set to S × 4%, the required SNR for transmitted light (S) is calculated to be 44dB (= input reference c4). The setting unit 7 converts the evaluation index e4 into the input reference c4 by performing such calculation on the evaluation index e4. For example, the setting unit 7 converts the evaluation index e5 into the input reference c5 by setting ΔA (SNR) = 20dB and performing the same calculation on the evaluation index e5.

[0068] The preprocessing unit 3 performs dark correction on the input signal x7 to generate a preprocessing signal f6. If the preprocessing signal f6 does not meet the input reference c4, the postprocessing unit 4 performs postprocessing operations on the preprocessing signal f6 to generate a postprocessing signal p6. The dark-corrected input signal x7 is finally extracted as the extracted data d4.

[0069] The preprocessing unit 3 performs dark correction and relative intensity correction on the input signal x7 to generate a preprocessed signal f7. If the preprocessed signal f7 does not meet the input reference c5, the postprocessing unit 4 performs postprocessing operations on the preprocessed signal f7 to generate a postprocessed signal p7. The input signal x7, after dark correction and relative intensity correction, is finally extracted as the extracted data d5.

[0070] like Figure 5 As shown in the example, the processing can also be branched based on the input signal x7, thereby extracting extraction data d4 and extraction data d5. Extraction data d4 and extraction data d5 can be used depending on the purpose of data analysis, depending on the desired substance to be captured due to variations in the culture medium.

[0071] It can be said that, Figure 5The example shown illustrates the processing of data obtained by performing different preprocessing operations on the same signal. The preprocessing unit 3 generates a first preprocessed signal (preprocessed signal f6) without calculating relative intensity, and generates a second preprocessed signal (preprocessed signal f7) with calculated relative intensity. The setting unit 7 sets a first input reference (input reference c4) as the input reference corresponding to the first preprocessed signal, and sets a second input reference (input reference c5) as the input reference corresponding to the second preprocessed signal. If the first preprocessed signal does not satisfy the first input reference, the postprocessing unit 4 performs postprocessing operations on the first preprocessed signal to generate a first postprocessed signal (postprocessed signal p6). If the first preprocessed signal satisfies the first input reference, the extraction unit 5 extracts the first preprocessed signal as first extracted data (extracted data d4); if the first postprocessed signal satisfies the first input reference, the extraction unit 5 extracts the first postprocessed signal as first extracted data. If the second preprocessed signal does not satisfy the second input reference, the postprocessing unit 4 performs postprocessing operations on the second preprocessed signal to generate a second postprocessed signal (postprocessed signal p7). When the second preprocessing signal satisfies the second input reference, the extraction unit 5 extracts the second preprocessing signal as the second extracted data (extracted data d5). When the second postprocessing signal satisfies the second input reference, the extraction unit 5 extracts the second postprocessing signal as the second extracted data.

[0072] Figure 6 This is a summary diagram illustrating yet another example of the actions of a data extraction system. Figure 6 In addition to showing Figure 5 In addition to the processing shown, examples of setting different input benchmarks are also illustrated. Figure 6 In this process, evaluation metrics e4 to e7 are used to evaluate SNR. Setting unit 7 sets the input reference c4 corresponding to evaluation metric e4. Setting unit 7 sets the input reference c5 corresponding to evaluation metric e5. Setting unit 7 sets the input reference c6 corresponding to evaluation metric e6. Setting unit 7 sets the input reference c7 corresponding to evaluation metric e7. Input unit 2 acquires the input signal x7 from spectrometer 23. Input signal x7 is a spectrum.

[0073] For example, the setting unit 7 calculates the required SNR (input reference c4 to input reference c7) of the transmitted light using the following formula.

[0074] • Input baseline c4 ΔA (SNR) = 10 dB 20×LOG(ΔS / N)-6(dB)=10dB • Input baseline c5 ΔA (SNR) = 20 dB 20×LOG(ΔS / N)-6(dB)=20dB • Input base c6 ΔA(SNR) = 0 dB 20×LOG(ΔS / N)-6(dB)=0dB • Input baseline c7 ΔA (SNR) = 10 dB 20×LOG(ΔS / N)-6(dB)=10dB Here, ΔA is the difference in absorbance. ΔS is the difference in transmitted light intensity. N is the floor noise of the optical sensor. 6dB is the SNR that degrades up to ΔA calculated from the transmitted light S. With input reference c4, when ΔS = S × 4%, the required SNR for transmitted light (S) is calculated to be 44dB. With input reference c5, when ΔS = S × 1%, the required SNR for transmitted light (S) is calculated to be 66dB. With input reference c6, when ΔS = S × 4%, the required SNR for transmitted light (S) is calculated to be 34dB. With input reference c7, when ΔS = S × 1%, the required SNR for transmitted light (S) is calculated to be 54dB. The setting unit 7 converts evaluation indices e4 to e7 into input references c4 to c7 by performing such calculations on evaluation indices e4 to e7.

[0075] The preprocessing unit 3 performs dark correction on the input signal x7 to generate a preprocessing signal f6. If the preprocessing signal f6 does not meet the input reference c4, the postprocessing unit 4 performs postprocessing operations on the preprocessing signal f6 to generate a postprocessing signal p6. The dark-corrected input signal x7 is finally extracted as the extracted data d4.

[0076] The preprocessing unit 3 performs dark correction and relative intensity correction on the input signal x7 to generate a preprocessed signal f7. If the preprocessed signal f7 does not meet the input reference c5, the postprocessing unit 4 performs postprocessing operations on the preprocessed signal f7 to generate a postprocessed signal p7. The input signal x7, after dark correction and relative intensity correction, is finally extracted as the extracted data d5.

[0077] The preprocessing unit 3 performs dark correction on the input signal x7 to generate a preprocessing signal f8. Preprocessing signal f8 is the same as preprocessing signal f6. If preprocessing signal f8 does not meet the input reference c6, the postprocessing unit 4 performs postprocessing operations on preprocessing signal f8 to generate a postprocessing signal p8. The dark-corrected input signal x7 is finally extracted as extracted data d6.

[0078] The preprocessing unit 3 performs dark correction and relative intensity correction on the input signal x7 to generate a preprocessing signal f9. The preprocessing signal f9 is identical to the preprocessing signal f7. If the preprocessing signal f9 does not meet the input reference c7, the postprocessing unit 4 performs postprocessing operations on the preprocessing signal f9 to generate a postprocessing signal p9. The input signal x7, after dark correction and relative intensity correction, is finally extracted as the extracted data d7.

[0079] Input references c4 and c6 are common in that they are both used for comparison with the pre-processed signal that has undergone dark correction, but their values ​​are different. That is, input references c4 and c6 can be said to be different input references corresponding to the pre-processed signal that has undergone dark correction. When the SNR of input reference c4 is greater than the SNR of input reference c6, the SNR of extracted data d4 is a higher-precision signal than the SNR of extracted data d6. The response of extracted data d6 is higher than the response of extracted data d4. The response is determined based on the time involved in the post-processing operations.

[0080] Input references c5 and c7 are common in that they are both used for comparison with the preprocessed signal that has undergone dark correction and relative intensity correction, but their values ​​are different. That is, it can be said that input references c5 and c7 are different input references corresponding to the preprocessed signal that has undergone dark correction and relative intensity correction. When the SNR of input reference c5 is greater than the SNR of input reference c7, the extracted data d5 has a higher SNR than the extracted data d7, indicating a higher precision signal. The response of extracted data d7 is higher than the response of extracted data d5.

[0081] It can be said that, Figure 6 The example shown illustrates a process that extracts data with different properties by setting different input criteria for the same signal. The setting unit 7 also sets other input criteria corresponding to the preprocessed signal. If the preprocessed signal does not meet the other input criteria, the post-processing unit 4 performs post-processing operations on the preprocessed signal to further generate other post-processed signals. If the preprocessed signal meets the other input criteria, the extraction unit 5 also extracts other preprocessed signals as additional extracted data; if other post-processed signals meet the other input criteria, the extraction unit 5 also extracts other post-processed signals as additional extracted data.

[0082] Reference Figure 7 (a) ~ Figure 7 (d) and Figure 8 (a) ~ Figure 8 (d) illustrates an example of the improved quality of the data used for analysis as an effect of this disclosure. Figure 7(a) ~ Figure 7 (d) is a diagram illustrating an example of the effect of preprocessing operations and input baseline settings. Figure 7 (a) ~ Figure 7 (d) shows an example of conducting a culture experiment for several days and observing lactic acid as the target substance.

[0083] Figure 7 (a) illustrates an example of sampling and determining the concentration of lactic acid using high-performance liquid chromatography (HPLC). Figure 7 In (a), the vertical axis represents the concentration of lactic acid, and the horizontal axis represents time. Figure 7 In (a), inflection points r1 and r2 appear.

[0084] Figure 7 (b) ~ Figure 7 (d) illustrates the following example: data is extracted under different conditions based on data observed in real time every minute using near-infrared spectroscopy. Figure 7 (b) ~ Figure 7 In (d), the vertical axis represents absorbance (the difference from the initial value), and the horizontal axis represents time.

[0085] Figure 7 (b) illustrates the following example: using SNR as the input reference, the value of 1700 nm (the calibration curve of lactic acid) is extracted from the preprocessed data after dark correction and relative intensity correction. When... Figure 7 (a) and Figure 7 When comparing (b), it can be seen that the inflection point t1 corresponding to the inflection point r1 and the inflection point t2 corresponding to the inflection point r2 are captured.

[0086] Figure 7 (c) illustrates the following example: using SNR as the input reference, the value of 1700 nm (the calibration curve for lactic acid) is extracted from the data that has undergone dark correction as a preprocessing operation. Figure 7 In (c), the signal increases monotonically, making it impossible to determine the change in lactic acid concentration. However, this monotonous increase reflects turbidity. Such a monotonous increase in turbidity is highly correlated with the yield of the product produced through cultivation.

[0087] Figure 7 (d) illustrates the following example: without applying an input reference, the value of 1700 nm (the calibration curve of lactic acid) is extracted from the preprocessed data after dark correction and relative intensity correction. Figure 7In (d), inflection points t11 and t21 are buried in noise. It can be said that such a signal is noisy but highly sensitive. In cases where large changes occur instantaneously during cultivation, using highly sensitive signals in data analysis is also useful.

[0088] Figure 8 (a) ~ Figure 8 (d) is a diagram that illustrates another example of how the effects of preprocessing operations and input baseline settings can be used to explain the results. Figure 8 (a) ~ Figure 8 (d) shows an example of conducting a culture experiment for several days and using glucose as the target substance for observation.

[0089] Figure 8 (a) shows an example of sampling and determining the concentration of glucose by HPLC. Figure 8 In (a), the vertical axis represents glucose concentration, and the horizontal axis represents time. Figure 8 In (a), inflection points r3 and r4 appear.

[0090] Figure 8 (b) ~ Figure 8 (d) illustrates the following example: data is extracted under different conditions based on data observed in real time every minute using near-infrared spectroscopy. Figure 8 (b) ~ Figure 8 In (d), the vertical axis represents absorbance (the difference from the initial value), and the horizontal axis represents time.

[0091] Figure 8 (b) illustrates the following example: using SNR as the input benchmark, the value of 1600 nm (the calibration curve for glucose) is extracted from the preprocessed data after dark correction and relative intensity correction. When... Figure 8 (a) and Figure 8 When comparing (b), it can be seen that the inflection point t3 corresponding to the inflection point r3 and the inflection point t4 corresponding to the inflection point r4 are captured.

[0092] Figure 8 (c) illustrates the following example: using SNR as the input benchmark, the value of 1600 nm (the calibration curve of glucose) is extracted from the data that has undergone dark correction as a preprocessing operation. Figure 8 In (c), the signal increases monotonically, making it impossible to determine changes in glucose concentration. However, this monotonous increase reflects turbidity. Such a monotonous increase in turbidity is highly correlated with the yield of the product produced through cultivation.

[0093] Figure 8Example (d) illustrates the following: without applying an input reference, the value of 1600 nm (the calibration curve of glucose) is extracted from the preprocessed data after dark correction and relative intensity correction. Figure 8 In (d), inflection points t31 and t41 are buried in noise. It can be said that such a signal is noisy but highly sensitive. In cases where large changes occur instantaneously during cultivation, using highly sensitive signals in data analysis is also useful.

[0094] [Example] Figure 9 This is an overall diagram illustrating an example of the application of the data extraction system 1 in the embodiment. Figure 9 The illustrated biological production system 50 includes a culture tank 20, multiple sensors 30 (pH sensor 31, DO sensor 32, temperature sensor 33, near-infrared sensor 34, RGB sensor 35, and potential sensor 36), a culture control device 51, a sensor control device 52, a control terminal 53, and a cloud system 54. The control terminal 53 is communicatively connected to the culture control device 51 and the sensor control device 52 via a communication network 59. The cloud system 54 is also communicatively connected to the culture control device 51 and the sensor control device 52 via the communication network 59. The communication network 59 can be either wired or wireless. The communication network 59 can be a non-dedicated line such as an internet line or a mobile communication network, or it can be a dedicated line.

[0095] pH sensor 31, DO sensor 32, and temperature sensor 33 acquire essential data for cultivation (hereinafter referred to as "cultivation data"). The cultivation data may also include the stirrer speed (Agit) acquired by a torque sensor, etc. The cultivation control device 51 controls the cultivation tank 20, pH sensor 31, DO sensor 32, and temperature sensor 33. The cultivation control device 51 adds sensor data information, etc., to the signals output from pH sensor 31, DO sensor 32, and temperature sensor 33, and sends these signals to the cloud system 54. Examples of sensor data information include, but are not limited to, version numbers and serial numbers.

[0096] Near-infrared sensor 34, RGB sensor 35, and potential sensor 36 are used to enable AI (Artificial Intelligence) in bioproduction. In AI-driven bioproduction, machine learning models are used to predict optimal control values ​​in real time and automatically control the culture according to these predictions. Sensor control device 52 controls the settings of near-infrared sensor 34, RGB sensor 35, and potential sensor 36. Sensor control device 52 adds sensor data information to the signals output from near-infrared sensor 34, RGB sensor 35, and potential sensor 36 and sends these signals to cloud system 54.

[0097] Control terminal 53 is one or more computers used by users of the biological production system 50. Control terminal 53 may be, for example, a personal computer, but is not limited to this. Control terminal 53 handles various settings and controls, for example, based on a web-based user interface (UI). Control terminal 53 modifies the settings of the culture control device 51, sensor control device 52, and cloud system 54. Control terminal 53 controls the culture control device 51 and sensor control device 52.

[0098] The cloud system 54 includes a data extraction system 1, a database 40, and a data analysis system 55. The database 40 can store signals received from the culture control device 51 and the sensor control device 52 as data. The data extraction system 1 can acquire signals read from the database 40 as input signals, process them, and store the extracted data in the database 40.

[0099] Regarding data extraction, multiple data points exist for each sensor (30). Furthermore, based on the patterns of preprocessing operations, input benchmarks, and post-processing operations, the extracted data is extracted in multiple data formats, resulting in a highly multidimensional set of parameters. The data analysis system (55) processes the extracted data read from the database (40) to perform data analysis. The data analysis system (55) includes a dimensionality reduction unit (56), an explanatory variable extraction unit (57), and a machine learning unit (58) as functional elements.

[0100] The dimensionality reduction unit 56 compresses the dimensions of the extracted data, for example, through principal component analysis. The explanatory variable extraction unit 57 determines explanatory variables from the dimensionally compressed signals based on correlation and regression analysis with the target variable in machine learning. The machine learning unit 58 uses the determined explanatory variables to perform machine learning and create a machine learning model. The bioproduction system 50 uses the machine learning model to control the cultivation process in real time and automatically.

[0101] The extracted data disclosed herein meets the input benchmark, thus ensuring data quality. By using such extracted data for machine learning, efficient machine learning can be achieved.

[0102] Reference Figure 10 An example of the operation of biological production system 50 will be used to illustrate this. Figure 10 This is a diagram illustrating an example of the overall operation of the biological production system 50, which includes the data extraction system 1.

[0103] In step S1, the control terminal 53 sets the parameters of the culture control device 51, the sensor control device 52, and the cloud system 54. Examples of parameters include, but are not limited to, the culture type, culture time, various sensor settings (sampling interval, integration time, and gain), and the ID used to determine the culture data.

[0104] In step S2, the cultivation control device 51, the sensor control device 52, and the cloud system 54 respectively respond to the parameter settings from the control terminal 53 and perform various settings.

[0105] In step S3, the control terminal 53 initiates the operation of the culture control device 51 and the sensor control device 52. For example, the control terminal 53 can initiate the operation by sending a command indicating the start of the operation to the culture control device 51 and the sensor control device 52 via a Web UI.

[0106] In step S4, the culture control device 51 and the sensor control device 52 begin operation in response to the start control from the control terminal 53. The culture control device 51 controls the pH sensor 31, DO sensor 32, and temperature sensor 33 to start the culture. The sensor control device 52 controls the near-infrared sensor 34, RGB sensor 35, and potential sensor 36 to start the measurement in the culture tank 20.

[0107] In step S5, the control terminal 53 performs stop control to halt the operations of the culture control device 51 and the sensor control device 52. For example, the control terminal 53 can perform stop control by sending a command to stop the operations to the culture control device 51 and the sensor control device 52 via a Web UI.

[0108] In step S6, the culture control device 51 and the sensor control device 52, in response to start and stop controls from the control terminal 53, respectively, transmit the collected data to the cloud system 54. For example, the culture control device 51 adds sensor data information to the signals output from the pH sensor 31, DO sensor 32, and temperature sensor 33, and sends the signals to the cloud system 54. The sensor control device 52 adds sensor data information to the signals output from the near-infrared sensor 34, RGB sensor 35, and potential sensor 36, and sends the signals to the cloud system 54. The culture control device 51 and the sensor control device 52 may also transmit various data to the cloud system 54 at predetermined time intervals during the culture period.

[0109] In step S7, database 40 stores various data received from culture control device 51 and sensor control device 52. Alternatively, when various data are received from culture control device 51 and sensor control device 52, cloud system 54 initiates subsequent processing via event-driven mechanisms.

[0110] In step S8, the data extraction system 1 reads various data from the database 40 as input signals and extracts the extracted data. The detailed processing of the data extraction system 1 will be described later. The data extraction system 1 stores the extracted data in the database 40.

[0111] In step S9, as a non-real-time process, the data analysis system 55 performs machine learning based on the extracted data. The dimensionality reduction unit 56 compresses the dimensions of the extracted data, for example, through principal component analysis. The explanatory variable extraction unit 57 determines explanatory variables from the dimensionally compressed signals based on correlation and regression analysis with the target variable in machine learning. The machine learning unit 58 uses the determined explanatory variables to perform machine learning, creating predictive models for predicting lactate levels, glucose levels, and product yields, as well as control models for controlling cultivation, and other machine learning models.

[0112] The processes in steps S1 through S9 can also be repeated. For example, the biological production system 50 can also collect data according to each mode of the parameters to create a machine learning model.

[0113] In step S10, as a real-time process, the cloud system 54 performs cultivation prediction. For example, the machine learning unit 58 uses a prediction model to perform cultivation prediction.

[0114] In step S11, as a real-time process, the cloud system 54 performs cultivation control. For example, the machine learning unit 58 uses the results of cultivation prediction and the control model to create control variables for controlling the cultivation control device 51. The data analysis system 55 sends the control variables to the cultivation control device 51.

[0115] In step S12, the culture control device 51, in response to control variables received from the cloud system 54, controls the amount of alkali added to control pH, the amount of oxygen injected to control DO, temperature, Agit, etc. The results are reflected in the culture tank 20, thus achieving automatic control of the culture.

[0116] Reference Figure 11 An example of the actions of data extraction system 1 will be used to illustrate this. Figure 11 This is a diagram illustrating an example of the actions of data extraction system 1. In Figure 11 Examples of measured culture data, near-infrared light data, RGB data, and potential data will be explained below. In one example, the culture data are DO, pH, Agit, and temperature. The following explanation assumes that the input reference has been set in the setting unit 7.

[0117] In step S101, multiple sensors 30 (e.g., pH sensor 31, DO sensor, torque sensor, and temperature sensor 33) send signals representing the measurement results of DO, pH, Agit, and temperature to the culture control device 51. The culture control device 51 adds sensor data information from the pH sensor 31, DO sensor, torque sensor, and temperature sensor 33 to the signal representing the measurement results, and then sends the signal to the cloud system 54. The database 40 stores the signals (culture data) received from the culture control device 51. The input unit 2 of the data extraction system 1 obtains culture data as input signals by reading the culture data from the database.

[0118] In step S102, the near-infrared sensor 34 sends a signal representing the measurement result of the spectrum to the sensor control device 52. The sensor control device 52 assigns sensor data information, etc., to the signal representing the measurement result to the near-infrared sensor 34 and sends the signal to the cloud system 54. The database 40 stores the signal (near-infrared light data) received from the sensor control device 52. The input unit 2 of the data extraction system 1 obtains near-infrared light data as an input signal by reading the near-infrared light data from the database 40.

[0119] In step S103, the RGB sensor 35 sends a signal representing the measurement result of RGB to the sensor control device 52. The sensor control device 52 adds sensor data information of the RGB sensor 35 to the signal representing the measurement result and sends the signal to the cloud system 54. The database 40 stores the signals (RGB data) received from the sensor control device 52. The input unit 2 of the data extraction system 1 obtains RGB data as an input signal by reading the RGB data from the database 40.

[0120] In step S104, the potential sensor 36 sends a signal representing the measurement result of the potential to the sensor control device 52. The sensor control device 52 adds sensor data information of the potential sensor 36 to the signal representing the measurement result and sends the signal to the cloud system 54. The database 40 stores the signals (potential data) received from the sensor control device 52. The input unit 2 of the data extraction system 1 obtains potential data as an input signal by reading the potential data from the database 40.

[0121] In step S105, the preprocessing unit 3 performs preprocessing operations such as gain adjustment and calibration on the input signal representing the culture data.

[0122] In step S106, the preprocessing unit 3 performs dark correction on the input signal representing near-infrared light data.

[0123] In step S107, the preprocessing unit 3 performs dark correction and relative intensity correction on the input signal, which is near-infrared light data.

[0124] In step S108, the preprocessing unit 3 performs preprocessing operations such as gain adjustment and calibration on the input signal representing RGB data.

[0125] In step S109, the preprocessing unit 3 performs preprocessing operations such as gain adjustment and calibration on the input signal representing potential data.

[0126] In step S110, the preprocessing unit 3 uses the signal obtained by performing preprocessing operations on the input signal representing the culture data to further calculate the SNR, thereby generating a preprocessing signal related to the culture data.

[0127] In step S111, the preprocessing unit 3 uses the signal obtained by performing dark correction on the input signal representing near-infrared light data to further calculate the SNR, thereby generating a first preprocessing signal related to the near-infrared light data.

[0128] In step S112, the preprocessing unit 3 uses the signal obtained by performing dark correction and relative intensity correction on the input signal representing near-infrared light data to further calculate the SNR, thereby generating a second preprocessing signal related to the near-infrared light data.

[0129] In step S113, the preprocessing unit 3 uses the signal obtained by performing preprocessing operations on the input signal representing RGB data to further calculate the SNR, thereby generating a preprocessing signal related to the RGB data.

[0130] In step S114, the preprocessing unit 3 uses the signal obtained by preprocessing the input signal representing the potential data to further calculate the SNR, thereby generating a preprocessing signal related to the potential data.

[0131] The method for calculating SNR in steps S110 to S114 is not limited. In one example, the preprocessing unit 3 may measure the noise level (N) of the sensor 30 for each range in advance, and calculate the SNR [dB] using the signal level (S) and 20log(S / N). In another example, the preprocessing unit 3 may obtain the signal (S1) obtained by removing noise from the actual signal level (S), extract the noise (N) by subtracting the noise-removed signal (S1) from the actual signal level (S), and calculate the SNR [dB] using 20log(S1 / N).

[0132] In step S115, the comparison unit 8 compares the preprocessing signal related to the culture data with the input benchmark corresponding to the signal, and generates a comparison result related to the culture data.

[0133] In step S116, the comparison unit 8 compares the first preprocessing signal associated with the near-infrared light data with the first input reference corresponding to the signal to generate a first comparison result associated with the near-infrared light data.

[0134] In step S117, the comparison unit 8 compares the second preprocessing signal associated with the near-infrared light data with the second input reference corresponding to the signal, and generates a second comparison result associated with the near-infrared light data.

[0135] In step S118, the comparison unit 8 compares the preprocessing signal related to the RGB data with the input reference corresponding to the signal to generate a comparison result related to the RGB data.

[0136] In step S119, the comparison unit 8 compares the preprocessing signal related to the potential data with the input reference corresponding to the signal, and generates a comparison result related to the potential data.

[0137] In step S120, if the preprocessing signal related to the culture data does not meet the input benchmark corresponding to the signal, the postprocessing unit 4 performs a postprocessing operation on the preprocessing signal to obtain an additive average or a moving average, thereby generating a postprocessing signal related to the culture data. If the preprocessing signal related to the culture data meets the input benchmark corresponding to the signal, the postprocessing unit 4 does not perform a postprocessing operation.

[0138] In step S121, the anomaly determination unit 9 determines the anomaly in the culture based on the input signal. The anomaly determination unit 9 may also sense the anomaly based on a preprocessing signal related to the culture data or a postprocessing signal related to the culture data. The processing in step S121 may also be performed before step S120.

[0139] The anomaly detection unit 9 can detect anomalies in the culture medium based on changes (decreases) in dissolved oxygen (DO). For example, the anomaly detection unit 9 can sense a sharp decrease in DO by comparing the change in DO over a specified time period with a specified threshold. A sharp decrease in DO may indicate solidification of the culture medium. However, solidification of the culture medium may not necessarily occur when Agit is at a low rotation level, so the anomaly detection unit 9 can also compare Agit with a threshold corresponding to Agit.

[0140] The anomaly detection unit 9 can also detect anomalies in electrode potential based on the correlation between dissolved oxygen (DO) or pH and electrode potential. The anomaly detection unit 9 can also monitor the correlation between DO or pH and electrode potential, and determine that electrode potentials deviating from the correlation are in an abnormal state. For example, a lower DO value (decreasing) or a higher pH value (increasing) indicates a more alkaline change in the culture medium. In potential measurements, there is a correlation that the electrode potential tends to decrease as the culture medium becomes more alkaline. The anomaly detection unit 9 can also determine that the electrode potential is in an abnormal state if an action occurs that differs from this correlation.

[0141] In the event of an anomaly, as a handling procedure, the anomaly determination unit 9 can send an error message to the control terminal 53 or similar device, or control the sensor 30. As a handling procedure, the anomaly determination unit 9 replaces all measurement data with null values. For example, the anomaly determination unit 9 replaces culture data, near-infrared light data, RGB data, and potential data with null values.

[0142] In step S122, if the first preprocessing signal related to the near-infrared light data does not meet the first input reference corresponding to the signal, the post-processing unit 4 performs a post-processing operation on the signal to obtain an additive average or a moving average, thereby generating a first post-processing signal related to the near-infrared light data. If the first preprocessing signal related to the near-infrared light data meets the first input reference corresponding to the signal, the post-processing unit 4 does not perform a post-processing operation.

[0143] In step S123, if the second preprocessing signal related to the near-infrared light data does not meet the input reference corresponding to the signal, the post-processing unit 4 performs a post-processing operation on the signal to obtain an additive average or a moving average, thereby generating a second post-processing signal related to the near-infrared light data. If the second preprocessing signal related to the near-infrared light data meets the second input reference corresponding to the signal, the post-processing unit 4 does not perform a post-processing operation.

[0144] In step S124, if the preprocessing signal related to the RGB data does not meet the input reference corresponding to the signal, the postprocessing unit 4 performs a postprocessing operation on the preprocessing signal to obtain an summation average or a moving average, generating a postprocessing signal related to the RGB data. If the preprocessing signal related to the RGB data meets the input reference corresponding to the signal, the postprocessing unit 4 does not perform a postprocessing operation.

[0145] In step S125, if the preprocessing signal related to the potential data does not meet the input reference corresponding to the signal, the postprocessing unit 4 performs a postprocessing operation on the preprocessing signal to obtain an summation average or a moving average, generating a postprocessing signal related to the potential data. If the preprocessing signal related to the potential data meets the input reference corresponding to the signal, the postprocessing unit 4 does not perform a postprocessing operation.

[0146] In one example, the input reference is the signal-to-noise ratio (SNR). When the sensor 30 captures changes in the culture medium, the difference from the initial value is monitored, so the signal level of the input signal is expected to be very small. If the input reference SNR is 10 dB and the preprocessed signal SNR is 0 dB, the preprocessed signal does not meet the input reference. Since the input signal is obtained once per minute, the post-processing unit 4 uses the input signals for ten minutes (ten times) to obtain a summed average. Therefore, the 10 dB difference in the comparison result can be eliminated.

[0147] In another example, when the input reference SNR is 20dB and the preprocessing signal SNR is 0dB, the preprocessing signal does not meet the input reference. With an input signal acquired once per minute, the post-processing unit 4 uses 100 ten-minute input signals to obtain a summed average. Thus, the 20dB difference in the comparison result can be eliminated. That is, acquiring the input signal for post-processing calculations requires 1000 minutes (approximately 16.7 hours). Here, if the signal level of the input signal increases tenfold after one hour (the difference relative to the initial value increases tenfold), then 20dB of extracted data can be acquired in ten-minute increments.

[0148] In step S126, if the preprocessing signal related to the culture data satisfies the input criteria corresponding to that signal, the extraction unit 5 extracts the preprocessing signal related to the culture data as extracted data related to the culture data. If the postprocessing signal related to the culture data satisfies the input criteria corresponding to that signal, the extraction unit 5 extracts the postprocessing signal related to the culture data as extracted data related to the culture data. If the postprocessing signal related to the culture data does not satisfy the input criteria corresponding to that signal, the extraction unit 5 removes the postprocessing signal related to the culture data. For example, the extraction unit 5 can remove the postprocessing signal related to the culture data by replacing it with null data. The extraction unit 5 can also extract null data as extracted data related to the culture data.

[0149] In step S127, if the first preprocessing signal related to the near-infrared light data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the first preprocessing signal related to the near-infrared light data to calculate the absorbance. The extraction unit 5 extracts the absorbance as the first extracted data related to the near-infrared light data. If the first postprocessing signal related to the near-infrared light data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the first postprocessing signal related to the near-infrared light data to calculate the absorbance. The extraction unit 5 extracts the absorbance as the first extracted data related to the near-infrared light data. If the first postprocessing signal related to the near-infrared light data does not satisfy the first input reference corresponding to that signal, the extraction unit 5 removes the first postprocessing signal related to the near-infrared light data. For example, the extraction unit 5 can remove the first postprocessing signal related to the near-infrared light data by replacing it with null data. The extraction unit 5 can also extract null data as the first extracted data related to the near-infrared light data.

[0150] In step S127, if the second preprocessing signal related to the near-infrared light data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the second preprocessing signal related to the near-infrared light data to calculate the absorbance. The extraction unit 5 extracts the absorbance as the second extracted data related to the near-infrared light data. If the second postprocessing signal related to the near-infrared light data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the second postprocessing signal related to the near-infrared light data to calculate the absorbance. The extraction unit 5 extracts the absorbance as the second extracted data related to the near-infrared light data. If the second postprocessing signal related to the near-infrared light data does not satisfy the second input reference corresponding to that signal, the extraction unit 5 removes the second postprocessing signal related to the near-infrared light data. For example, the extraction unit 5 can remove the second postprocessing signal related to the near-infrared light data by replacing it with null data. The extraction unit 5 can also extract null data as the second extracted data related to the near-infrared light data.

[0151] In step S128, if the preprocessing signal related to the RGB data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the preprocessing signal related to the RGB data to calculate the value of ΔE*ab in the L*a*b* color space. The extraction unit 5 extracts the value of ΔE*ab as the extracted data related to the RGB data. If the postprocessing signal related to the RGB data satisfies the input reference corresponding to that signal, the extraction unit 5 uses the postprocessing signal related to the RGB data to calculate the value of ΔE*ab in the L*a*b* color space. The extraction unit 5 extracts the value of ΔE*ab as the extracted data related to the RGB data. If the postprocessing signal related to the RGB data does not satisfy the input reference corresponding to that signal, the extraction unit 5 removes the postprocessing signal related to the RGB data. For example, the extraction unit 5 can remove the postprocessing signal related to the RGB data by replacing it with null data. The extraction unit 5 can also extract null data as the extracted data related to the RGB data.

[0152] In step S129, if the preprocessing signal related to the potential data satisfies the input reference corresponding to that signal, the extraction unit 5 extracts the preprocessing signal related to the potential data as extracted data related to the potential data. If the postprocessing signal related to the potential data satisfies the input reference corresponding to that signal, the extraction unit 5 extracts the postprocessing signal related to the potential data as extracted data related to the potential data. If the postprocessing signal related to the potential data does not satisfy the input reference corresponding to that signal, the extraction unit 5 removes the postprocessing signal related to the potential data. For example, the extraction unit 5 can remove the postprocessing signal related to the potential data by replacing it with null data. The extraction unit 5 can also extract null data as extracted data related to the potential data.

[0153] In step S130, the database 40 stores each extracted data. The extracted data stored in the database 40 can be used for data analysis. The extracted data can also be used by a data analysis system 55 or the like for machine learning. For example, when the extracted data is used as learning data in machine learning, the efficiency related to building the prediction model will improve. The anomaly detection unit 9 can also determine an anomaly if there is an input signal, preprocessing signal, or postprocessing signal that deviates from the prediction model.

[0154] [Hardware Configuration] Figure 12 This is a diagram illustrating an example of the hardware configuration associated with data extraction system 1. Figure 12 A computer 100 is shown that functions as a data extraction system 1. The computer 100 includes a processor 101, main storage 102, auxiliary storage 103, communication control unit 104, input device 105, and output device 106. The data extraction system 1 is composed of one or more computers 100, wherein the one or more computers 100 are composed of this hardware and software such as programs.

[0155] In the case where the data extraction system 1 consists of multiple computers 100, these computers 100 can be connected locally or via a communication network 59 such as the Internet or intranet.

[0156] Processor 101 is a CPU (Central Processing Unit) that executes the operating system, applications, etc. Main storage unit 102 consists of ROM (Read Only Memory) and RAM (Random Access Memory). Auxiliary storage unit 103 is a storage medium consisting of a hard disk or flash memory. Generally, auxiliary storage unit 103 stores a larger amount of data than main storage unit 102. Communication control unit 104 consists of a network interface card (NIC) or a wireless communication module. At least a portion of the communication functions with other devices in data retrieval system 1 can be implemented through communication control unit 104. Input device 105 consists of a keyboard, mouse, touch panel, and microphone for voice input. Output device 106 consists of a display and printer.

[0157] The auxiliary storage unit 103 pre-stores the program 110 (data extraction program) and the data required for processing. The program 110 causes the computer 100 to execute the various functional elements of the data extraction system 1. Through the program 110, for example, the processing involved in the data extraction method described above is executed in the computer 100. For example, the program 110 is read by the processor 101 or the main storage unit 102, and the program 110 causes at least one of the processor 101, the main storage unit 102, the auxiliary storage unit 103, the communication control unit 104, the input device 105, and the output device 106 to operate. For example, the program 110 reads and writes data from the main storage unit 102 and the auxiliary storage unit 103.

[0158] Program 110 may be provided, for example, on a tangible storage medium such as a CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disc Read Only Memory), or semiconductor memory. Program 110 may also be provided as a data signal via a communication network 59.

[0159] As described above, a data extraction system 1 according to one aspect of this disclosure includes: a setting unit 7 for setting an input benchmark representing the performance required for data analysis; an input unit 2 for acquiring an input signal representing the measurement result of a target substance in biological production; a preprocessing unit 3 for performing preprocessing operations on the input signal for comparison with the input benchmark to generate a preprocessed signal; a postprocessing unit 4 for performing postprocessing operations on the preprocessed signal using other signals representing the measurement result of the target substance if the preprocessed signal does not meet the input benchmark to generate a postprocessed signal; and an extraction unit 5 for extracting the preprocessed signal as extracted data if the preprocessed signal meets the input benchmark, and extracting the postprocessed signal as extracted data if the postprocessed signal meets the input benchmark.

[0160] One aspect of this disclosure is a data extraction method performed by a data extraction system, wherein the data extraction system includes at least one processor. The data extraction method comprises the following steps: setting an input benchmark representing the performance required for data analysis; acquiring an input signal representing the measurement result of a analyte in biological production; performing a preprocessing operation on the input signal for comparison with the input benchmark to generate a preprocessed signal; if the preprocessed signal does not meet the input benchmark, performing a postprocessing operation on the preprocessed signal using other signals representing the measurement result of the analyte to generate a postprocessed signal; and if the preprocessed signal meets the input benchmark, extracting the preprocessed signal as extracted data, and if the postprocessed signal meets the input benchmark, extracting the postprocessed signal as extracted data.

[0161] One aspect of this disclosure is a data extraction procedure that enables a computer to perform the following steps: setting an input benchmark representing the performance required for data analysis; acquiring an input signal representing the measurement result of a analyte in biological production; performing a preprocessing operation on the input signal for comparison with the input benchmark to generate a preprocessed signal; performing a postprocessing operation on the preprocessed signal using other signals representing the measurement result of the analyte if the preprocessed signal does not meet the input benchmark to generate a postprocessed signal; and extracting the preprocessed signal as extracted data if the preprocessed signal meets the input benchmark, and extracting the postprocessed signal as extracted data if the postprocessed signal meets the input benchmark.

[0162] In one aspect of the data extraction system, method, and procedure disclosed herein, a preprocessing operation is performed on the input signal to generate a preprocessed signal. Then, the preprocessed signal is compared with an input benchmark representing the performance required for data analysis. If the preprocessed signal meets the input benchmark, the preprocessed signal is extracted as extracted data. Conversely, if the preprocessed signal does not meet the input benchmark, a postprocessing operation is performed on the preprocessed signal to generate a postprocessed signal. Then, the postprocessed signal is compared with the input benchmark. If the postprocessed signal meets the input benchmark, the postprocessed signal is extracted as extracted data. Thus, extracted data with ensured quality can be extracted. As a result, the quality of the data used for analysis can be improved.

[0163] The input signal is a spectrum measured by spectral analysis in bioprocessing. The preprocessing unit 3 generates a preprocessing signal by calculating the relative intensity of the signal level in a wavelength region where the transmittance is not zero and does not respond to changes in the culture medium. In conventional bioprocessing, the spectral signal is differentiated twice in the wavelength direction to suppress variation noise while capturing changes in the culture medium, etc. However, in this method, due to the double differentiation, it is difficult to directly capture wavelength changes. In contrast, in the data extraction system of this disclosure, the relative intensity of the signal level in a wavelength region where the transmittance is not zero and does not respond to changes in the culture medium is calculated. By using a preprocessing signal representing the relative intensity, variations in signal levels caused by hardware settings, etc., are eliminated. This improves the detection accuracy of target substances with low sensitivity. As a result, the quality of the data used for analysis is improved.

[0164] The spectral measurement is near-infrared spectroscopy. The pretreatment unit 3 generates a pretreatment signal by calculating the relative intensity of the signal level in a wavelength region that is unresponsive to changes in organic molecules and where the light intensity required for near-infrared spectroscopy is not affected by water absorption. In this case, for example, it is possible to detect target substances with low sensitivity, such as glucose concentration or lactic acid concentration.

[0165] The preprocessing unit 3 generates a first preprocessed signal with calculated relative intensity and a second preprocessed signal without calculated relative intensity. The setting unit 7 sets a first input reference as an input reference corresponding to the first preprocessed signal and a second input reference as an input reference corresponding to the second preprocessed signal. If the first preprocessed signal does not satisfy the first input reference, the postprocessing unit 4 performs postprocessing operations on the first preprocessed signal to generate a first postprocessed signal. If the first preprocessed signal satisfies the first input reference, the extraction unit 5 extracts the first preprocessed signal as first extracted data; if the first postprocessed signal satisfies the first input reference, the extraction unit 5 extracts the first postprocessed signal as first extracted data. If the second preprocessed signal does not satisfy the second input reference, the postprocessing unit 4 performs postprocessing operations on the second preprocessed signal to generate a second postprocessed signal. If the second preprocessed signal satisfies the second input reference, the extraction unit 5 extracts the second preprocessed signal as second extracted data; if the second postprocessed signal satisfies the second input reference, the extraction unit 5 extracts the second postprocessed signal as second extracted data. In this case, extracted data is obtained based on a first preprocessing signal without calculated relative intensity and second extracted data is obtained based on a second preprocessing signal with calculated relative intensity. That is, data obtained by performing different preprocessing operations on the same signal is extracted. The first and second extracted data can be used depending on the purpose of data analysis, even when the desired substance to be captured varies with the culture medium. This improves the convenience of data analysis.

[0166] The input benchmark is the signal-to-noise ratio (SNR). By using SNR as the input benchmark, unwanted noise components are removed. As a result, the efficiency and accuracy of data analysis are improved.

[0167] As part of the post-processing operation, the post-processing unit 4 performs summation averaging or moving average averaging. In this case, the necessary accuracy of the post-processed signal is improved.

[0168] The setting unit 7 also sets other input references different from the input reference. If the preprocessed signal does not meet these other input references, the post-processing unit 4 performs post-processing operations on the preprocessed signal to further generate other post-processed signals. If the preprocessed signal meets these other input references, the extraction unit 5 extracts other preprocessed signals as additional extracted data; if other post-processed signals meet these other input references, the extraction unit 5 also extracts other post-processed signals as additional extracted data. In this case, both extracted data based on the input reference and additional extracted data based on other input references are extracted. That is, by setting different input references for the same signal, data with different properties are extracted.

[0169] The system includes an anomaly detection unit 9, which determines anomalies based on input signals, preprocessing signals, or postprocessing signals. If an anomaly is detected, the anomaly detection unit 9 invalidates the extracted data. This ensures the extraction of high-quality data. Consequently, the quality of the data used for analysis is improved.

[0170] The anomaly detection unit 9 detects anomalies in the culture medium based on a decrease in dissolved oxygen levels. In this case, it can sense with high precision whether the culture medium has solidified.

[0171] The anomaly detection unit 9 detects anomalies in electrode potential based on the correlation between dissolved oxygen levels or pH and electrode potential. In this case, it can detect anomalies such as a change in the culture medium towards alkalinity with high precision.

[0172] One aspect of this disclosure is a data extraction system that acquires a spectrum measured by spectrometry in biological production and outputs a signal obtained by calculating the relative intensity of the spectrum to the signal level of the spectrum in a wavelength region that does not respond to changes in the culture medium and has a non-zero transmittance.

[0173] In one aspect of the data extraction system disclosed herein, the relative intensity of the signal level in a wavelength region that is not responsive to changes in the culture medium and has non-zero transmittance is calculated. By representing the signal with relative intensity, variations in signal levels caused by hardware settings, etc., are eliminated. This improves the detection accuracy of target substances with low sensitivity. As a result, the quality of the data used for analysis is improved.

[0174] [Variation Example] This disclosure is not necessarily limited to the above-described embodiments, and various modifications may be made without departing from its spirit.

[0175] In bioproduction, evaluation metrics can be constant or variable. For example, evaluation metrics can vary to match the substance being evaluated throughout the cultivation process. In one example, the evaluation metric might correspond to substance A in the early stages of cultivation, to substance B in the middle stages, and to substance C in the later stages. In this case, the data extraction system extracts the relevant data for each stage.

[0176] [Postscript] The following outlines the main points of this disclosure.

[0177] [1] A data extraction system comprising: a setting unit for setting an input benchmark representing the performance required for data analysis; an input unit for acquiring an input signal representing the measurement result of a target substance in biological production; a preprocessing unit for performing a preprocessing operation on the input signal for comparison with the input benchmark to generate a preprocessed signal; a postprocessing unit for performing a postprocessing operation on the preprocessed signal using other signals representing the measurement result of the target substance when the preprocessed signal does not satisfy the input benchmark to generate a postprocessed signal; and an extraction unit for extracting the preprocessed signal as extracted data when the preprocessed signal satisfies the input benchmark, and extracting the postprocessed signal as extracted data when the postprocessed signal satisfies the input benchmark.

[0178] [2] According to the data extraction system of [1], wherein the input signal is a spectrum measured by spectral measurement in biological production, and the pretreatment unit generates the pretreatment signal by calculating the relative intensity of the signal level in a wavelength region that does not respond to changes in the culture medium and has a non-zero transmittance.

[0179] [3] According to the data extraction system described in [2], wherein the spectral measurement is a near-infrared spectral measurement, the preprocessing unit generates the preprocessing signal by calculating the relative intensity of the signal level in a wavelength region that is not responsive to changes in organic molecules and where the light intensity required for the near-infrared spectral measurement is not affected by water absorption.

[0180] [4] According to the data extraction system described in [3], the preprocessing unit generates a first preprocessing signal without calculating the relative intensity and generates a second preprocessing signal with calculated relative intensity. The setting unit sets a first input reference as an input reference corresponding to the first preprocessing signal and sets a second input reference as an input reference corresponding to the second preprocessing signal. If the first preprocessing signal does not satisfy the first input reference, the postprocessing unit performs the postprocessing operation on the first preprocessing signal to generate a first postprocessing signal. If the first preprocessing signal satisfies the first input reference, the extraction unit extracts the first preprocessing signal as first extracted data. If the first postprocessing signal satisfies the first input reference, the extraction unit extracts the first postprocessing signal as first extracted data. If the second preprocessing signal does not satisfy the second input reference, the postprocessing unit performs the postprocessing operation on the second preprocessing signal to generate a second postprocessing signal. If the second preprocessing signal satisfies the second input reference, the extraction unit extracts the second preprocessing signal as second extracted data. If the second postprocessing signal satisfies the second input reference, the extraction unit extracts the second postprocessing signal as second extracted data.

[0181] [5] The data extraction system according to any one of [1] to [4], wherein the input reference is the signal-to-noise ratio.

[0182] [6] The data extraction system according to any one of [1] to [5], wherein, as the post-processing operation, the post-processing unit performs an additive average or a moving average.

[0183] [7] According to the data extraction system described in [1] to [6], the setting unit further sets other input benchmarks that are different from the input benchmarks. When the preprocessing signal does not meet the other input benchmarks, the postprocessing unit performs the postprocessing operation on the preprocessing signal to further generate other postprocessing signals. When the preprocessing signal meets the other input benchmarks, the extraction unit further extracts the other preprocessing signals as other extracted data. When the other postprocessing signals meet the other input benchmarks, the extraction unit further extracts the other postprocessing signals as other extracted data.

[0184] [8] The data extraction system according to [1] to [7], wherein the data extraction system includes an anomaly determination unit, wherein the anomaly determination unit determines an anomaly based on the input signal, the preprocessing signal or the postprocessing signal, and in the case of sensing an anomaly, the anomaly determination unit invalidates the extracted data.

[0185] [9] According to the data extraction system described in [8], the anomaly determination unit detects anomalies in the culture medium based on a decrease in dissolved oxygen.

[0186]

[10] According to the data extraction system described in [8], the anomaly determination unit detects anomalies in electrode potential based on the correlation between dissolved oxygen or pH and electrode potential.

[0187]

[11] A data extraction method, executed by a data extraction system, wherein the data extraction system includes at least one processor, wherein the data extraction method comprises the following steps: setting an input benchmark representing the performance required for data analysis; acquiring an input signal representing the measurement result of a analyte in biological production; performing a preprocessing operation on the input signal for comparison with the input benchmark to generate a preprocessed signal; if the preprocessed signal does not satisfy the input benchmark, performing a postprocessing operation on the preprocessed signal using other signals representing the measurement result of the analyte to generate a postprocessed signal; and if the preprocessed signal satisfies the input benchmark, extracting the preprocessed signal as extracted data, and if the postprocessed signal satisfies the input benchmark, extracting the postprocessed signal as extracted data.

[0188]

[12] A data extraction program that causes a computer to perform the following steps: setting an input benchmark representing the performance required for data analysis; acquiring an input signal representing the measurement result of a measured object in biological production; performing a preprocessing operation on the input signal for comparison with the input benchmark to generate a preprocessing signal; if the preprocessing signal does not meet the input benchmark, performing a postprocessing operation on the preprocessing signal using other signals representing the measurement result of the measured object to generate a postprocessing signal; and if the preprocessing signal meets the input benchmark, extracting the preprocessing signal as extracted data, and if the postprocessing signal meets the input benchmark, extracting the postprocessing signal as extracted data.

[0189]

[13] A data extraction system, wherein the data extraction system acquires a spectrum measured by spectrometry in bioproduction, and the data extraction system outputs a signal obtained by calculating the relative intensity of the signal level of the spectrum with respect to the spectrum in a wavelength region where the transmittance is not zero and does not respond to changes in the culture medium.

[0190] Explanation of reference numerals in the attached figures 1: Data extraction system; 2: Input unit; 3: Preprocessing unit; 4: Postprocessing unit; 5: Extraction unit; 6: Control unit; 7: Setting unit; 8: Comparison unit; 9: Anomaly detection unit; 10: Control block; 20: Culture tank; 21: Light source; 22: Flow cell; 23: Spectrometer; 24: Chart; 30: Sensor; 40: Database; 50: Biological production system; 51: Culture control device; 52: Sensor control device; 53: Control terminal; 54: Cloud system; 55: Data analysis system; 56: Dimension reduction unit; 57: Explanatory variable extraction unit; 58: Machine learning unit; 59: Communication network; 100: Computer; 101: Processor; 102: Main storage unit; 103: Auxiliary storage unit; 104: Communication control unit; 105: Input device; 106: Output device; 110: Program.

Claims

1. A data extraction system, comprising: The setting section sets the input benchmarks that represent the performance requirements for data analysis. The input unit acquires input signals representing the measurement results of the objects being measured in biological production; The preprocessing unit performs preprocessing operations on the input signal for comparison with the input reference, and generates a preprocessed signal; The post-processing unit performs post-processing operations on the pre-processing signal using other signals representing the measurement result of the object being measured, when the pre-processing signal does not meet the input reference, and generates a post-processing signal. as well as The extraction unit extracts the preprocessing signal as extracted data when the preprocessing signal meets the input reference, and extracts the postprocessing signal as extracted data when the postprocessing signal meets the input reference.

2. The data extraction system according to claim 1, wherein, The input signal is a spectrum measured through spectral determination in biological production. The pretreatment unit generates the pretreatment signal by calculating the relative intensity of the signal level in a wavelength region that does not respond to changes in the culture medium and has a non-zero transmittance.

3. The data extraction system according to claim 2, wherein, The spectral measurement is a near-infrared spectroscopy measurement. The preprocessing unit generates the preprocessing signal by calculating the relative intensity of the signal level in a wavelength region that is not responsive to changes in organic molecules and whose light intensity required for the near-infrared spectroscopy measurement is not affected by water absorption.

4. The data extraction system according to claim 3, wherein, The preprocessing unit generates a first preprocessing signal without calculating the relative intensity, and generates a second preprocessing signal with the relative intensity calculated. The setting unit sets a first input reference as the input reference corresponding to the first preprocessing signal, and sets a second input reference as the input reference corresponding to the second preprocessing signal. If the first preprocessing signal does not meet the first input reference, the postprocessing unit performs the postprocessing operation on the first preprocessing signal to generate a first postprocessing signal. If the first preprocessing signal satisfies the first input benchmark, the extraction unit extracts the first preprocessing signal as first extracted data; if the first postprocessing signal satisfies the first input benchmark, the extraction unit extracts the first postprocessing signal as first extracted data. If the second preprocessing signal does not meet the second input reference, the postprocessing unit performs the postprocessing operation on the second preprocessing signal to generate a second postprocessing signal. When the second preprocessing signal satisfies the second input reference, the extraction unit extracts the second preprocessing signal as the second extracted data. When the second postprocessing signal satisfies the second input reference, the extraction unit extracts the second postprocessing signal as the second extracted data.

5. The data extraction system according to claim 4, wherein, The input reference is the signal-to-noise ratio.

6. The data extraction system according to claim 5, wherein, As part of the post-processing operation, the post-processing unit performs summation averaging or moving average.

7. The data extraction system according to claim 1, wherein, The setting unit also sets other input references that are different from the input reference. If the preprocessed signal does not meet the other input criteria, the postprocessing unit performs the postprocessing operation on the preprocessed signal to further generate other postprocessed signals. If the preprocessing signal satisfies the other input criteria, the extraction unit also extracts the other preprocessing signal as other extracted data. If the other postprocessing signal satisfies the other input criteria, the extraction unit also extracts the other postprocessing signal as other extracted data.

8. The data extraction system according to claim 1, wherein, The data extraction system includes an anomaly determination unit, which determines anomalies based on the input signal, the preprocessing signal, or the postprocessing signal. If an anomaly is detected, the anomaly determination unit invalidates the extracted data.

9. The data extraction system according to claim 8, wherein, The anomaly detection unit detects anomalies in the culture medium based on a decrease in dissolved oxygen levels.

10. The data extraction system according to claim 8, wherein, The anomaly detection unit detects anomalies in electrode potential based on the correlation between dissolved oxygen or pH and electrode potential.

11. A data extraction method, executed by a data extraction system, wherein, The data extraction system includes at least one processor, and the data extraction method includes the following steps: Set an input benchmark that represents the performance requirements for data analysis; Acquire input signals representing the measurement results of the objects to be measured in biological production; The input signal is subjected to preprocessing operations for comparison with the input reference to generate a preprocessed signal; If the preprocessing signal does not meet the input reference, postprocessing operations are performed on the preprocessing signal using other signals representing the measurement result of the object being measured, to generate a postprocessing signal; and If the preprocessing signal satisfies the input benchmark, the preprocessing signal is extracted as extracted data; if the postprocessing signal satisfies the input benchmark, the postprocessing signal is extracted as extracted data.

12. A data extraction program that causes a computer to perform the following steps: Set an input benchmark representing the performance requirements for data analysis; Acquire input signals representing the measurement results of the objects to be measured in biological production; The input signal is subjected to preprocessing operations for comparison with the input reference to generate a preprocessed signal; If the preprocessing signal does not meet the input reference, the preprocessing signal is subjected to postprocessing operations using other signals representing the measurement result of the object being measured, to generate a postprocessing signal; as well as If the preprocessing signal satisfies the input benchmark, the preprocessing signal is extracted as extracted data; if the postprocessing signal satisfies the input benchmark, the postprocessing signal is extracted as extracted data.

13. A data extraction system, wherein, The data extraction system acquires spectra measured through spectral determination in biological production. The data extraction system outputs a signal obtained by calculating the relative intensity of the signal level of the spectrum with respect to the wavelength region where the transmittance is not zero and does not respond to changes in the culture medium.

Images