Near infrared absorption detection spectroscopy system

Abstract

The application discloses a near-infrared absorption detection spectrum analysis system and belongs to the technical field of intelligent manufacturing detection equipment. The system comprises a dual-mode magnetic-optical collaborative sensing probe, a signal conditioning circuit, a magnetic field inversion noise dynamic decoupling and compensation module and a spectrum analysis module. The magnetic sensing element and the near-infrared spectrum detection element are heterogeneously integrated and coaxially and homologously arranged in the probe, are used for synchronously collecting three-dimensional magnetic field vector signals and near-infrared spectrum signals, the compensation module generates a magnetic noise compensation signal based on a pre-trained nonlinear mapping model and compensates the spectrum signals by using the magnetic field data, and the analysis module obtains the chemical component concentration. The system further comprises a multi-physical quantity online structure-activity relationship collaborative modeling module. The application solves the problems of insufficient spectrum detection anti-interference ability in a strong magnetic field environment and mutual separation of component detection and magnetic performance detection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing testing equipment technology, specifically relating to a near-infrared absorption detection spectroscopy analysis system. It is applicable to the online synchronous detection of the chemical composition and physical magnetic properties of materials in intelligent manufacturing production lines for magnetic materials such as new energy vehicle drive motors, wind power core components, rare earth permanent magnets, and soft magnetic alloys. In particular, it relates to a near-infrared spectroscopy online detection and multi-physical quantity fusion evaluation technology that can operate stably under strong DC or alternating magnetic field interference environments. Background Technology

[0002] Near-infrared absorption spectroscopy is widely used in the online analysis of the chemical composition and properties of materials in intelligent manufacturing processes due to its non-destructive, online, and rapid detection characteristics. Near-infrared absorption spectroscopy refers to the characteristic spectrum formed when electromagnetic waves with wavelengths between 780 nm and 2500 nm are absorbed by specific chemical bonds (such as carbon-hydrogen bonds, oxygen-hydrogen bonds, and nitrogen-hydrogen bonds) in the molecules of a material. It is used to characterize the chemical composition of materials. The basic principle of its detection is as follows: after incident light passes through the material to be tested, light of a specific wavelength is absorbed by the chemical bonds inside the material. By measuring the degree of attenuation of the transmitted light intensity relative to the incident light intensity, the concentration of the chemical composition of the material can be obtained by inversion according to the Lambert-Beer absorption law.

[0003] In intelligent manufacturing production lines for magnetic materials, the existing technical solution closest to online near-infrared spectroscopy detection consists of two independent detection systems. One is an online near-infrared spectroscopy detection device, which typically includes a light source module, a sample chamber, a micro-spectral detection chip, and a signal conditioning circuit, used for online spectral acquisition and analysis of the chemical composition of the material. The other is an independently deployed magnetic property detection device, which typically uses a magnetometer, fluxmeter, or Hall sensor array to measure the coercivity, remanence, permeability, and other magnetic property parameters of the material in an offline or random sampling manner. The two systems are usually located at different workstations and at different time points on the production line, and there is no data interaction or collaborative processing mechanism between them. Furthermore, the spectral detection system itself does not have the ability to sense and compensate for the external magnetic field environment.

[0004] The existing technologies have the following drawbacks: First, in strong magnetic fields or complex electromagnetic environments, the magnetic field can alter the trajectory of charge carriers inside the photodetector through the Lorentz force, and induce nonlinear interference electromotive force in the weak signal conditioning and amplification circuit. This leads to an increase in the dark current of the photodetector and a drift in the spectral absorption baseline, ultimately causing a significant decrease in the signal-to-noise ratio of the spectral composition analysis results, or even causing detection failure. Since existing near-infrared spectroscopy detection systems are not equipped with magnetic field sensing and compensation modules, they cannot identify and eliminate this type of magneto-induced noise. This drawback has remained unresolved in the existing technologies. Second, because chemical composition detection and magnetic property detection belong to two independent systems, located at different workstations and at different time points, it is impossible to obtain the composition data and magnetic property data of the same batch of materials at the same location and at the same time. This prevents the production line from establishing an integrated online structure-property relationship model between the chemical composition and physical magnetic properties of the materials, which restricts the efficiency of quality closed-loop optimization and real-time adjustment of process parameters during the production process. Therefore, it is necessary to provide a near-infrared absorption detection spectroscopy analysis system that can operate stably in strong magnetic field interference environments and can simultaneously acquire the chemical composition and physical magnetic property parameters of materials. Summary of the Invention

[0005] The technical problem to be solved by this invention is: in the intelligent manufacturing production line of magnetic materials such as drive motors for new energy vehicles, core components of wind power, rare earth permanent magnets and soft magnetic alloys, how to achieve stable operation of the near-infrared spectroscopy online detection system under strong DC or alternating magnetic field interference environment, and how to achieve synchronous coupling measurement of the chemical composition and physical magnetic properties of materials at the same detection site and at the same time, so as to overcome the technical problem that the composition detection and magnetic property detection are separated in the existing production line and that online structure-property relationship modeling cannot be performed;

[0006] To address the aforementioned technical problems, this invention provides a near-infrared absorption detection spectral analysis system, comprising a dual-modal magneto-optical co-sensing probe. The dual-modal magneto-optical co-sensing probe includes a magnetic sensing element, a near-infrared spectral detection element, a near-infrared light source, and a shielding housing. The magnetic sensing element, the near-infrared spectral detection element, and the near-infrared light source are heterogeneously integrated and packaged within the shielding housing, and are coaxially and at the same location along the same substrate. The magnetic sensing element is used to acquire the three-dimensional magnetic field vector signal at the location of the dual-modal magneto-optical co-sensing probe. The near-infrared light source is used to emit near-infrared light towards the material under test. The near-infrared spectral detection element is used to simultaneously acquire the near-infrared spectral signal reflected or transmitted by the material under test.

[0007] A first signal conditioning circuit and a second signal conditioning circuit are provided. The first signal conditioning circuit is electrically connected to the magnetic sensing element and is used to preprocess the three-dimensional magnetic field vector signal. The second signal conditioning circuit is electrically connected to the near-infrared spectral detection element and is used to preprocess the near-infrared spectral signal. The first signal conditioning circuit and the second signal conditioning circuit are electrically isolated from each other.

[0008] The magnetic field inversion noise dynamic decoupling and compensation module is connected to the first signal conditioning circuit and the second signal conditioning circuit respectively. It is used to input the pre-processed three-dimensional magnetic field vector signal into the pre-trained nonlinear mapping model to obtain the magneto-induced noise compensation signal, and to subtract the pre-processed near-infrared spectral signal from the magneto-induced noise compensation signal to obtain the compensated spectral signal.

[0009] The spectral analysis module is connected to the magnetic field inversion noise dynamic decoupling and compensation module, and is used to perform Lambert-Beer law inversion analysis on the compensated spectral signal to obtain the chemical composition concentration information of the material under test.

[0010] Based on the above technical solution, the present invention also provides a multi-physical quantity online structure-property relationship collaborative modeling module, which is used to extract features from the compensated spectral signal to obtain a spectral feature vector, extract features from the time-series data of the three-dimensional magnetic field vector signal to obtain a magnetic field feature vector, and fuse the two at the feature level to obtain a fused feature vector, which is then input into a pre-trained structure-property relationship mapping model and outputs a joint evaluation result of the chemical composition content and physical and magnetic properties parameters of the tested material. This result can be fed back to the host computer through the production line control interface for closed-loop adjustment of production line process parameters.

[0011] The technical effects that this invention can achieve include:

[0012] 1. This invention integrates a magnetic sensing element and a near-infrared spectral detection element heterogeneously into the same shielded probe and establishes a dynamic decoupling compensation mechanism between the magnetic field vector and the optical baseline drift. This enables the signal-to-noise ratio and detection accuracy of near-infrared spectral detection to be improved through magneto-induced noise compensation calculations in a strong magnetic field environment, thereby achieving stable operation of the spectral detection system against magnetic noise.

[0013] 2. Because the magnetic sensing element and the spectral detection element are coaxially and at the same point on the same probe, the present invention can realize the synchronous acquisition of two types of parameters, namely chemical composition and physical magnetic properties of materials, at the same detection point and at the same time, thus avoiding the data misalignment problem caused by separate detection positions and times in the prior art;

[0014] 3. This invention, through online collaborative modeling of structure-property relationships using multiple physical quantities, can break down the data barriers between component measurement and magnetic property measurement, providing technical support for online modeling of structure-property relationships in production lines and closed-loop optimization of quality. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the overall system architecture of the present invention.

[0016] Figure 2 This is a flowchart of the dual-modal magneto-optical collaborative sensing probe structure of the present invention;

[0017] Figure 3 This is a flowchart of the magnetic field inversion noise dynamic decoupling and compensation method of the present invention. Detailed Implementation

[0018] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0019] System Overall Architecture

[0020] The near-infrared absorption detection spectroscopy analysis system provided by this invention includes a dual-modal magneto-optical co-sensing probe, a signal conditioning and preamplifier circuit, a magnetic field inversion noise dynamic decoupling and compensation module, a spectral analysis module, a multi-physical quantity online structure-property relationship co-modeling module, and a production line control interface. The dual-modal magneto-optical co-sensing probe integrates a magnetic sensing element and a near-infrared spectral detection element for simultaneously acquiring the three-dimensional magnetic field vector signal at the probe's location and the near-infrared spectral signal of the material under test. The signal conditioning and preamplifier circuit preprocesses the two raw signals respectively. The magnetic field inversion noise dynamic decoupling and compensation module is deployed on an edge computing unit. The system receives preprocessed magnetic field and spectral signals, outputs compensated spectral signals, and performs Lambert-Beer law inversion on the compensated signals to obtain chemical component concentrations. The multi-physical quantity online structure-property relationship collaborative modeling module performs feature-level fusion of spectral and magnetic field features and outputs integrated evaluation results of chemical components and physical magnetic properties. The production line control interface feeds back the evaluation results to the host computer or production line programmable logic controller for process closed-loop adjustment. The modules are connected by data paths, and the data types transmitted include raw magnetic field vector data, raw spectral signals, compensated signals, fused feature vectors, and evaluation results.

[0021] Dual-modal magneto-optical co-sensing probe structure

[0022] The dual-modal magneto-optical co-sensing probe comprises, from the outside in, a probe shell, a multi-layer permalloy magnetic shielding layer, an active magnetic field shielding coil, an optical-magnetic isolation barrier, a magnetic sensing element, a near-infrared light source emission window, a near-infrared spectral detection element, and a sample contact window. The probe shell serves as a shielding and encapsulation layer, and is covered with a multi-layer permalloy passive shielding layer. An active magnetic field shielding coil is disposed inside this passive shielding layer. The active magnetic field shielding coil generates a compensating magnetic field through a reverse current, which performs hardware-level primary cancellation of the strong external magnetic field entering the probe, thereby reducing the original magnetic field strength entering the probe and reducing the processing difficulty for subsequent software-level dynamic compensation.

[0023] The magnetic sensing element and the near-infrared spectral detection element are integrated coaxially and at the same point on the same substrate. Their signal pins are connected to their respective independent signal conditioning circuits to ensure that the magnetic signal acquisition path and the optical signal acquisition path are electrically isolated from each other and do not interfere with each other. At the same time, they are aligned with the same detection point in space to achieve coaxial, synchronous, and same-point acquisition. An optical-magnetic isolation partition is set between the magnetic sensing element and the near-infrared spectral detection element to further isolate the electromagnetic and optical crosstalk that may exist between them.

[0024] In this embodiment, the magnetic sensing element is preferably a TMR tunnel magnetoresistive sensor array. The TMR tunnel magnetoresistive sensor is a high-sensitivity magnetic field sensor based on the tunneling magnetoresistive effect of a magnetic tunnel junction. It can detect weakly changing magnetic field vector components and has high sensitivity and a wide dynamic range. In other alternative embodiments, the magnetic sensing element can also be a Hall sensor array, a fluxgate sensor array, or an anisotropic magnetoresistive sensor array. It can also achieve real-time acquisition of the three-dimensional magnetic field vector at the probe location and, in conjunction with the magnetic field inversion compensation method described below, achieve the same technical effect as in this embodiment.

[0025] Near-infrared spectral detection elements can be miniature spectral chips manufactured using microelectromechanical systems (MEMS) technology. These chips integrate optical components such as gratings and detector arrays to perform spectral dispersion and intensity detection of near-infrared light reflected or transmitted from the surface of the material under test. The sample contact window is used to align the sample with the surface of the material under test, and the near-infrared light source emission window is used to emit near-infrared light onto the material under test.

[0026] A Dynamic Decoupling and Compensation Method for Spectral Noise Based on Magnetic Field Inversion

[0027] The magnetic field inversion noise dynamic decoupling and compensation method provided by this invention includes a calibration stage and an online operation stage.

[0028] During the calibration phase, a controlled gradient magnetic field generator is used to apply a frequency-sweeping magnetic field sequence covering the actual operating conditions of the production line to the dual-mode magneto-optical co-sensing probe. According to the preset frequency sweeping scheme, magnetic fields of different intensities and directions are applied sequentially, and the baseline drift of the spectral detection channel output at the corresponding time is recorded simultaneously. This forms a calibration database containing magnetic field vector samples and baseline drift samples. Based on this calibration database, a nonlinear feature mapping model between the magnetic field vector and the optical baseline drift is constructed. This model is used to learn the dynamic nonlinear relationship between the transient three-dimensional magnetic field vector and the dark current and baseline drift of the photodetector. After training, the model parameters are saved to the edge computing unit. The edge computing unit refers to computing hardware deployed at the detection site with real-time data processing capabilities, such as field-programmable gate arrays or digital signal processors, to replace the cloud in completing low-latency online computing.

[0029] During the online operation phase, the three-dimensional magnetic field vector data of the probe's location is acquired in real time by the magnetic sensing element in the dual-modal probe. This real-time magnetic field vector data is input into the trained nonlinear mapping model, and the model outputs the magnetostrictive noise compensation signal at the corresponding moment in real time. Then, the original spectral signal acquired synchronously by the near-infrared spectral detection element in the dual-modal probe is subtracted from the compensation signal to obtain the compensation spectral signal after deducting the magnetostrictive noise. Finally, the compensation spectral signal is analyzed by Lambert-Beer law inversion to obtain the chemical composition concentration information of the tested material, completing a near-infrared online detection with strong magnetic interference resistance. The calibration phase and the online operation phase are connected by model parameter loading to form a closed-loop processing flow of offline calibration and online compensation.

[0030] The Lambert-Beer absorption law involved in the above steps can be expressed as:

[0031] ;

[0032] In the formula, This indicates the absorbance of a material for a specific wavelength of near-infrared light. This indicates the intensity of transmitted light after passing through the material being tested. This represents the initial light intensity before it is incident on the material being tested. This indicates the molar absorptivity of the chemical component being measured at that wavelength. Indicates the concentration of the chemical component being measured. This indicates the effective optical path length of light in the material being tested.

[0033] The compensation calculation described in the online operation phase can be expressed as:

[0034] ;

[0035] In the formula, Indicates time Spectral signal after magneto-induced noise compensation Indicates time The uncompensated spectral signal originally acquired by the probe. This indicates that the nonlinear mapping model is based on time... The magneto-induced noise compensation signal is obtained by predicting the magnetic field vector data.

[0036] In one implementation, the nonlinear mapping model is implemented using a Long Short-Term Memory (LSTM) network. An LSM network is a recurrent neural network with gated memory units, suitable for nonlinear mapping modeling of time-series data such as transient magnetic field vectors. Its internal gated operation can be expressed as:

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] ;

[0043] ;

[0044] In the formula, Indicates time The magnetic field vector characteristics of the input network, This indicates the hidden state at the previous moment. Indicates time The hidden state, , , These represent the activation and output of the forget gate, input gate, and output gate, respectively. Indicates time The state of the cells, Indicates the state of candidate cells. , , , , These represent the weight matrices corresponding to each gate and the output layer, respectively. , , , , These represent the corresponding bias vectors. This represents the Sigmoid activation function. Represents the hyperbolic tangent activation function, symbol This indicates that vectors are multiplied element by element. This indicates a concatenation operation between two vectors.

[0045] In another implementation, the nonlinear mapping model is implemented using an adaptive Kalman filter. An adaptive Kalman filter is a filtering algorithm that can dynamically adjust the noise covariance parameter based on observed data and recursively estimate the system state. Its recursive estimation process can be expressed as:

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] In the formula, Indicates the first The next step is based on the prior estimate obtained from the state in the previous step, i.e., the predicted baseline drift state. Indicates the first The posterior estimate of the step, Represents the state transition matrix. Represents the control input matrix. Indicates the first The control input for the step, i.e., the real-time magnetic field measurement value, Denotes the prior error covariance matrix. Let represent the posterior error covariance matrix from the previous step. Represents the process noise covariance matrix. Represents the observation matrix. Represents the observation noise covariance matrix. Indicates the first Kalman gain of the step, Indicates the first The actual observed value of the step, i.e., the measured baseline drift. Indicates the first The posterior estimate of the step, i.e., the compensation signal of the final output, Represents the identity matrix, superscript Indicates matrix transpose, superscript This represents finding the inverse of a matrix.

[0052] In other alternative implementations, the nonlinear mapping model between the magnetic field and the optical baseline drift can also be implemented using a support vector regression model or a radial basis function neural network. Alternatively, in production line conditions where the magnetic field variation is relatively simple and repeatable, a lookup table of magnetic field strength and compensation amount can be pre-constructed, and the compensation signal can be obtained by interpolation through the lookup table during online operation. This can also achieve dynamic decoupling and compensation of magneto-induced noise, serving as an effective supplementary solution to the present invention.

[0053] Multi-physical quantity online structure-property relationship collaborative modeling architecture

[0054] The compensated spectral signal and the three-dimensional magnetic field vector time series data output from the dual-mode probe are respectively fed into the spectral feature extraction submodule and the magnetic field feature extraction submodule. The spectral feature extraction submodule performs feature peak extraction or principal component dimensionality reduction on the compensated spectral signal to obtain a spectral feature vector characterizing the distribution of the chemical composition of the material. The magnetic field feature extraction submodule extracts features from the real-time magnetic field vector time series data, including features such as coercivity estimation, magnetic field fluctuation amplitude, and magnetic field change rate, to obtain a magnetic field feature vector characterizing the physical and magnetic properties and potential defects of the material.

[0055] The two feature vectors mentioned above then enter the feature-level fusion module for weighted concatenation to obtain a fused feature vector. The fusion method can be expressed as:

[0056] ;

[0057] In the formula, This represents the fused feature vector. Represents the spectral eigenvector. Represents the eigenvectors of the magnetic field. , These represent the weighting coefficients corresponding to the spectral and magnetic field features, respectively. These weighting coefficients can be adaptively learned through an attention mechanism. The symbols are... This indicates a concatenation operation between two vectors.

[0058] The feature vectors are fused and input into a pre-trained offline structure-function mapping model. The output of this model can be expressed as:

[0059] ;

[0060] In the formula, This represents the combined evaluation results of the material's chemical composition and physical and magnetic properties output by the model. This represents a nonlinear mapping function, which can be implemented using a random forest regression model or a deep neural network regression model. This represents the aforementioned fused feature vector. This represents the set of parameters obtained by training the mapping model offline.

[0061] The joint evaluation results are fed back to the host computer or the production line programmable logic controller through the production line control interface. This is used to determine in real time whether the product quality meets the standards and to make closed-loop adjustments to the production line process parameters, such as heat treatment temperature and sintering time.

[0062] Technical Effect Description

[0063] As can be seen from the above embodiments, the present invention heterogeneously integrates magnetic sensing elements and near-infrared spectral detection elements into the same shielded probe, and, in conjunction with the magnetic field inversion noise dynamic decoupling and compensation method, can suppress spectral detection baseline drift through the subtraction compensation operation in a strong magnetic field environment. At the same time, through the coaxial and co-located integrated structure, it realizes the synchronous acquisition of material chemical composition and physical magnetic property parameters at the same detection site and at the same time. Furthermore, by combining online structure-property relationship collaborative modeling of multiple physical quantities, it realizes the transformation from passive anti-interference to active multi-modal fusion, providing technical support for the quality closed-loop optimization of intelligent manufacturing production lines for magnetic materials. The above embodiments are only some examples of the present invention. All equivalent substitutions or improvements made on the basis of the technical solution of the present invention should fall within the protection scope of the present invention.

Claims

1. A near-infrared absorption detection spectroscopy analysis system, characterized in that, include: A dual-modal magneto-optical co-sensing probe includes a magnetic sensing element, a near-infrared spectral detection element, a near-infrared light source, and a shielding housing. The magnetic sensing element, the near-infrared spectral detection element, and the near-infrared light source are heterogeneously integrated and packaged within the shielding housing, and are coaxially and at the same location along the same substrate. The magnetic sensing element is used to acquire the three-dimensional magnetic field vector signal at the location of the dual-modal magneto-optical co-sensing probe. The near-infrared light source is used to emit near-infrared light to the material under test. The near-infrared spectral detection element is used to simultaneously acquire the near-infrared spectral signal reflected or transmitted by the material under test. A first signal conditioning circuit and a second signal conditioning circuit are provided. The first signal conditioning circuit is electrically connected to the magnetic sensing element and is used to preprocess the three-dimensional magnetic field vector signal. The second signal conditioning circuit is electrically connected to the near-infrared spectral detection element and is used to preprocess the near-infrared spectral signal. The first signal conditioning circuit and the second signal conditioning circuit are electrically isolated from each other. The magnetic field inversion noise dynamic decoupling and compensation module is connected to the first signal conditioning circuit and the second signal conditioning circuit respectively. It is used to input the pre-processed three-dimensional magnetic field vector signal into the pre-trained nonlinear mapping model to obtain the magneto-induced noise compensation signal, and to subtract the pre-processed near-infrared spectral signal from the magneto-induced noise compensation signal to obtain the compensated spectral signal. The spectral analysis module is connected to the magnetic field inversion noise dynamic decoupling and compensation module, and is used to perform Lambert-Beer law inversion analysis on the compensated spectral signal to obtain the chemical composition concentration information of the material under test.

2. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The shielding shell includes a multilayer permalloy passive shielding layer and an active magnetic field shielding coil. The active magnetic field shielding coil is disposed inside the multilayer permalloy passive shielding layer and is used to generate a compensating magnetic field through a reverse current to perform primary cancellation of the external magnetic field entering the shielding shell.

3. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The magnetic sensing element is a TMR tunnel magnetoresistive sensing array.

4. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The dual-mode magneto-optical co-sensing probe also includes an optical-magnetic isolation plate, which is disposed between the magnetic sensing element and the near-infrared spectral detection element to isolate signal interference between the magnetic sensing element and the near-infrared spectral detection element.

5. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The nonlinear mapping model is a long short-term memory network model. The long short-term memory network model performs nonlinear mapping on the temporal characteristics of the three-dimensional magnetic field vector signal based on the forget gate, input gate, output gate and cell state, and outputs the magneto-noise compensation signal.

6. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The nonlinear mapping model is an adaptive Kalman filter. The adaptive Kalman filter recursively estimates the three-dimensional magnetic field vector signal based on the state transition matrix, the observation matrix, and the dynamically adjusted noise covariance parameter, and outputs the magneto-noise compensation signal.

7. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, The nonlinear mapping model is pre-trained as follows: a frequency-sweeping magnetic field sequence covering the actual operating magnetic field intensity range of the production line is applied to the dual-mode magneto-optical co-sensing probe using a controlled gradient magnetic field generator. The intensity coverage range of the frequency-sweeping magnetic field sequence is consistent with the DC or alternating magnetic field intensity range actually experienced by the material under test in the production line. Corresponding spectral baseline drift samples are collected simultaneously, and a calibration database containing magnetic field vector samples and spectral baseline drift samples is constructed. The nonlinear mapping model is then trained based on the calibration database.

8. The near-infrared absorption detection spectral analysis system according to claim 1, characterized in that, It also includes a multi-physical quantity online structure-property relationship collaborative modeling module. The multi-physical quantity online structure-property relationship collaborative modeling module is used to extract features from the compensated spectral signal to obtain a spectral feature vector, extract features from the time-series data of the three-dimensional magnetic field vector signal to obtain a magnetic field feature vector, perform feature-level fusion of the spectral feature vector and the magnetic field feature vector to obtain a fused feature vector, and input the fused feature vector into a pre-trained structure-property relationship mapping model to output the joint evaluation results of the chemical composition content and physical magnetic property parameters of the tested material.

9. The near-infrared absorption detection spectral analysis system according to claim 8, characterized in that, The online structure-property relationship collaborative modeling module for multiple physical quantities assigns weight coefficients to the spectral feature vector and the magnetic field feature vector respectively, and then concatenates them to obtain the fused feature vector. The weight coefficients are learned through an attention mechanism.

10. The near-infrared absorption detection spectral analysis system according to claim 8, characterized in that, It also includes a production line control interface, which is connected to the multi-physical quantity online structure-property relationship collaborative modeling module, and is used to feed back the joint evaluation results to the host computer for closed-loop adjustment of production line process parameters.

Images