Monitoring method and device for double-measuring-point near infrared spectrum signal separation and storage medium
By merging and processing near-infrared spectral signals on an industrial production line and utilizing constrained independent component analysis (CICA) methods, the problems of equipment complexity and data consistency in multi-point monitoring were solved, enabling high-frequency continuous monitoring and accurate prediction.
Patent Information
- Application Number
- CN202510761674.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-09-23
AI Technical Summary
Existing technologies for multi-point near-infrared spectroscopy monitoring have problems such as high equipment complexity, high maintenance costs, measurement timing delays, and poor data consistency, making them particularly unsuitable for high-frequency continuous monitoring.
A dual-point near-infrared spectral signal separation method was adopted. By setting two measuring points on the industrial production line, the near-infrared spectral signals were merged and smoothed. The constrained independent component analysis method was used for mathematical separation, and a quantitative analysis model for the feed end and the discharge end was constructed.
It realizes simultaneous in-situ monitoring of multiple measuring points, reduces equipment complexity and maintenance costs, improves data consistency and modeling accuracy, and is suitable for high-frequency continuous monitoring.
Smart Images
Figure CN120685594A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to near-infrared spectrum intelligent analysis technology, and in particular to a monitoring method, device and storage medium for separating double-measurement-point near-infrared spectrum signals. Background Art
[0002] Near-infrared spectroscopy is widely used in process monitoring due to its rapid and non-destructive nature. However, due to the high cost of near-infrared spectrometers, in industrial applications where multiple measurement points need to be measured simultaneously, using a single spectrometer to monitor multiple points simultaneously is the best solution to reduce equipment investment costs. Traditional methods typically rely on fiber optic multiplexers, which collect spectral data from different measurement points through mechanical switching devices or controlled timing. This method relies on movable parts (such as mechanical switching devices), which not only increases system complexity and maintenance costs, but also may cause measurement delays, making it unsuitable for high-frequency continuous monitoring.
[0003] At the same time, due to the loss and deviation during the optical path switching process, the data consistency between different measurement points may be reduced, thus affecting the modeling accuracy and long-term stability. In many monitoring processes, the spectral signals at different measurement points (such as the feed end and the discharge end) may have significant differences. Summary of the Invention
[0004] In view of the shortcomings of the existing technology, the present invention proposes a monitoring method for separating dual-point near-infrared spectral signals, comprising:
[0005] Two measuring points are set up on the industrial production line to collect near-infrared spectral signals at the feeding end and the discharging end respectively;
[0006] The two near-infrared spectral signals of the two measuring points are merged into one to form a mixed spectral signal, and the mixed spectral signal is guided to the slit entrance of the near-infrared spectrometer, and the collected mixed spectral signal is smoothed;
[0007] Using a constrained independent component analysis method to mathematically separate the mixed spectral signal, and extracting the spectral features of the feed end and the discharge end;
[0008] Combining the spectral characteristics of the feed end and the discharge end, quantitative analysis models of the feed end and the discharge end are constructed respectively, and the composition is predicted through the real-time collected spectral data.
[0009] In some embodiments, the near-infrared spectrum signals of the two measuring points are guided to a spectroscope via an optical fiber, and the spectroscope combines the two near-infrared spectrum signals into one to form a mixed spectrum signal.
[0010] In some embodiments, the two near-infrared spectral signals are symmetrically arranged at an angle of 45° to the spectroscope.
[0011] In some embodiments, the constrained independent component analysis method includes the following steps:
[0012] Assume that the near-infrared spectrum signal X consists of the feed end spectrum X1 of measuring point 1 and the discharge end spectrum X2 of measuring point 2, that is:
[0013] X(λ,t)=a(t)S1(λ)+b(t)S2(λ)+∈(λ)
[0014] Where: S1 and S2 correspond to the standard spectra of the materials at the feed end and the discharge end respectively; a(t) and b(t) are the time-varying mixing coefficients respectively, and ∈ is the noise term. First, SG or smoothing filtering is used to remove the noise term ∈;
[0015] An arbitrary standard spectrum R(λ) at the feed end is introduced as a constraint to establish the objective function, J(w) = non-Gaussianity measure - γ·constraint term, where γ is the constraint weight, ranging from 0 to 1;
[0016] The objective function is solved by Newton iteration method to separate the spectral signals of the feed end and the discharge end.
[0017] In some embodiments, an arbitrary standard spectrum R(λ) at the feed end is introduced as a constraint to establish the objective function as follows:
[0018] maxJ(w)=E{G(w T X)}-ρ·Corr(w T X,R)
[0019] Where G is a nonlinear function that measures non-Gaussianity, and ρ is a constraint weight that ranges from 0.1 to 0.9. In some embodiments, the Newton iteration method is used to solve:
[0020]
[0021] Iterate n times and separate in,
[0022] In some embodiments, the constrained independent component analysis method further comprises the following steps:
[0023] The separated signals are verified for correlation using Euclidean distance or similarity coefficient, with the similarity coefficient being preferred:
[0024]
[0025] The value of δ is generally greater than 0.95. Depending on the signal quality, both signals can be selected to meet the verification requirements, or any one of them can be selected.
[0026] If the verification condition is still not met within n times, the adaptive mixing matrix update is triggered:
[0027] [a,b] T =(S T S) -1 S T X
[0028] Where S = [S1, S2] is the calibration spectrum matrix composed of the standard spectra of the inlet and outlet.
[0029] In some embodiments, the method for constructing the quantitative analysis model includes multiple linear regression, principal component regression, partial least squares regression, support vector regression or deep learning method.
[0030] The present invention also provides a monitoring device for separating dual-point near-infrared spectral signals, comprising:
[0031] The acquisition module sets two measurement points on the industrial production line to collect near-infrared spectral signals at the feed end and the discharge end respectively;
[0032] a preprocessing module, which combines the two near-infrared spectral signals of the two measuring points into one to form a mixed spectral signal, guides the mixed spectral signal to the slit entrance of the near-infrared spectrometer, and performs smoothing on the collected mixed spectral signal;
[0033] A feature extraction module, which uses a constrained independent component analysis method to mathematically separate the mixed spectral signal and extract the spectral features of the feed end and the discharge end;
[0034] The prediction module combines the spectral characteristics of the feed end and the discharge end to construct quantitative analysis models for the feed end and the discharge end respectively, and predicts the composition through the spectral data collected in real time.
[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.
[0036] From the above scheme, it can be seen that the advantages of the present invention are:
[0037] The monitoring method based on dual-point near-infrared spectral signal separation provided by the present invention is suitable for simultaneous in-situ monitoring of signals at multiple measuring points with different selections during the production process. The technical solution provided by the present invention optimizes the optical fiber design so that the optical signals of multiple measuring points enter the spectrometer at the same time, and then uses mathematical methods to achieve signal separation, thereby establishing near-infrared prediction models for different measuring points. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 As shown in the embodiment of the present invention Figure 1, a schematic diagram of the overall process of the monitoring method for dual-point near-infrared spectral signal separation shown in an embodiment of the present invention;
[0039] Figure 2 Schematic diagram of a light mixing device according to an embodiment of the present invention;
[0040] Figure 3 This is a flowchart of the steps of the constrained independent component analysis method shown in an embodiment of the present invention;
[0041] Figure 4 This is a diagram of a PLS regression model established by using SG derivatives and MSC for spectral preprocessing of straw and CARS for feature engineering as shown in the embodiment of the present invention;
[0042] Figure 5 For the straw shown in the embodiment of the present invention, SG derivative and SNV were used for spectral preprocessing, CARS was used for feature engineering, and a PLS regression model model diagram was established;
[0043] Figure 6 Schematic diagram of the overall structure of a monitoring device for dual-point near-infrared spectral signal separation according to an embodiment of the present invention;
[0044] Wherein, the reference numerals:
[0045] 300-Dual-point near-infrared spectroscopy signal separation monitoring device;
[0046] 310-acquisition module;
[0047] 320-preprocessing module;
[0048] 330-feature extraction module;
[0049] 340-Prediction module;
[0050] S1-S4: steps. DETAILED DESCRIPTION
[0051] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.
[0052] See also Figure 1 An embodiment of the present invention provides a monitoring method for separating near-infrared spectral signals at two measurement points, comprising:
[0053] S1. Set up two measuring points on the industrial production line to collect near-infrared spectral signals at the feed end and the discharge end respectively;
[0054] S2. Combining the two near-infrared spectral signals of the two measuring points into one to form a mixed spectral signal, and guiding the mixed spectral signal to the slit entrance of the near-infrared spectrometer, and smoothing the collected mixed spectral signal; specifically, guiding the near-infrared spectral signals of the two measuring points to a spectroscope through an optical fiber, and the spectroscope combining the two near-infrared spectral signals into one to form a mixed spectral signal.
[0055] S3. Using a constrained independent component analysis method to mathematically separate the mixed spectral signal and extract the spectral features of the feed end and the discharge end;
[0056] S4. Combining the spectral characteristics of the feed end and the discharge end, quantitative analysis models of the feed end and the discharge end are constructed respectively, and the composition is predicted through the real-time collected spectral data.
[0057] In this embodiment, a dual-probe structure is used, which is arranged at the feed end and the discharge end respectively. The near-infrared light signal of each measuring point is collected through optical fiber. In order to reduce the volume of the equipment and optimize the way the light signal enters the spectrometer, see Figure 2 This is an optical mixing device. Specifically, a beamsplitter is used to combine two optical signals into one, which is then guided to the spectrometer's slit entrance via a spatial optical path or optical fiber. In this embodiment, the beamsplitter is a non-polarizing beamsplitter cube with a specific splitting ratio, operating in the near-infrared region of 900-2500nm. The two near-infrared spectral signals are arranged symmetrically at a 45° angle with the beamsplitter to optimize optical signal combining efficiency.
[0058] Furthermore, since the spectral signals from different measuring points are mixed and then enter the same spectrometer, the directly measured spectral data is a mixed optical signal from the feed end and the discharge end; therefore, the present invention first smoothes the spectral signal in the data processing stage to reduce the impact of noise on subsequent analysis, and then uses the constrained independent component analysis (cICA) method to mathematically separate the spectral signal. cICA is based on the principle of independent component analysis (ICA) and combines known prior information to perform signal unmixing, which can effectively improve the separation accuracy of spectral signals from different measuring points and ensure the accurate extraction of spectral data from each measuring point. For details, see Figure 3 , the constrained independent component analysis method comprises the following steps:
[0059] Assume that the near-infrared spectrum signal X consists of the feed end spectrum X1 of measuring point 1 and the discharge end spectrum X2 of measuring point 2, that is:
[0060] X(λ,t)=a(t)S1(λ)+b(t)S2(λ)+∈(λ)
[0061] Where: S1 and S2 correspond to the standard spectra of the materials at the feed end and the discharge end respectively; a(t) and b(t) are the time-varying mixing coefficients respectively, and ∈ is the noise term. First, SG or smoothing filtering is used to remove the noise term ∈;
[0062] An arbitrary standard spectrum R(λ) at the feed end is introduced as a constraint to establish the objective function, J(w) = non-Gaussianity measure - γ·constraint term, where γ is the constraint weight, ranging from 0 to 1;
[0063] The objective function is solved by Newton iteration method to separate the spectral signals of the feed end and the discharge end.
[0064] Among them, an arbitrary standard spectrum R(λ) at the feed end is introduced as a constraint to establish the objective function as follows:
[0065] maxJ(w)=E{G(w T X)}-ρ·Corr(w T X,R)
[0066] Among them, G is a nonlinear function that measures non-Gaussianity, and ρ is the constraint weight, which ranges from 0.1 to 0.9.
[0067] Among them, the solution is obtained by Newton's iteration method:
[0068]
[0069] Iterate n times and separate in,
[0070] Furthermore, the separated signals are verified for correlation using Euclidean distance or similarity coefficient, where the similarity coefficient is preferred:
[0071]
[0072] The value of δ is generally greater than 0.95. Depending on the signal quality, both signals can be selected to meet the verification requirements, or any one of them can be selected.
[0073] If the verification condition is still not met within n times, the adaptive mixing matrix update is triggered:
[0074] [a,b] T =(S T S) -1 S T X
[0075] Where S = [S1, S2] is the calibration spectrum matrix composed of the standard spectra of the inlet and outlet.
[0076] The method for constructing the quantitative analysis model described in this embodiment includes multiple linear regression, principal component regression, partial least squares regression, support vector regression or deep learning method.
[0077] The standard model is established and predicted as follows:
[0078] First, the absorbance is calculated:
[0079] The blank spectrum collected at any measuring point is used as the reference I0, and the absorbance is calculated according to the Lambert-Beer law:
[0080] A=-log 10 (I / I0)
[0081] The reference can be fixed or a blank spectrum collected at regular intervals. The blank spectrum can be collected by the light source and detector facing each other, or the spectrum of a standard diffuse reflection white board (which can be a ceramic board, a polytetrafluoroethylene white board, a gold-plated board, etc.) can be collected.
[0082] Furthermore, a quantitative analysis model is constructed:
[0083] Using the separated spectra and the reference values of the corresponding materials (usually property data obtained by standard laboratory methods), a quantitative analysis model is established through spectral preprocessing, feature engineering and regression methods (multivariate linear regression, principal component regression, partial least squares regression, support vector regression, univariate linear regression, etc.). The quantitative analysis model can also be established end-to-end through deep learning.
[0084] Finally, make a prediction:
[0085] The material spectrum of unknown property values is collected, and after signal separation and absorbance calculation, the same spectrum preprocessing and feature engineering are carried out and brought into the quantitative analysis model to obtain the required quantitative analysis data.
[0086] As described above, the monitoring method based on dual-point near-infrared spectral signal separation provided in this embodiment is suitable for scenarios with a number of measuring points ≥ 2. The merging and separation of multiple optical signals are achieved through a cascade of spectroscopes or a multi-channel spectroscopic device. The spectrum type can be a near-infrared spectrum, a Raman spectrum, or an ultraviolet spectrum. The reference spectrum can be a fixed reference or a blank spectrum collected regularly. The blank spectrum collection method includes light source and detector cross-reflection collection or standard diffuse reflection whiteboard collection. The standard diffuse reflection whiteboard is a ceramic plate, a polytetrafluoroethylene whiteboard, or a gold-plated plate. The optical fiber core diameter is 200-600 μm, and the numerical aperture is 0.22-0.39 to ensure the optical signal coupling efficiency.
[0087] The following is a detailed description of the technical solution of the present invention using the online monitoring of the biochar production process as an example:
[0088] (1) Monitoring content: fixed carbon of straw and calorific value of biochar
[0089] (2) Experimental setup
[0090] Two measurement points were set up on the biochar production line: one on the feeder conveyor belt at the inlet and one on the discharge pipe. The signals from the two probes were transmitted to an optical signal mixer via optical fiber with a core diameter of 400 μm. The beam splitter used in the mixer had a 30:70 split ratio, with the inlet measurement point at 30 and the discharge measurement point at 70. A near-infrared spectrometer was used to collect the combined signal. The mixer and spectrometer were connected via a 600 μm core diameter optical fiber. A reference plate was installed at the feeder measurement point and swung to the lower end of the measurement port by a motor.
[0091] (3) Data Collection
[0092] The spectrum is collected every 5 minutes, and the reference signal is collected every 8 hours. In the modeling stage, the time point of the sampling moment is recorded, and the three spectra before and after are captured to establish the correction model.
[0093] (4) Signal processing
[0094] The original signal was smoothed using the SG method, with the window width set to 13, the fitting order set to 2, and the number of derivatives set to 0. The cICA method was used to separate the signals of the feed port and the discharge port.
[0095] (5) Model construction
[0096] Calculate the absorbance of the separated spectrum and the latest updated reference spectrum, see Figure 4 , in which straw was spectrally preprocessed using SG derivatives and MSC, feature engineering was performed using CARS, and a PLS regression model was established; see Figure 5 , straw was spectrally preprocessed using SG derivative and SNV, feature engineering was performed using CARS, and a PLS regression model was established.
[0097] (6) Online monitoring
[0098] The spectra were collected in real time and separated using cICA, absorbance was calculated, and spectra were preprocessed. The characteristic bands were selected and introduced into the above model to obtain the predicted values.
[0099] Further, see Figure 6 Another embodiment of the present invention provides a monitoring device 300 for separating near-infrared spectral signals at two measurement points, comprising:
[0100] The acquisition module 310 sets two measurement points on the industrial production line to respectively collect near-infrared spectrum signals at the feed end and the discharge end;
[0101] The preprocessing module 320 combines the two near-infrared spectral signals of the two measuring points into one mixed spectral signal, guides the mixed spectral signal to the slit entrance of the near-infrared spectrometer, and performs smoothing on the collected mixed spectral signal.
[0102] A feature extraction module 330 is configured to perform mathematical separation on the mixed spectral signal using a constrained independent component analysis method to extract spectral features of the feed end and the discharge end;
[0103] The prediction module 340 combines the spectral characteristics of the feed end and the discharge end to construct quantitative analysis models for the feed end and the discharge end respectively, and performs composition prediction based on the spectral data collected in real time.
[0104] The embodiment of this device can be implemented in conjunction with the implementation of the above method embodiment. The relevant technical details mentioned in the implementation of the above method embodiment are still valid in the implementation of this device embodiment, and will not be repeated here to reduce repetition.
[0105] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the online monitoring method based on dual-point near-infrared spectral signal separation in the above embodiment are implemented.
[0106] In addition, it should be understood that the storage medium in the embodiments of the apparatus of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).
[0107] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only.
[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A monitoring method for separating dual-point near-infrared spectroscopy signals, characterized in that: include: Two measuring points are set up on the industrial production line to collect near-infrared spectral signals at the feeding end and the discharging end respectively; The two near-infrared spectral signals of the two measuring points are merged into one to form a mixed spectral signal, and the mixed spectral signal is guided to the slit entrance of the near-infrared spectrometer, and the collected mixed spectral signal is smoothed; Using a constrained independent component analysis method to mathematically separate the mixed spectral signal, and extracting the spectral features of the feed end and the discharge end; Combining the spectral characteristics of the feed end and the discharge end, quantitative analysis models of the feed end and the discharge end are constructed respectively, and the composition is predicted through the real-time collected spectral data.
2. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 1, characterized in that: The near-infrared spectrum signals of the two measuring points are guided to a spectroscope through an optical fiber, and the spectroscope combines the two near-infrared spectrum signals into one to form a mixed spectrum signal.
3. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 2, characterized in that: The two near-infrared spectral signals are symmetrically arranged at an angle of 45° to the spectroscope.
4. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 1, characterized in that: The constrained independent component analysis method comprises the following steps: Assume that the near-infrared spectrum signal X consists of the feed end spectrum X1 of measuring point 1 and the discharge end spectrum X2 of measuring point 2, that is: X(λ,t)=a(t)S1(λ)+b(t)S2(λ)+∈(λ) Where: S1 and S2 correspond to the standard spectra of the materials at the feed end and the discharge end respectively; a(t) and b(t) are the time-varying mixing coefficients respectively, and ∈ is the noise term. First, SG or smoothing filtering is used to remove the noise term ∈; An arbitrary standard spectrum R(λ) at the feed end is introduced as a constraint to establish the objective function, J(w) = non-Gaussianity measure - γ·constraint term, where γ is the constraint weight, ranging from 0 to 1; The objective function is solved by Newton iteration method to separate the spectral signals of the feed end and the discharge end.
5. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 4, characterized in that: Introducing any standard spectrum R(λ) at the feed end as a constraint to establish the objective function is as follows: maxJ(w)=E{G(w T X)}-ρ·Corr(w T X,R) Among them, G is a nonlinear function that measures non-Gaussianity, and ρ is the constraint weight, which ranges from 0.1 to 0.
9.
6. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 4, characterized in that: in, Solve using Newton's method: Iterate n times and separate in, 7. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 6, characterized in that: The constrained independent component analysis method further comprises the following steps: The separated signals are verified for correlation using Euclidean distance or similarity coefficient, with the similarity coefficient being preferred: The value of δ is generally greater than 0.
95. Depending on the signal quality, both signals can be selected to meet the verification requirements, or any one of them can be selected. If the verification condition is still not met within n times, the adaptive mixing matrix update is triggered: [a,b] T =(S T S) -1 S T X Where S = [S1, S2] is the calibration spectrum matrix composed of the standard spectra of the inlet and outlet.
8. The monitoring method for dual-point near-infrared spectroscopy signal separation according to claim 1, characterized in that: The construction method of the quantitative analysis model includes multiple linear regression, principal component regression, partial least squares regression, support vector regression or deep learning method.
9. A monitoring device for dual-point near-infrared spectral signal separation, characterized in that: include: The acquisition module sets two measurement points on the industrial production line to collect near-infrared spectral signals at the feed end and the discharge end respectively; a preprocessing module, which combines the two near-infrared spectral signals of the two measuring points into one to form a mixed spectral signal, guides the mixed spectral signal to the slit entrance of the near-infrared spectrometer, and performs smoothing on the collected mixed spectral signal; A feature extraction module, which uses a constrained independent component analysis method to mathematically separate the mixed spectral signal and extract the spectral features of the feed end and the discharge end; The prediction module combines the spectral characteristics of the feed end and the discharge end to construct quantitative analysis models for the feed end and the discharge end respectively, and predicts the composition through the spectral data collected in real time.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.