Method and system for detecting a radiation dose of an irradiated cross-linked region

Abstract

This invention discloses a method and system for detecting radiation dose in irradiated cross-linked regions, belonging to the field of radiation dose detection technology. Specifically, it includes: inputting the coordinates of the core point to be measured and the coordinates of its surrounding associated points, and simultaneously acquiring transient fluorescence spectra, thermoluminescence signals, and conductivity variation data of each point within the irradiation window; constructing a physical configuration association model based on the distribution of associated points to form a scattered field association data network; when the residual between the measured data of the core point and the model prediction exceeds a dynamic threshold, activating the inverse solution process of the scattered field, using the associated node data to reconstruct the theoretical data sequence of the core point, replacing the disturbed original data; finally, inputting the reconstructed sequence into the dose calculation model, and outputting the corrected radiation dose value. This invention can effectively reduce measurement errors caused by local material inhomogeneity and transient interference, improving the reliability of dose detection.

Description

Technical Field

[0001] This invention relates to the field of radiation dose detection technology, and specifically to a method and system for detecting radiation dose in an irradiated cross-linked region. Background Technology

[0002] In the field of radiation processing technology, precise radiation dose irradiation of materials is a key process for achieving performance modification, such as polymer crosslinking, material sterilization, and semiconductor device performance tuning. As a core process parameter, the accuracy and uniformity of the spatial distribution of the irradiation dose directly determine the quality and performance consistency of the final product. Therefore, developing technologies capable of accurately and reliably measuring and evaluating the radiation dose absorbed by specific regions of materials has always been an important direction in radiation application research and industrial quality control.

[0003] Currently, various measurement methods based on the physical effects of radiation-matter interaction have been developed for detecting dose in irradiated areas. Common techniques include using thermoluminescent detectors to record cumulative dose, retrieving dose by measuring transient fluorescence spectra generated after material irradiation, or inferring energy deposition by monitoring transient changes in the material's electrical properties (such as conductivity). These methods typically rely on deploying sensors at or near the measurement point to collect single or limited physical signals and performing calculations based on pre-calibrated dose-response relationships. Furthermore, to improve measurement reliability, some techniques introduce reference points or employ multi-point measurements to reduce local measurement errors through spatial interpolation or statistical averaging.

[0004] However, in actual radiation processing environments, especially when the material itself has microstructural inhomogeneities, internal defects, or local fluctuations in the irradiation field, the aforementioned existing methods still face significant challenges. Because they rely on direct measurement signals from single or limited points, when the measurement point happens to be located in an abnormal region of the material or is subject to transient disturbances, the acquired signal may not accurately reflect the overall radiation energy absorption of that region, but only reflects a localized, specific physical response. This measurement bias introduced by local material inhomogeneities or transient disturbances is difficult to effectively eliminate through simple spatial interpolation or correction using a few reference points, leading to systematic errors in the assessment of the true dose in the core region, affecting the accuracy of process control and the uniformity of product performance. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting radiation dose in irradiated cross-linked regions, thereby solving the following technical problems:

[0006] Existing radiation dose detection methods rely on direct measurement signals from single or limited points when materials are non-uniform or have local interference. These signals are easily affected by local specificity and cannot effectively distinguish between the true radiation absorption signal and the spurious signal introduced by local anomalies or transient disturbances in the material. This leads to deviations in the assessment of the true dose in the core area.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for detecting radiation dose in an irradiated cross-linked region includes the following steps:

[0009] S1. Input the coordinates of the core point of the irradiated crosslinked region to be detected, and obtain the coordinates of multiple associated points within a preset spatial range centered on the core point;

[0010] S2. Obtain transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all associated points within the same irradiation time window;

[0011] S3. Based on the spatial distribution of all associated points, construct a physical configuration association model describing the radiation scattering path, take the data of the core point and associated points as nodes, connect the nodes according to the scattering path parameters, and form a scattering field association data network.

[0012] S4. When the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold, the data reconstruction process of the inverse solution of the scattered field is activated.

[0013] S5. The data reconstruction process starts from the time point when the residual exceeds the threshold. The physical configuration association model and the associated node data are used to reverse solve the theoretical data of the core point, and the corresponding data segments in the original measured data of the core point are replaced to generate the reconstructed core point data sequence.

[0014] S6. Input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point.

[0015] As a further aspect of the present invention: in step S2, the process of acquiring transient fluorescence spectrum, transient thermoluminescence signal, and transient conductivity change data is as follows:

[0016] Multi-sensor probes are deployed at the core point coordinates and the coordinates of each associated point. The multi-sensor probes integrate a miniature fiber optic spectrometer, a thermoluminescent detector sheet doped with rare earth elements, and an interdigitated microelectrode array. The miniature fiber optic spectrometer collects transient fluorescence spectra, the thermoluminescent detector sheet collects transient thermoluminescent signals, and the microelectrode array collects transient conductivity change data.

[0017] A unified irradiation pulse synchronization trigger signal is sent to all multi-sensor probes, and all multi-sensor probes start data acquisition synchronously. After the acquisition is completed, the spectral intensity sequence, thermoluminescence curve and conductivity change time sequence from the core point and all associated points are time-series aligned according to the timestamp recorded by the acquisition device, and integrated into a multimodal transient dataset according to the coordinate point index.

[0018] As a further aspect of the present invention: in S3, the process of constructing the physical configuration correlation model is as follows:

[0019] Based on the coordinates of the core point and each associated point, calculate the spatial straight-line distance from the core point to each associated point, and calculate the spatial azimuth and elevation angles with the core point as the origin; read the mass attenuation coefficient and Compton scattering cross section data of the tested material to incident radiation from the material database.

[0020] Using the core point as the radiation source, and combining the mass attenuation coefficient and Compton scattering cross section data, the theoretical signal attenuation factor and theoretical propagation time delay of the radiation reaching each associated point after being scattered by the material are calculated. The correspondence between the theoretical attenuation factor and theoretical time delay of all core point and associated point pairs is organized into a matrix to form the initial scattering transmission matrix. The initial scattering transmission matrix is ​​then normalized to generate a standardized scattering transmission matrix.

[0021] As a further aspect of the present invention: the process of calculating the theoretical signal attenuation factor and theoretical propagation time delay of the radiation after scattering by the material to each associated point is as follows:

[0022] For each pair of core points and associated points, the calculated straight-line distance, spatial azimuth angle, and elevation angle are read, and the material parameter database is queried to obtain the mass attenuation coefficient and Compton scattering cross section value of the tested material under the current irradiation energy; based on the straight-line distance and mass attenuation coefficient, the primary attenuation factor of radiation propagating along the straight path is calculated.

[0023] Based on the spatial azimuth, elevation angle, and Compton scattering cross section values, the secondary attenuation factor and additional path delay caused by Compton scattering are calculated using a single scattering physical model. The primary attenuation factor and the secondary attenuation factor are multiplied to obtain the total theoretical attenuation factor. The theoretical direct arrival time calculated based on the straight-line distance and the speed of light is added to the additional scattering path delay to obtain the total theoretical time delay.

[0024] As a further aspect of the present invention: in step S4, the process of determining the dynamic threshold is as follows:

[0025] Retrieve historical experimental records from the historical measurement database that are the same type of material being tested as the current material and were completed under similar irradiation geometry conditions. For each historical experimental record, extract the measured data sequence of the historical core points. Based on the associated point data corresponding to the historical experimental record and the constructed physical configuration association model, calculate the predicted data sequence of the historical core points.

[0026] The absolute value sequence of residuals between historical measured sequences and historical predicted sequences is calculated. The absolute value sequence of residuals from all historical records is collected and statistically distributed to obtain the cumulative empirical distribution function of the absolute value of residuals. The preset high quantile values ​​are extracted from the cumulative empirical distribution function as the benchmark residual threshold. The nominal value of the beam intensity of the current irradiation mission is read, and the benchmark residual threshold is scaled according to the linear relationship between the nominal value of the beam intensity and the preset scaling factor to generate a dynamic threshold.

[0027] As a further aspect of the present invention: in step S5, the data reconstruction process of inverse solution of the scattered field specifically includes:

[0028] The signal propagation relationship described by the physical configuration association model is expressed as a linear forward propagation operator. The signal propagation relationship is the relationship from the core point to each associated point. The linear forward propagation operator is solved by inverse operation, and regularization constraints are introduced in the process to obtain the inverse solution operator.

[0029] When the scattered field associated data network determines that the residual of the core point data exceeds the dynamic threshold at a set time T, a continuous data window of fixed time length is extracted with time T as the endpoint. The complete historical observation data sequence of all associated nodes within the data window is extracted, and the historical observation data sequence of the associated nodes is input into the inverse solver. The inverse solver outputs a core point theoretical data sequence corresponding to the length of the data window. The core point theoretical data sequence is used to completely replace the data of the same time period in the original measured data sequence of the core point.

[0030] As a further aspect of the present invention: the expression and inversion process of the linear forward propagation operator is as follows:

[0031] The signal propagation relationship defined by the physical configuration association model is expressed as a set of linear equations. The signal propagation relationship is the relationship from the core point to each associated point. The coefficients are extracted from the set of linear equations and arranged in order to form the system matrix. The system matrix is ​​the mathematical expression of the linear forward propagation operator. The matrix inversion operation is performed on the system matrix. During the operation, a Tikhonov regularization term based on the matrix eigenvalues ​​is introduced. The result of the matrix inversion operation is the inverse system matrix, which is the inverse solution operator.

[0032] As a further aspect of the present invention: in step S6, the working process of the dose calculation model is as follows:

[0033] The dose calculation model employs a deep feedforward neural network structure, which includes an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer is the same as the length of the reconstructed core point data sequence, which is then fed into the input layer. Multiple hidden layers sequentially perform nonlinear feature transformations and hierarchical abstractions on the input data. The output layer consists of linear neurons, whose output values ​​are the calculated core point corrected radiation dose values. The dataset used to train the deep feedforward neural network comes from calibration experiments in a standard radiation field environment. These calibration experiments use standard dosimeters and generate multimodal transient data and corresponding known dose values ​​for various combinations of materials and irradiation conditions.

[0034] The present invention also includes a radiation dose detection system for an irradiated cross-linked region, used to implement the above-described radiation dose detection method for an irradiated cross-linked region, comprising:

[0035] The coordinate input module is used to input the coordinates of the core point of the irradiated cross-linked region to be detected, and to obtain the coordinates of multiple associated points centered on the core point within a preset spatial range;

[0036] The multi-source data acquisition module is used to acquire transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all related points within the same irradiation time window;

[0037] The associated network construction module constructs a physical configuration associated model describing the radiation scattering path based on the spatial distribution of all associated points. It uses the data of the core points and associated points as nodes and connects the nodes according to the scattering path parameters to form a scattering field associated data network.

[0038] The residual decision module is used to activate the data reconstruction process of inverse solution of the scattered field when the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold.

[0039] The data reconstruction module is used to reverse solve the theoretical data of the core point by starting from the time point when the residual exceeds the threshold, using the physical configuration association model and the associated node data, and replacing the corresponding data segments in the original measured data of the core point to generate the reconstructed core point data sequence.

[0040] The dose output module is used to input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point.

[0041] The beneficial effects of this invention are:

[0042] This invention constructs a physical configuration correlation model describing the radiation scattering path by simultaneously acquiring the coordinates of multiple associated points centered on the core point and transient multimodal data. This places the core point measurement within a physical constraint network composed of spatially associated point data. When the deviation between the measured core point data and the predicted value calculated based on this model and associated point data exceeds a dynamic threshold, it indicates that the data at that point may be affected by local non-uniformity. Subsequently, the inverse solution process of the scattered field is activated, using historical observation data of the associated points to inversely deduce the theoretical data sequence that the core point should present during that period, and using this to replace the disturbed original measured data. This mechanism effectively removes spurious signals introduced by local defects in the material or transient disturbances. Finally, a well-trained deep neural network is used to perform dose mapping on the reconstructed clean data sequence, ultimately outputting a more accurate and reliable core point corrected radiation dose value, significantly improving the anti-interference capability and accuracy of dose assessment in complex materials and non-uniform irradiation fields. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1 This is a schematic flowchart of a radiation dose detection method for an irradiated cross-linked region according to the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a radiation dose detection system for an irradiated cross-linked region according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figure 1 As shown, the present invention provides a method for detecting radiation dose in an irradiated cross-linked region, comprising the following steps:

[0048] S1. Input the coordinates of the core point of the irradiated crosslinked region to be detected, and obtain the coordinates of multiple associated points within a preset spatial range centered on the core point;

[0049] S2. Obtain transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all associated points within the same irradiation time window;

[0050] S3. Based on the spatial distribution of all associated points, construct a physical configuration association model describing the radiation scattering path, take the data of the core point and associated points as nodes, connect the nodes according to the scattering path parameters, and form a scattering field association data network.

[0051] S4. When the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold, the data reconstruction process of the inverse solution of the scattered field is activated.

[0052] S5. The data reconstruction process starts from the time point when the residual exceeds the threshold. The physical configuration association model and the associated node data are used to reverse solve the theoretical data of the core point, and the corresponding data segments in the original measured data of the core point are replaced to generate the reconstructed core point data sequence.

[0053] S6. Input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point.

[0054] In a preferred embodiment of the present invention, the process of acquiring transient fluorescence spectrum, transient thermoluminescence signal, and transient conductivity change data in step S2 is as follows:

[0055] First, an integrated multi-sensor probe with the same structure is deployed at the pre-determined coordinates of the core point of the irradiated cross-linked region to be detected, as well as at the coordinates of each associated point. The probe adopts a miniaturized packaging design, which integrates three independent sensing units: a miniature fiber optic spectrometer with an operating wavelength range covering 250 nm to 850 nm, a thermoluminescent detector sheet based on lithium fluoride material and doped with a specific concentration of terbium ions, and an interdigitated gold microelectrode array with an electrode spacing of 20 micrometers.

[0056] The miniature fiber optic spectrometer receives fluorescence signals from the material surface via a 100-micrometer-diameter silica fiber. The fluorescence signal is generated by a 355-nanometer nanosecond pulsed laser, synchronized with the main irradiation source. An internal grating disperses the incident light, which is then received by a linear CCD detector with 2048 pixels, each corresponding to a specific wavelength channel. The final output is a time-varying spectral intensity matrix.

[0057] A rare-earth-doped thermoluminescent detector, 0.5 mm thick, is mounted close to the material surface. When exposed to radiation, traps in the detector's crystal structure capture electrons. Under subsequent excitation by a synchrotron laser pulse, the captured electrons gain energy and undergo transitions, releasing light of a specific wavelength as they return to their ground state. The resulting emission curve is recorded by a high-sensitivity photomultiplier tube mounted adjacent to the detector, forming a transient thermoluminescent signal sequence.

[0058] The interdigitated microelectrode array has electrodes in direct contact with the material surface. A constant 1-volt bias voltage is applied between the two electrodes, and a high-precision current sensing unit is connected in series. When the material is irradiated, its internal conductivity changes transiently, causing a change in the current flowing through the electrodes. The current sensing unit records the transient fluctuations of the current at a sampling rate of 1 MHz, and this fluctuation data is directly converted to correspond to the transient conductivity change of the material.

[0059] To achieve strict time synchronization across multiple points, a central synchronization controller generates a unified irradiation pulse synchronization trigger signal. This signal is a standard 5-volt TTL level pulse, simultaneously distributed to the control interfaces of all multi-sensor probes via a low-latency coaxial cable network. Upon receiving the rising edge of the trigger signal, all probes synchronously initiate their respective data acquisition processes within a time deviation of less than 1 nanosecond. The acquisition duration for each probe is 10 milliseconds, covering the entire process of a single irradiation pulse.

[0060] After data acquisition, each probe packages and transmits the recorded data, along with a high-precision timestamp, to the data processing unit. The timestamp for each data point is provided by the probe's built-in synchronization clock chip, with a time accuracy of 0.1 microseconds. The data processing unit first performs time-series alignment on all spectral intensity sequences, thermoluminescence curves, and conductivity change time series from the core point and various associated points based on the timestamps. The alignment process ensures that the readings from different sensors are correctly matched at the same physical moment, ultimately correcting the time deviation of each channel's data to less than 0.1 milliseconds.

[0061] After time-series alignment, the data is indexed and integrated according to the spatial coordinates of its source. Data from the core point is labeled as one group, and data from each associated point is labeled as an independent group. Each group contains three strictly synchronized time series: a spectral sequence consisting of intensity values ​​from 2048 wavelength channels, a thermoluminescence intensity sequence consisting of photomultiplier tube voltage values, and a conductivity variation sequence converted from current values. These multiple groups of time-series data, organized by coordinate points, are ultimately packaged and stored into a structured multimodal transient dataset, which fully records the transient physical response of each point within a spatial range centered on the core point under a single irradiation pulse event.

[0062] In another preferred embodiment of the present invention, the process of constructing the physical configuration correlation model in step S3 is as follows:

[0063] The construction of the physical configuration association model begins with the precise calculation of spatial geometric relationships. First, the processing unit reads the input three-dimensional spatial coordinates of the core point to be detected. These coordinates are typically in millimeters and represented as a vector containing X, Y, and Z components, for example, (10.5, 25.3, 0.0). Simultaneously, it reads the coordinate set of all associated points. The number of associated points can be set according to the required detection accuracy, such as 16 or 32. For each associated point, the processing unit performs a vector subtraction operation, subtracting the core point's coordinate vector from the associated point's coordinate vector to obtain a spatial displacement vector. The magnitude of this vector is the straight-line distance between the core point and the associated point. For example, an associated point with coordinates (12.5, 27.8, 1.2) will have a calculated straight-line distance of approximately 3.65 millimeters.

[0064] Furthermore, the processing unit calculates the spherical coordinate system parameters with the core point as the origin based on this displacement vector. Specifically, the spatial azimuth angle is obtained by calculating the angle between the projection of the displacement vector onto the XY plane and the positive X-axis, typically ranging from 0 to 360 degrees. The pitch angle is obtained by calculating the angle between the displacement vector and the positive Z-axis (usually defined as the direction perpendicular to the material surface), ranging from 0 to 180 degrees. All angle values ​​are stored in radians or degrees for subsequent physical calculations. For example, for the above displacement vector, the calculated azimuth angle is approximately 63.4 degrees, and the pitch angle is approximately 70.5 degrees.

[0065] After completing the geometric parameter calculations for all point pairs, the process proceeds to the material physics parameter lookup stage. The processing unit accesses a pre-generated material parameter database. This database stores the fundamental physical parameters of various commonly used materials (such as polyethylene, polypropylene, and silicone rubber) at different radiation energies. Key parameters include the material's mass attenuation coefficient for photons of specific energies, typically measured in square centimeters per gram, and the Compton scattering cross section, typically measured in square centimeters per gram or Barn per atom. For example, for photons with an energy of 1.25 MeV (e.g., a cobalt-60 gamma source), the mass attenuation coefficient of polyethylene is approximately 0.065 square centimeters per gram, and its Compton scattering cross section can be obtained by consulting the X-ray and gamma-ray attenuation database published by the International Atomic Energy Agency. Based on the known energy of the current irradiation mission (e.g., 1.25 MeV) and the identification code of the material being tested (e.g., "polyethylene-high density"), the processing unit extracts the corresponding mass attenuation coefficient and Compton scattering cross section data from the database.

[0066] After obtaining the material parameters, the core physical calculations begin, which involve calculating the theoretical signal attenuation factor and theoretical propagation time delay for each pair of core points and associated points. This calculation consists of two main parts: primary (linear) propagation contribution and secondary (single scattering) propagation contribution.

[0067] First, the primary attenuation factor is calculated. It is assumed that the core point is an ideal point source, and the radiated photons propagate along a straight path to the correlation point. The processing unit uses the calculated straight-line distance, combined with the material's mass attenuation coefficient and density, to perform the calculation. The material density is also obtained from a material database; for example, the density of polyethylene is approximately 0.95 grams per cubic centimeter. The primary attenuation factor characterizes the remaining fraction of the photon after it has penetrated the material directly to the correlation point without any scattering. Its value is equal to a negative power of the natural constant e, where the exponent is the product of the mass attenuation coefficient, the material density, and the straight-line distance. For example, for polyethylene with a distance of 3.65 mm, the calculated primary attenuation factor is approximately 0.78, indicating that approximately 78% of the original signal strength is retained.

[0068] Next, the secondary attenuation factor and additional path delay caused by the single Compton scattering are calculated. This calculation is based on a simplified physical model: assuming that photons emitted from the core radiation source undergo only one Compton scattering event along their propagation path, changing direction exactly after scattering and heading towards the associated point. The processing unit uses the previously calculated spatial azimuth and elevation angles to determine the possible directions and path changes of the scattering.

[0069] The calculation process requires differential cross-sectional data from the Kleinrenko formula, which describes the probability of a photon undergoing Compton scattering at different scattering angles. The scattering angle is obtained through the geometric relationship between the core point, the virtual scattering point, and the associated point. The processing unit, near the straight path from the core point to the associated point, calculates the sum of probabilities of a photon being scattered towards the associated point within all possible spatial volume elements where single scattering might occur, using numerical integration. This is combined with the energy loss factor of the scattered photon (due to the Compton effect, the energy of the scattered photon is lower than that of the incident photon) to finally obtain the secondary attenuation factor. This factor is typically much smaller than the primary attenuation factor, for example, it might be 0.05. Simultaneously, since the scattering path is longer than the straight path, the total optical path from the core point to the virtual scattering point, and then from the scattering point to the associated point, is calculated. Subtracting the optical path of the straight path, and then dividing by the speed of light in vacuum (approximately 3 x 10⁸ meters per second), yields the additional path delay caused by scattering. This delay is very small, typically in the picosecond to nanosecond range, for example, calculated to be 2.3 picoseconds.

[0070] After obtaining the primary and secondary contributions, the total theoretical attenuation factor and the total theoretical time delay can be calculated. The total theoretical attenuation factor is equal to the product of the primary attenuation factor and the secondary attenuation factor. This is because, in the single scattering model, the photon must simultaneously satisfy the superposition of the probabilities of two independent events: arriving directly without scattering and arriving after one scattering (approximately additive under weak scattering conditions, but more accurate models often use multiplication or more complex combinations after considering coherence). For example, the product of the primary factor 0.78 and the secondary factor 0.05 yields a total attenuation factor of approximately 0.039.

[0071] The total theoretical time delay is obtained by adding the theoretical direct arrival time to the scattering-induced path delay. The theoretical direct arrival time equals the straight-line distance divided by the speed of light in the material. The speed of light in materials is slightly lower than the speed of light in a vacuum, so the refractive index of the material must be considered; for most polymers, the refractive index is close to 1.5. For example, for a distance of 3.65 mm, the theoretical direct arrival time is approximately 18.3 picoseconds. Adding the scattering-induced path delay of 2.3 picoseconds, the total theoretical time delay is approximately 20.6 picoseconds.

[0072] After performing the above calculations on all core-point-associated point pairs, the processing unit organizes the results into a mathematical matrix called the initial scattering transfer matrix. The row indices of this matrix represent core points; since there is usually only one core point, it is actually a row vector. The column indices represent individual associating points. Each element in the matrix contains two data points: the total theoretical attenuation factor and the total theoretical time delay. For example, a configuration with 16 associating points will generate a 1x16 matrix, where each element is a data structure containing two floating-point numbers.

[0073] Finally, a normalization operation is performed on the initial scattering transfer matrix. Row normalization aims to convert the relative weights of the correlated point signals from absolute physical quantities into probability distributions or relative intensity distributions, facilitating consistency comparisons in the subsequent network. Specifically, the operation involves: first, calculating the sum of the total theoretical attenuation factors for all elements in the row; then, dividing the total theoretical attenuation factor for each element in the row by the calculated sum. After this operation, the sum of the new attenuation factors for all elements in the row will equal 1. Time delay values ​​are typically not included in the normalization process, retaining their original physical quantity values. The normalized matrix is ​​the standardized scattering transfer matrix, which constitutes the core mathematical expression of the physical configuration correlation model. It describes the normalized theoretical signal intensity distribution and relative time delay relationship from the core radiation source to each correlated point in space, providing a physical basis for subsequently constructing a scattering field correlation data network.

[0074] In another preferred embodiment of the present invention, the process of determining the dynamic threshold in step S4 is as follows:

[0075] First, a dedicated data retrieval module accesses a historical measurement database. This database stores complete data packages of numerous past irradiation experiments in a structured format. Each data package contains experimental metadata, such as the unique identifier of the tested material, the type and energy of the irradiation source, the geometric coordinates of the core point and associated points, and all acquired transient raw data. The retrieval module sets query conditions based on the parameters of the current detection task: the material identifier must match exactly; for example, if the current detection is of high-density polyethylene, only records with the material identifier "HDPETypeIV" in the database are selected. The irradiation geometry must be similar, which is determined by comparing the relative spatial distribution patterns of the core point to the associated point group. For example, if the current arrangement is a spherical array with a radius of 5 mm, then experimental records in the database where the maximum distance between the core point and the associated points is between 4 mm and 6 mm are selected. Assuming there are 23 historical experimental records in the database that meet the conditions, the retrieval module extracts them all for subsequent analysis.

[0076] Next, for each historical experimental record, the processing unit performs the following operations. It extracts the historical core point measured data sequence from the record. This sequence is typically a vector containing thousands of time points, each representing a scalar observation value obtained by fusing multimodal data collected at a specific moment. Simultaneously, the processing unit reads the historical observation data sequences of all associated points corresponding to that historical record. Using the physical configuration association model constructed or used at the time of the historical experiment—the parameters of which can be obtained from a database, such as a 1x16 normalized scattering transfer matrix—the processing unit takes the historical observation data sequence of the associated points as input and calculates the historical core point predicted data sequence through the forward propagation relationship defined by the model. The predicted sequence is strictly aligned with the measured sequence in terms of time points.

[0077] Then, the residuals between the two are calculated. The processing unit calculates the absolute value of the difference between the measured sequence value and the predicted sequence value point by point, generating a residual absolute value sequence of the same length as the original sequence. This process is repeated for all 23 historical experimental records to obtain 23 residual absolute value sequences.

[0078] Subsequently, statistical aggregation and distribution fitting are performed. The processing unit aggregates all residual values ​​from these 23 sequences into a single dataset, which may contain tens of thousands of residual data points. Analysis of the distribution characteristics of these residual values ​​reveals that they typically do not follow a standard normal distribution, but rather are closer to a right-skewed distribution, such as a Weiber distribution or a log-normal distribution. The processing unit uses maximum likelihood estimation to fit all the aggregated residual values ​​onto a Weiber distribution, obtaining the specific values ​​of its shape and scale parameters. Based on the cumulative distribution function of the fitted Weiber distribution, the cumulative empirical distribution function of the absolute values ​​of the residuals can be obtained.

[0079] Next, a preset high quantile value is extracted from the cumulative empirical distribution function as the baseline residual threshold. This high quantile is usually set in the configuration file, such as the 95th percentile or the 98th percentile. The processing unit calculates the residual value corresponding to a cumulative probability of 0.95. Assuming that the residual value corresponding to the 95th percentile is calculated to be 0.15 based on the fitted Weiber distribution, this value is set as the baseline residual threshold, meaning that under ideal historical conditions, 95% of the residual fluctuations are below this level.

[0080] Finally, the dynamic threshold is dynamically adjusted based on the real-time parameters of the current irradiation mission. The processing unit reads the nominal beam intensity value of the current irradiation mission, which is provided in real time by the control unit of the irradiation facility. The unit may be grays per second (Gys per second), for example, the current value is 5.0 Gys per second. A preset linear scaling factor relationship is defined, for example, the scaling factor equals the nominal beam intensity value divided by a reference beam intensity, which can be set to 10 Gys per second. The processing unit multiplies the baseline residual threshold of 0.15 by the scaling factor of 0.5 to obtain the final dynamic threshold of 0.075. This dynamic threshold will be used in the current detection mission to determine whether there are significant anomalies in the real-time data of the core points. Through this process, the threshold can be adaptively adjusted according to the key operating parameter of irradiation intensity, improving the rationality and environmental adaptability of the criterion.

[0081] In another preferred embodiment of the present invention, the data reconstruction process of inverse solution of the scattered field in step S5 is specifically as follows:

[0082] Based on the established physical configuration correlation model, the deterministic signal propagation relationship it describes is expressed as a linear forward propagation operator. The core of this model is the normalized scattering transfer matrix, which defines the signal propagation weights from a single core radiation source to multiple correlated points in space. Assuming there are 16 correlated points, this matrix is ​​a 1x16 matrix, denoted as A. Each element A{1,j} of matrix A is a complex number or a numerical value containing amplitude and phase information. The amplitude part comes from the normalized theoretical attenuation factor, and the phase part is related to the theoretical time delay (e.g., the delay time is converted into a phase shift at a specific frequency). This matrix A is the compact mathematical expression of the linear forward propagation operator, establishing a linear transformation relationship between the core point transmitted signal vector (a scalar time-varying sequence) and the received signal vectors at each correlated point (a set of 16-channel time-varying sequences).

[0083] To infer the signal at the core point from the observation data of the associated points, it is necessary to obtain the inverse of the forward propagation operator. This process is achieved by solving the matrix inverse problem. Specifically, the physical relationship is expressed as a set of linear equations. Let the signal amplitude at the core point at discrete time point k be sk, and the observations at the same time points of the 16 associated points constitute a 16-dimensional column vector dk. Ideally, there exists a relationship dk = A. T ×sk, where A T This is the transpose of matrix A. Since the core signal is a scalar, A needs to be transposed to accommodate matrix multiplication. Expanding this relationship over time, for a sequence containing N time points, a large system of linear equations can be constructed. However, under steady-state or slowly varying assumptions, a more efficient approach is to directly manipulate the frequency domain or the statistical eigenvalues. In practice, matrix A is often augmented to construct a square matrix or the least squares principle is utilized.

[0084] The key step in the inverse solution is constructing and inverting the system matrix. The processing unit first extracts matrix A from the physical configuration correlation model. To construct an invertible system matrix H, the processing unit calculates A and its transpose AH. T The product of these two elements yields a new 16x16 square matrix H = A × A. T This H matrix represents a certain autocorrelation mapping from the signal space of the correlated points to itself, and is the basis for subsequent inversion operations.

[0085] Directly inverting matrix H can be mathematically ill-conditioned, meaning the condition number of H is too large, and even small observation errors can cause drastic oscillations in the inversion results. To address this issue, regularization constraints must be introduced. In this implementation, a Tikhonov regularization method based on matrix eigenvalues ​​is used. The processing unit first performs eigenvalue decomposition on matrix H to obtain its eigenvalues ​​λi (i=1,…,16) and corresponding eigenvectors. These eigenvalues ​​typically vary in magnitude, and some values ​​may be very close to zero, causing the matrix to approach singularity.

[0086] The processing unit sets a regularization parameter λ according to preset rules. A common strategy is to set λ to a small proportion of the largest eigenvalue, such as one-thousandth. Simultaneously, an eigenvalue truncation threshold is set, for example, treating all eigenvalues ​​less than one ten-thousandth of the largest eigenvalue as zero, to avoid numerical instability. Then, the regularized inverse matrix Hreginv is constructed. For each non-zero eigenvalue λi, its corresponding contribution in the inverse matrix is ​​corrected to 1 / (λi+λ), instead of the original 1 / λi. For eigenvalues ​​truncated to zero, their corresponding contributions are directly set to zero. Using the eigenvector matrix obtained from the decomposition, the processing unit calculates the stable regularized inverse matrix Hreginv through inverse combination.

[0087] Finally, the inverse solution operator G is obtained by calculating G=A. T The result is obtained by multiplying the vector of observed data from the associated points by G. This G is a 16-row, 1-column matrix (or vector), which is the inverse solution operator. The meaning of operator G is: multiplying the vector of observed data from the associated points by G yields an optimal linear unbiased estimate of the signal amplitude at the core point.

[0088] When the scattered field correlation data network is conducting real-time monitoring, and at a specific time T, it determines that the residual of the core point data exceeds the dynamic threshold, the data reconstruction process is immediately activated. Assume T is the 102.5th millisecond calculated from the start of monitoring.

[0089] After activation, the processing unit extracts a continuous data window of a fixed time length, starting from the current moment T. This window length is set during algorithm initialization, for example, 20 milliseconds. This means extracting all data within the time interval from 82.5 milliseconds to 102.5 milliseconds.

[0090] Next, the processing unit extracts the complete historical observation data sequence of all associated nodes within this 20-millisecond data window from the cache. The data for each associated point is a time-varying sequence recorded at a high sampling rate (e.g., 1 MHz), thus containing 20,000 data points within the 20-millisecond window. The 16 associated points then form a 16-channel spatiotemporal observation data block with 20,000 points per channel.

[0091] Then, the historical observation data sequences of these associated nodes are input into the inverse solver G. The operation is performed point-by-point across time. For each sampling time t within the window (a total of 20,000 time points), the processing unit extracts the observation values ​​of 16 associated points at that time, forming a 16-dimensional column vector d(t). This vector is multiplied by the inverse solver G to obtain a scalar value s estimated(t) = G. T ×d(t). This operation is repeated for each time point within the window, ultimately generating a 20-millisecond sequence of theoretical core point data containing 20,000 estimates. This sequence represents the signal that the core point "should" have generated during that time period, calculated based on observations of 16 surrounding related points and constrained by a physical model.

[0092] Finally, data replacement is performed. The processing unit locates the portion of the original measured data sequence corresponding to the same time period, specifically the original data segment from 82.5 milliseconds to 102.5 milliseconds. This segment of original data is completely replaced with the theoretical data sequence of the core point calculated above. After the replacement, the 20 milliseconds of disturbed or abnormal data in the original data sequence are covered by the theoretical data reconstructed based on physical correlations, thus generating a locally reconstructed core point data sequence that better conforms to overall physical consistency, used for subsequent dose calculations. The entire reverse solution and data reconstruction process is completed within milliseconds, meeting the requirements of real-time or near-real-time processing.

[0093] In another preferred embodiment of the present invention, the working process of the dose calculation model in step S6 is as follows:

[0094] The dose calculation model is implemented using a pre-trained deep feedforward neural network. This deep feedforward neural network has a specific layered structure. The input layer consists of a variable number of neurons, with the number of nodes strictly equal to the length of the reconstructed core point data sequence. For example, if the reconstructed sequence is obtained within a 20-millisecond time window at a sampling rate of 1 MHz, the sequence contains 20,000 data points, and the corresponding input layer is designed to have 20,000 neurons. The reconstructed core point data sequence, i.e., the time-varying signal amplitude array corrected by the aforementioned inverse solution process, is directly and in parallel input to these 20,000 input neurons, with each neuron receiving the value at a specific time point in the sequence.

[0095] The input layer is followed by multiple hidden layers. The number of hidden layers and the number of neurons in each layer are pre-defined according to the model complexity requirements. For example, one configuration includes 5 hidden layers with 1024, 512, 256, 128, and 64 neurons respectively. All neurons between adjacent layers are fully connected. Each hidden layer neuron performs a weighted summation of all its received inputs, applies a bias term, and then produces an output through a non-linear activation function. The activation function is typically a modified linear unit function, meaning the output equals the input value when the input value is greater than zero, and the output is zero when the input value is less than or equal to zero. This hierarchical structure allows the network to perform progressive non-linear feature transformations and abstractions on the high-dimensional time-series input data, gradually extracting deep pattern features related to radiation dose.

[0096] The output of the last hidden layer is passed to the output layer. The output layer is designed as a single linear neuron. This neuron performs a linear weighted sum of all inputs from the previous hidden layer and adds a bias, but without any nonlinear activation function. Its output value is a continuous real number, which is directly interpreted as the calculated core-point corrected radiation dose value, typically in kilogras or megagras. Therefore, the mapping function of the entire network is to transform a time-series input of length 20,000 into a single dose value.

[0097] The parameters of this deep feedforward neural network, namely all connection weights and biases, were obtained through supervised training on a large calibration dataset. The training dataset consisted of numerous calibration experiments conducted in controlled standard radiation field environments. For example, in a secondary standard dosimetry laboratory, using a finger-shaped ionization chamber or alanine dosimeter traceable by the National Institute of Metrology as the standard dosimeter, various material samples, such as polyethylene, polypropylene, and polyvinyl chloride of different densities, were irradiated under different irradiation conditions. Variations in irradiation conditions included changing the type of irradiation source, such as a cobalt-60 gamma source or an electron accelerator; changing the dose rate, for example, from 0.1 Gy / s to 10 Gy / s; and changing the cumulative dose, for example, from 1 kGy to 100 kGy. During each calibration experiment, multimodal transient data at a specific point on the surface of the irradiated material were simultaneously acquired. This involved recording transient fluorescence spectra, thermoluminescence signals, and conductivity changes using the same sensing devices as in the detection method of this invention, and recording the true value of the absorbed dose at that point as measured by the standard dosimeter. The resulting dataset contains tens of thousands, even hundreds of thousands, of corresponding samples of "multimodal transient data sequences" and "known absorbed dose values." Using this dataset, the weights and biases of the neural network are iteratively adjusted through backpropagation and gradient descent optimizers such as the Adam optimizer to minimize the mean square error between the network's predicted dose values ​​and the actual dose values. This results in a set of optimized model parameters, enabling the network to accurately deduce radiation dose from transient data sequences. The trained model parameters are then embedded into the processing unit of the detection device for online dose calculation.

[0098] Please see Figure 2 As shown, the present invention also includes a radiation dose detection system for an irradiated cross-linked region, used to implement the above-described radiation dose detection method for an irradiated cross-linked region, comprising:

[0099] The coordinate input module is used to input the coordinates of the core point of the irradiated cross-linked region to be detected, and to obtain the coordinates of multiple associated points centered on the core point within a preset spatial range;

[0100] The multi-source data acquisition module is used to acquire transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all related points within the same irradiation time window;

[0101] The associated network construction module constructs a physical configuration associated model describing the radiation scattering path based on the spatial distribution of all associated points. It uses the data of the core points and associated points as nodes and connects the nodes according to the scattering path parameters to form a scattering field associated data network.

[0102] The residual decision module is used to activate the data reconstruction process of inverse solution of the scattered field when the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold.

[0103] The data reconstruction module is used to reverse solve the theoretical data of the core point by starting from the time point when the residual exceeds the threshold, using the physical configuration association model and the associated node data, and replacing the corresponding data segments in the original measured data of the core point to generate the reconstructed core point data sequence.

[0104] The dose output module is used to input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point.

[0105] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for detecting radiation dose in an irradiated cross-linked region, characterized in that, Includes the following steps: S1. Input the coordinates of the core point of the irradiated crosslinked region to be detected, and obtain the coordinates of multiple associated points within a preset spatial range centered on the core point; S2. Obtain transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all associated points within the same irradiation time window; S3. Based on the spatial distribution of all associated points, construct a physical configuration association model describing the radiation scattering path, take the data of the core point and associated points as nodes, connect the nodes according to the scattering path parameters, and form a scattering field association data network. S4. When the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold, the data reconstruction process of the inverse solution of the scattered field is activated. S5. The data reconstruction process starts from the time point when the residual exceeds the threshold. The physical configuration association model and the associated node data are used to reverse solve the theoretical data of the core point, and the corresponding data segments in the original measured data of the core point are replaced to generate the reconstructed core point data sequence. S6. Input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point. In S3, the process of constructing the physical configuration association model is as follows: Based on the coordinates of the core point and each associated point, calculate the spatial straight-line distance from the core point to each associated point, and calculate the spatial azimuth and elevation angles with the core point as the origin; read the mass attenuation coefficient and Compton scattering cross section data of the tested material to incident radiation from the material database. Using the core point as the radiation source, and combining the mass attenuation coefficient and Compton scattering cross section data, the theoretical signal attenuation factor and theoretical propagation time delay of the radiation after scattering by the material to each associated point are calculated. The correspondence between the theoretical signal attenuation factor and theoretical propagation time delay of all core point and associated point pairs is organized into a matrix to form the initial scattering transmission matrix. The initial scattering transmission matrix is ​​normalized to generate a standardized scattering transmission matrix. In step S4, the process of determining the dynamic threshold is as follows: Retrieve historical experimental records from the historical measurement database that are the same type of material being tested as the current material and were completed under similar irradiation geometry conditions. For each historical experimental record, extract the measured data sequence of the historical core points. Based on the associated point data corresponding to the historical experimental record and the constructed physical configuration association model, calculate the predicted data sequence of the historical core points. The absolute value sequence of residuals between historical measured sequences and historical predicted sequences is calculated. The absolute value sequence of residuals from all historical records is collected and statistically distributed to obtain the cumulative empirical distribution function of the absolute value of residuals. The preset high quantile values ​​are extracted from the cumulative empirical distribution function as the benchmark residual threshold. The nominal value of the beam intensity of the current irradiation mission is read, and the benchmark residual threshold is scaled according to the linear relationship between the nominal value of the beam intensity and the preset scaling factor to generate a dynamic threshold.

2. The method for detecting radiation dose in an irradiated cross-linked region according to claim 1, characterized in that, In step S2, the process of acquiring transient fluorescence spectrum, transient thermoluminescence signal, and transient conductivity change data is as follows: Multi-sensor probes are deployed at the core point coordinates and each associated point coordinates. The multi-sensor probes integrate a miniature fiber optic spectrometer, a thermoluminescent detector sheet doped with rare earth elements, and an interdigitated microelectrode array. A miniature fiber optic spectrometer acquires transient fluorescence spectra, a thermoluminescence detector acquires transient thermoluminescence signals, and a microelectrode array acquires transient conductivity change data. A unified irradiation pulse synchronization trigger signal is sent to all multi-sensor probes, and all multi-sensor probes start data acquisition synchronously. After the acquisition is completed, the spectral intensity sequence, thermoluminescence curve and conductivity change time sequence from the core point and all associated points are time-series aligned according to the timestamp recorded by the acquisition device, and integrated into a multimodal transient dataset according to the coordinate point index.

3. The method for detecting radiation dose in an irradiated cross-linked region according to claim 1, characterized in that, The process of calculating the theoretical signal attenuation factor and theoretical propagation time delay of the radiation after scattering by the material to each associated point is as follows: For each pair of core points and associated points, the calculated straight-line distance, spatial azimuth angle, and elevation angle are read, and the material parameter database is queried to obtain the mass attenuation coefficient and Compton scattering cross section value of the tested material under the current irradiation energy; based on the straight-line distance and mass attenuation coefficient, the primary attenuation factor of radiation propagating along the straight path is calculated. Based on the spatial azimuth, elevation angle, and Compton scattering cross section values, the secondary attenuation factor and additional path delay caused by Compton scattering are calculated using a single scattering physical model. The primary attenuation factor and the secondary attenuation factor are multiplied to obtain the total theoretical signal attenuation factor. The theoretical direct arrival time calculated based on the straight-line distance and the speed of light is added to the additional scattering path delay to obtain the total theoretical propagation time delay.

4. The method for detecting radiation dose in an irradiated cross-linked region according to claim 1, characterized in that, In S5, the data reconstruction process of inverse solution of the scattered field is as follows: The signal propagation relationship described by the physical configuration association model is expressed as a linear forward propagation operator. The signal propagation relationship is the relationship from the core point to each associated point. The linear forward propagation operator is solved by inverse operation, and regularization constraints are introduced in the process to obtain the inverse solution operator. When the scattered field associated data network determines that the residual of the core point data exceeds the dynamic threshold at a set time T, a continuous data window of fixed time length is extracted with time T as the endpoint. The complete historical observation data sequence of all associated nodes within the data window is extracted, and the historical observation data sequence of the associated nodes is input into the inverse solver. The inverse solver outputs a core point theoretical data sequence corresponding to the length of the data window. The core point theoretical data sequence is used to completely replace the data of the same time period in the original measured data sequence of the core point.

5. The method for detecting radiation dose in an irradiated cross-linked region according to claim 4, characterized in that, The expression and inverse process of the linear forward propagation operator are as follows: The signal propagation relationship defined by the physical configuration association model is expressed as a set of linear equations. The signal propagation relationship is the relationship from the core point to each associated point. The coefficients are extracted from the set of linear equations and arranged in order to form the system matrix. The system matrix is ​​the mathematical expression of the linear forward propagation operator. The matrix inversion operation is performed on the system matrix. During the operation, a Tikhonov regularization term based on the matrix eigenvalues ​​is introduced. The result of the matrix inversion operation is the inverse system matrix, which is the inverse solution operator.

6. The method for detecting radiation dose in an irradiated cross-linked region according to claim 1, characterized in that, In S6, the working process of the dose calculation model is as follows: The dose calculation model adopts a deep feedforward neural network structure, which includes an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer is the same as the length of the reconstructed core point data sequence. The reconstructed core point data sequence is fed into the input layer. Multiple hidden layers sequentially perform nonlinear feature transformation and hierarchical abstraction on the input data. The output layer consists of linear neurons, and the output value of the linear neurons is the calculated core point corrected radiation dose value. The dataset used to train the deep feedforward neural network comes from the calibration experiment in the standard radiation field environment. The calibration experiment uses a standard dosimeter to generate multimodal transient data and corresponding known dose values ​​for various combinations of materials and irradiation conditions.

7. A radiation dose detection system for an irradiated cross-linked region, used to implement the radiation dose detection method for an irradiated cross-linked region as described in any one of claims 1-6, characterized in that, include: The coordinate input module is used to input the coordinates of the core point of the irradiated cross-linked region to be detected, and to obtain the coordinates of multiple associated points centered on the core point within a preset spatial range; The multi-source data acquisition module is used to acquire transient fluorescence spectra, transient thermoluminescence signals, and transient conductivity changes of the core point and all related points within the same irradiation time window; The associated network construction module constructs a physical configuration associated model describing the radiation scattering path based on the spatial distribution of all associated points. It uses the data of the core points and associated points as nodes and connects the nodes according to the scattering path parameters to form a scattering field associated data network. The residual decision module is used to activate the data reconstruction process of inverse solution of the scattered field when the residual between the measured data of the core point in the scattered field association data network and the predicted value calculated based on the physical configuration association model exceeds the dynamic threshold. The data reconstruction module is used to reverse solve the theoretical data of the core point by starting from the time point when the residual exceeds the threshold, using the physical configuration association model and the associated node data, and replacing the corresponding data segments in the original measured data of the core point to generate the reconstructed core point data sequence. The dose output module is used to input the reconstructed core point data sequence into the dose calculation model and output the corrected radiation dose value of the core point.

Images