Device and method for measuring an object comprising beta activity

Abstract

Method for characterizing an object comprising at least one radionuclide emitting beta radiation, the method comprising: - placing a detector facing the object; - detecting the radiation emitted by the object; - forming, from the radiation detected by the detector, an input spectrum comprising a beta component; - applying to the input spectrum an identification algorithm associated with a radionuclide, the identification algorithm determining the presence in the object of the radionuclide associated with the identification algorithm; - identifying, from d), the radionuclide contained in the object; - applying a deconvolution algorithm to the input spectrum in order to estimate the contribution of each radionuclide identified in e) to the input spectrum.

Description

Technical Field

[0001] The technical field of this invention is the measurement of the β activity of an object using energy dispersive spectroscopy. Background Technology

[0002] Understanding the radiation status of processes and equipment in nuclear facilities is essential for establishing reliable decommissioning programs and managing waste (and particularly for waste classification and disposal). In-situ nondestructive nuclear measurement techniques, combined with modeling techniques, enable the creation of radioactive inventories for processes, fixed equipment, and civil engineering structures.

[0003] Gamma spectroscopy is one of the most commonly used passive, non-destructive nuclear measurement techniques for obtaining qualitative and quantitative information about the presence of gamma-ray radionuclides in an object. Gamma spectroscopy is used to obtain gamma spectra in which characteristic peaks can be identified, corresponding to characteristic markers that allow for the identification of radionuclides.

[0004] However, some radionuclides emit little or no gamma radiation. One example of these is the so-called pure beta emitter—its in-situ radiographic characterization is difficult due to the very short path of freedom of electrons in dense media. An example of a pure beta emitter is... 90 In the laboratory, destructive analysis of field-collected samples is commonly used to identify pure β-emitters and quantify their activities. However, these destructive laboratory measurements have certain drawbacks: issues regarding the representativeness of the collected samples, analytical costs, and time.

[0005] The publication Vetter K, "In-situ quantification of gamma-ray and beta-only emitting radionuclides", arXiv, Apr. 09, 2023, http: / / arxiv.org / abs / 2304.07632, hereinafter referred to as [Vetter], describes a spectral measurement method in which a compact CdZnTe semiconductor detector is used to acquire the spectrum. The device is placed close enough to the object to be characterized to detect the spectrum emitted by gamma-ray and beta-only emitting radionuclides. 137 The gamma radiation produced by Cs, by 137 Cs-emitted internal conversion electrons, and those emitted by 90 Sr-emitted β - Radiation. The following components are extracted from this spectrum: components representing interactions occurring in the surface region of the detector, including interactions originating from electrons (internal conversion electrons, β-electrons). -The spectrum consists of most of the radiation; and a component representing interactions occurring deep within the detector, including interactions caused by gamma photons. A screen is placed between the detector and the object to acquire the spectrum in order to determine which interactions in the surface regions are caused by gamma photons. The spectrum generated by electrons and photons is then deconvolved using a maximum likelihood algorithm to quantify the activity of radionuclides.

[0006] The method described in the publication [Vetter] has the advantage of enabling the estimation of the activity of a pure beta emitter in an object without sampling and without destructive analysis. However, this method requires determining the depth of interactions within the detector, which is relatively complex. Decomposing the spectrum into components representing interactions originating from electrons and components representing interactions originating from photons can be tedious, especially when performed in-situ (i.e., under non-laboratory conditions). Furthermore, interpreting these spectra is relatively complex.

[0007] In fact, using a CdZnTe detector is another drawback: this type of detector is limited by its size of only a few centimeters. 3 The crystal. Finally, the need to perform two measurements (one with a screen and one without) is another limitation of implementation.

[0008] The inventors have developed an alternative method with the same purpose as the method described in the publication [Vetter]. The inventors' method does not require consideration of the depth of interaction within the detector. Furthermore, it is not limited to the use of semiconductor detectors and can be advantageously implemented in scintillator detectors. Summary of the Invention

[0009] The first subject of this invention is a method for characterizing an object containing at least one radioactive nuclide that emits beta radiation, the method comprising:

[0010] - a) A detector is placed facing an object and configured to acquire a spectrum representing the distribution of energy released in the detector by radiation emitted by the object;

[0011] - b) During the acquisition, the radiation emitted by the object is detected by a detector, and the spectrum of the detected radiation is acquired;

[0012] - c) Based on the spectrum of the detected radiation, form an input spectrum containing a β component, which corresponds to the distribution of energy released in the detector by the β radiation;

[0013] - d) Apply an identification algorithm associated with a radionuclide to the input spectrum. The identification algorithm is configured to determine the presence of the radionuclide associated with the identification algorithm in the object. Step d) is repeated for various radionuclides by implementing different identification algorithms.

[0014] - e) Identify each radionuclide contained in the object according to d);

[0015] - f) Apply the input spectrum deconvolution algorithm to estimate the contribution of each radionuclide identified in e) to the input spectrum;

[0016] - g) For each radionuclide identified in e), estimate the activity of the radionuclide and / or the depth to which the radionuclide extends in the object, based on the contribution of the radionuclide to the input spectrum estimated in f).

[0017] Steps d), f), and g) are implemented by the processing unit based on the input spectrum. Step e) can be implemented by the processing unit.

[0018] According to one possibility, in step d), each identification algorithm is an artificial intelligence identification algorithm associated with each radionuclide, and at least two different radionuclides are associated with two different corresponding identification algorithms.

[0019] Each recognition algorithm can be a neural network.

[0020] According to one possibility, the deconvolution algorithm is based on a deconvolution database containing at least one spectrum representing each radionuclide identified in e).

[0021] Based on one possibility:

[0022] - For a given radionuclide, the deconvolution database contains various spectra representing the various distributions of that radionuclide in an object;

[0023] Step g) includes determining the distribution of the radionuclide in the object.

[0024] Based on one possibility:

[0025] - For a given radionuclide, the deconvolution database contains various spectra representing the radionuclide at various depths in the object, from the surface of the object facing the detector;

[0026] - Step g) includes determining the depth to which the radionuclide extends within the object.

[0027] At least one radionuclide associated with the identification algorithm can be a pure beta emitter.

[0028] The detector may include an organic scintillator material for detecting radiation emitted by an object.

[0029] The detector may include a volume of inorganic scintillator or semiconductor with a thickness of less than 10 mm, which is placed facing the object and the thickness is measured in a direction perpendicular to the object.

[0030] Based on the possibility that the object possesses natural activity, step c) includes:

[0031] - Estimate the natural activity spectrum of an object;

[0032] - Subtract the object’s natural activity spectrum from the spectrum obtained in step b) to form the input spectrum.

[0033] The detector may include a movable screen configured to be positioned between the detector and the object, and the method includes:

[0034] - Obtain the background spectrum with the screen positioned between the detector and the object;

[0035] - In step c), the background spectrum is subtracted from the spectrum obtained in step b) to form the input spectrum.

[0036] The second aspect of the invention is a detection device comprising a detector configured to acquire a spectrum of beta radiation emitted by an object, the spectrum representing the distribution of energy released in the detector during ionizing radiation interactions. The device includes a processing unit configured to implement steps d) to f) of the method according to the first aspect of the invention.

[0037] The detector may include an organic scintillator material for detecting radiation emitted by an object.

[0038] The detector may include an inorganic scintillator or a semiconductor with a thickness of less than 10 mm.

[0039] The detector may include a movable screen configured to be positioned between the detector and the object.

[0040] The invention will be better understood after reading the disclosure of exemplary embodiments presented in the following sections of this specification and in conjunction with the accompanying drawings. Attached Figure Description

[0041] Figure 1 The measuring device that allows the implementation of the present invention is illustrated schematically.

[0042] Figure 2A The laboratory measurements are shown. 137 Cs spectra. Unless otherwise stated, for each spectrum described in this patent application, the x-axis corresponds to energy (in MeV) and the y-axis corresponds to the number of interactions detected.

[0043] Figure 2B The measurements taken in the laboratory are shown. 137 γ spectrum of Cs.

[0044] Figure 2C The measurements taken in the laboratory are shown. 90 The spectrum of Sr.

[0045] Figure 3 It shows that it contains 137 Cs and 90 The βγ spectrum of an Sr object.

[0046] Figure 4 One modeling configuration is illustrated schematically.

[0047] Figure 5A and Figure 5B The modeling spectrum is shown, along with the artificial radionuclides in two different configurations. 137 Cs and 90 The contributions of Sr and natural radionuclides.

[0048] Figure 6 The main steps of the method according to the invention are illustrated schematically.

[0049] Figure 7A Another modeling configuration is illustrated schematically.

[0050] Figure 7B An example of the training spectrum is shown.

[0051] Figure 8A It shows that it contains 90 Modeling spectrum of objects in Sr.

[0052] Figure 8B It shows Figure 8A The probability of the presence of various radionuclides in the spectrum. This is related to the output of the identification algorithm.

[0053] Figure 9A It shows that it contains 14 C and 36 Modeling spectrum of objects with Cl.

[0054] Figure 9B It shows Figure 9A The probability of the presence of various radioactive nuclides in the spectrum.

[0055] Figure 10A It shows the specific values ​​for a given 90 The spectrum of Sr activity is modeled by a depth distribution that results in an exponential activity gradient.

[0056] Figure 10B The diagram shows the results after integral normalization. Figure 10A Modeling spectrum.

[0057] Figure 10C It shows Figure 10B Each spectrum in the middle represents uniformity.90 The ratio between the spectra of Sr activity (normalized by its integral).

[0058] Figure 10D It is for depth distribution 137 Cs activity, and Figure 10C Equivalent diagram.

[0059] Figure 11 It has a depth of up to 40 mm and an exponential gradient distribution. 90 Modeling spectrum and uniformity of Sr activity 90 Comparison of Sr activity spectra.

[0060] Figures 12A to 12D The stepwise fitting of the spectrum generated by the deconvolution algorithm to the measured spectrum is shown based on the number of iterations.

[0061] Figure 13 The illustration schematically shows a sample collected laterally in a channel within a graphite-moderated nuclear reactor.

[0062] Figures 14A to 14D The graphite sample is shown 137 Cs and 90 The probability of Sr's existence: Each probability is established by applying various recognition neural networks 100 times to each sample.

[0063] Figure 15 An example of an implementation of spectral deconvolution to determine the contribution of various radionuclides to the measured spectrum is shown. Figure 15 In the diagram, the x-axis corresponds to each channel.

[0064] Figure 16A The spectrum measured on the graphite sample is shown.

[0065] Figure 16B The spectrum measured with a screen placed between the detector and the graphite sample is shown.

[0066] Figure 16C It shows how to subtract Figure 16A and Figure 16B The spectrum obtained is shown in the figure. Detailed Implementation

[0067] Figure 1 A measuring device is shown that allows for the measurement of the activity of an object 2. The device includes a scintillator detector 10 containing a scintillator material 11, preferably an organic scintillator material, and more preferably a polyvinyl toluene (PVT)-based scintillator material, as described in the publication Venara J. et al., "Design and development of a portable β-spectrometer for..."90 As described in "Sr activity measurements in contaminated matrices", Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. 953, p.163081, Feb. 2020, light pulses are formed under the influence of the interaction between ionizing radiation and the scintillator material. These light pulses are converted into electrical pulses by one or more photodetectors 12. The electrical pulses are then processed by an energy spectrum circuit 13. The energy spectrum circuit 13 is configured to form an amplitude histogram of the pulses detected by the organic scintillator during acquisition.

[0068] The relationship between pulse amplitude and energy value can typically be determined using an energy calibration function generated by energy calibration. When an ionized particle deposits its full energy into a scintillator, the resulting pulse amplitude corresponds to the particle's energy before interacting with the detector material. The detector's energy calibration, i.e., the relationship between pulse amplitude and energy, is performed using the energy values ​​of electrons emitted via internal conversion to discrete energy values. 207 Bi and / or 137 The calibration is performed using a Cs source. When the source used for calibration emits gamma radiation, certain characteristic energies can be utilized, such as the energy corresponding to the Compton edge or the energy of the photoelectric peak.

[0069] Organic scintillator detectors are suitable for measuring the energy spectrum of beta-type charged particles. Organic scintillators are also sensitive to X-ray or gamma-ray ionized photons. However, the low atomic numbers of the constituent materials of organic scintillators make them less likely to undergo photoelectric interactions compared to inorganic scintillators or semiconductor detectors.

[0070] The scintillator detector is covered with a thin, opaque sheath 14, such as an 18 μm thick aluminized polyethylene terephthalate (PET) film, to ensure impermeability to ambient light. This small thickness minimizes the probability of beta radiation interactions. In the remainder of the specification, the term beta particle refers to beta... - particle.

[0071] The thickness e of the scintillator material is, for example, 4 mm. The diameter of the example scintillator material is 76 mm. Organic scintillator materials have the advantage of being relatively insensitive to gamma radiation due to their low atomic number. Furthermore, this type of scintillator limits the backscattering of β particles. Another advantage is the stability of the scintillator material response in terms of thermal changes. When exposed to a given radiation, the resulting light intensity response is stable in the range of 0° to 50°C, meeting the requirements for field use.

[0072] The measuring device includes a processing unit 20 configured to implement the spectral processing steps described below. The processing unit 20 is programmed to execute instructions stored in a memory connected to the processing unit via a wired or wireless link. The processing unit 20 may, in particular, include a microprocessor.

[0073] According to one variation, the detector may include a semiconductor suitable for beta spectroscopy measurements. For example, it may be silicon, such as planar silicon.

[0074] The advantage of organic scintillator materials is that they can be manufactured in various shapes and sizes. When the object is a sample, the shape of the organic scintillator material can match the shape of the sample.

[0075] According to one possibility, detector 10 includes a movable screen 15, which serves as a shutter and is configured to be placed in:

[0076] - Close the position between detector 10 and object 2;

[0077] - Or open the position to release the space between detector 10 and object 2, such as Figure 1 As shown.

[0078] The presence of the movable screen 15 is not necessary.

[0079] In this example, screen 15 can be translated in a plane parallel to detector material 11. The screen is made of aluminum, for example, with a thickness of 4 mm.

[0080] This invention is based on the detection of β particles and optionally γ particles emitted by object 2 by detector 10. Object 2 is the object to be inspected, due to β and γ emitters (such as...) 137 Cs) or pure β emitters (such as ... 14 C 90 Sr or 36 Radioactive nuclides such as Cl may have volume activity or surface activity.

[0081] To achieve sufficient sensitivity to β particles, the detector is preferably placed at a very short distance *d* from the object to be characterized. This distance *d* is preferably non-zero and between *mm* and *cm*—for example, 1 cm. This allows for the formation of a βγ spectrum, as the small distance between the detector and the object increases the contribution of β particles to the spectrum.

[0082] One possibility is to limit the contribution of gamma radiation to the β-γ spectrum by inserting screen 15 between detector 10 and object 2. This makes it possible to acquire the gamma spectrum. The gamma spectrum is then subtracted from the β-γ spectrum, which allows the formation of a β spectrum that is considered to represent only the β radiation emitted by the object.

[0083] Therefore, the present invention can be implemented on the βγ spectrum or the β spectrum. Typically, the present invention is implemented on a spectrum where the contribution of β radiation is greater than 10%, or even 20%, 30%, or more. The contribution refers to the amount of interaction caused by β radiation that is taken into account when forming the spectrum.

[0084] In the example described below, the object is a concrete wall. For materials such as concrete, an additional difficulty is the presence of natural activity. While natural activity is low, it can complicate the interpretation of measurements, especially when dealing with low levels of artificial activity, such as approximately 1 Bq / g or lower. Among the naturally occurring radioactive elements that may be present in the object, for example, can be mentioned... 40 K, and 232 Th and 238 A daughter product of U. Such radioactive elements can be found, for example, in objects made of concrete.

[0085] Another well-known difficulty in nuclear measurements is the potential presence of background gamma noise caused by artificial radioactivity in the detector's environment (outside the object being analyzed).

[0086] Figures 2A to 2C The spectrum was measured in a laboratory using known... 137 Cs or 90 The Sr point source is placed 10 mm in front of the detector at a distance d. Figure 2A It shows 137 The spectrum of the Cs source, which is composed of 137 Cs and 137m The β and γ radiation emitted during Ba equilibrium is formed. Figure 2B It shows 137 The gamma radiation spectrum of Cs. Figure 2B Through as described above 137 This is achieved by placing a screen between the Cs source and the detector to absorb the beta radiation emitted by the source. This is done through comparison. Figure 2A and Figure 2B As can be seen, a large portion of the spectrum is due to the contribution of beta radiation. This is because the distance between the detector and the source is very small.

[0087] Figure 2C It shows 90 Sr source (and) 90 The spectrum (of γ equilibrium) is entirely due to the interaction of β particles in the detector. The activity of the standard source is 9 kBq, and the acquisition time is 10 minutes.

[0088] Figure 3 Demonstrates use in the laboratory 137 Cs and 90 Spectra measured from two standard Sr point sources (activity approximately 9 kBq).

[0089] Based on such Figure 3 The method described below for the spectrum shown below aims to:

[0090] - Identify β or βγ emitters present in the object being analyzed;

[0091] - Estimate the distribution of each identified radionuclide in the object: surface distribution or volume distribution, and if possible, estimate the thickness of each radionuclide present from the surface;

[0092] - and / or quantify the activity of each identified radionuclide.

[0093] This invention is particularly advantageous when the presence of at least one pure beta emitter in the object being analyzed is suspected. The inventors observed that the short free path of electrons in the detector material (in this case, an organic scintillator) results in significantly different spectra depending on whether the activity is distributed near the surface of the object being analyzed or at a deeper depth. Indeed, using low-density organic scintillator materials allows for different spectra to be obtained depending on the depth at which the activity is distributed within the object, especially in cases involving radionuclides emitting high-energy beta particles. Therefore, while inorganic scintillator materials, such as NaI, CsI, or LaBr3, or semiconductors (Si, Ge), can be used, the use of organic scintillator materials allows for better differentiation of spectra corresponding to activities distributed at various depths within the object being analyzed. Using organic scintillators allows for a higher ratio of beta to gamma contributions to the measured spectrum. Furthermore, compared to scintillators made of inorganic materials, the size of organic scintillators can be designed to expose a larger area to the object being examined. Additionally, organic scintillators can be manufactured in various planar and non-planar shapes, enabling good matching with the geometry of the object being examined.

[0094] Although preferred, the use of an organic scintillator is not essential for carrying out the invention. For example, a semiconductor detector or an inorganic scintillator detector that is thin enough to limit the sensitivity to gamma radiation can be used. For example, a Si crystal-based semiconductor detector or an inorganic scintillator can be used.

[0095] The effect of natural activity

[0096] One of the target applications of this invention is to perform low-level inspections of concrete civil engineering structures to verify whether the activity level meets predetermined targets. For example, this could involve achieving a so-called cleanliness level (for...). 137 Cs or 90 The activity level of Sr is 1 Bq / g. At such levels, the natural activity of some materials (such as concrete) can complicate the interpretation of the spectrum, as mentioned above.

[0097] The inventors modeled the natural activity against the reference Figure 1 The effect of the spectrum measured by the described device. Figure 4 A modeled geometry is shown, in which the detector is confined within a metal sheath 16. Figure 4 In the model, a uniform thickness of 1 cm starting from the surface of the object was created. 90 Sr and 137 Cs activity, the object is facing the detector. Modeling was performed using the MCNP transport code (MCNP stands for Monte Carlo N particle). Figure 5A The modeling of the β spectrum is shown, taking into account 90 Sr and 137 Cs activity is 1 Bq / g, and 40 K, 232 Th and 238 The U activities are 0.5 Bq / g, 0.03 Bq / g, and 0.02 Bq / g, respectively. This β spectrum is modeled considering only the interaction of β particles within the detector. Figure 5A The contribution of each radionuclide is shown. Natural activity accounts for 26% of the total β spectrum. Figure 5A (The "natural total β" spectrum). The value of 26% corresponds to the ratio of the integral of the natural total spectrum to the integral of the total spectrum.

[0098] Figure 5B The modeled γ spectrum is shown, taking into account 137 Cs activity is 1 Bq / g, and 40 K, 232 Th and 238 U activities were 0.5 Bq / g, 0.03 Bq / g, and 0.02 Bq / g, respectively. This gamma spectrum was modeled considering only the interactions of gamma particles within the detector. Native activity accounts for 18% of the total gamma spectrum. Figure 5B The "natural total γ" spectrum in the image.

[0099] Table 1 shows various specific distributions uniformly distributed over a 1cm thickness of concrete. 137 Cs and 90 Sr activity ( 137 Cs activity = 90Sr activity), the proportion of native activity in the β and γ spectra.

[0100] Table 1

[0101]

[0102] The results presented in Table 1 show the proportion of natural radioactivity in the measured spectrum. When this proportion is considered too large, for example, for low levels of artificial activity, the acquired spectrum can be corrected to remove the contribution of natural activity. This can be done by estimating the contribution of the object's natural activity to the measured spectrum (β-γ spectrum or β spectrum). Natural activity (assuming it is homogeneous throughout the object) may be:

[0103] - It either stems from the analysis of samples taken from an object or from another object that is considered representative;

[0104] - It either originates from measurements, such as measurements of an object or another object considered representative, performed using high-resolution (e.g., germanium) gamma spectroscopy.

[0105] Then, the contribution of natural activity is subtracted from the spectrum to obtain a βγ spectrum or β spectrum in which the contribution of natural activity is considered negligible.

[0106] Figure 6 The main steps of the present invention are illustrated schematically.

[0107] Step 100 The device 10 is positioned facing the object to be inspected, and the spectrum of radiation emitted by the object is acquired. This can be a β-γ spectrum when the object contains gamma emitters, or a β spectrum when the object contains only pure β emitters.

[0108] Step 110 : Obtain the γ spectrum and correct the γ contribution in the obtained spectrum.

[0109] In step 110 (optional step), screen 15 is placed between the object and the detector. This allows the acquisition of a spectrum representing the γ component of the βγ spectrum acquired in step 110. The β spectrum is formed by subtraction, as described below. Figures 16A to 16C As stated above.

[0110] Step 110 is optional. It is performed when gamma radiation contributes too much to the acquired spectrum.

[0111] Step 120 Correcting natural activity

[0112] In step 120, the contribution of natural activity to the β or βγ spectrum obtained in step 110 or step 100 is estimated. This contribution is subtracted from the spectrum obtained in step 100 or the spectrum obtained in step 110. Step 120 is optional. It is performed when the contribution of natural activity to the obtained spectrum or to the spectrum obtained in step 110 is too large.

[0113] After steps 100 to 120, the input spectrum Sp is obtained. in The input spectrum Sp in The spectrum is either the one obtained in step 100, or the spectrum formed after optional corrections as described in steps 110 to 120. Input spectrum Sp in Input data is formed for the algorithms described in processing steps 130 and 140, which are implemented by processing unit 20.

[0114] Step 130 Identifying radionuclides

[0115] A key aspect of this invention is the combination of two consecutive analysis steps of the input spectrum: the first is an identification step to identify radionuclides present in the object based on the input spectrum, without quantification. The aim is to identify radionuclides present in a pre-determined list. Following this first step, and based on the identification performed, a second step is conducted to estimate the contribution of the identified radionuclides to the spectrum.

[0116] The identification steps are implemented using identification algorithms, which are supervised learning algorithms based on artificial intelligence. Each identification algorithm aims to identify the presence of a radionuclide i in the input spectrum. Each identification algorithm can be, for example, a neural network associated with a radionuclide i, such as a convolutional neural network (CNN). i A Bayesian convolutional network. The subscript 'i' refers to the radionuclide associated with this neural network. Such neural networks are typically used to process structured data, such as images or histograms. A series of convolutional layers enables the extraction of features from the input spectrum. These convolutional layers are fed into multilayer perceptron-type layers, which allow the determination of features related to the convolutional neural network (CNN) based on the features extracted by the convolutional layers. i The probability of the presence of associated radionuclides.

[0117] The output of each identification algorithm is the probability of the radionuclide associated with that algorithm being present in the object. Therefore, the number of identification algorithms equals the number of radionuclides present in the list and potentially present in the object. Input spectrum Sp inThe input data forms the basis of each of these algorithms. For each radionuclide in the list, the output of each algorithm allows the radionuclide to be classified as identified or unidentified based on the input spectrum. Each identification algorithm has been previously trained using training spectra obtained from modeling or acquiring cases where a radionuclide associated with the algorithm is present or absent in an object or a comparable object.

[0118] Each recognition algorithm CNN i The output is the probability P that radionuclide i is present in the object being inspected. i If the value exceeds a certain threshold (e.g., 0.5), then radionuclide i is considered to be present in the object. Conversely, if the value is below the threshold, then radionuclide i is considered not to be present in the object.

[0119] Preferably, the convolutional neural network implements Monte Carlo Dropout, which is equivalent to deactivating certain neurons either randomly or according to a probability law during the training phase and during the use of the neural network. In this way, the network can provide different results each time it is applied to the same input data. This allows for the acquisition of measurement statistics.

[0120] Figure 7A A schematic model of the detector is shown, which is used to identify various radionuclides (in this example, MCNP codes) using the MCNP code. 14 C 36 Cl、 90 Sr、 137 Cs) established various training spectra. The detector was assumed to be 10 mm from the standard source 2. Starting with four initial spectra obtained from modeling the surface distribution of each radionuclide, a training database was created by combining these spectra, which were weighted by various randomly defined parameters: the number of radionuclides in the spectrum, the proportion of each radionuclide, the number of interactions included in the spectrum, the minimum energy, and the maximum energy. In this way, 500,000 training spectra were generated, where the number of interactions considered (counts) ranged from 1. E 3 to 1 E 7.

[0121] Figure 7B The training spectrum for parameterizing the identification algorithm associated with various corresponding radionuclides is shown. Figure 7B The proportion of each radionuclide for each spectrum is shown.

[0122] Figure 8A and Figure 8B This shows a first example of the application of the recognition algorithm. Figure 8A It shows 90 The β spectrum of the Sr source was measured. This spectrum was targeted at a radionuclide. 14 C 36Cl、 90 Sr、 137 Each corresponding recognition algorithm for each definition of Cs was used 100 times. Figure 8B The presence probability (y-axis) for each radionuclide is shown in box plot form. 90 Sr is systematically identified, while 14 C 36 Cl and 137 The median probability of Cs is always less than 0.5.

[0123] Figure 9A and Figure 9B A second example of the application of the recognition algorithm is shown. Figure 9A It shows that it contains 5% 14 C and 95% 36 The β spectrum of a mixture of Cl. This spectrum is for radionuclides. 14 C 36 Cl、 90 Sr、 137 Cs is used 100 times in each definition of each recognition algorithm. Figure 9B The probability of presence for each radionuclide is shown in box plot form. 14 C and 36 Cl is systematically identified, while 90 Sr and 137 The probability of Cs's existence is always less than 0.5.

[0124] The performance of the recognition algorithm is evaluated using a test spectrum, which is obtained by using... 14 C 36 Cl、 90 Sr and 137 The Cs source was obtained by combining four spectra measured in the experiment. By combining the four measured spectra and changing the following characteristics, 10,000 test spectra were generated: relative proportion, activity, number of interactions contained in the spectrum, minimum energy, and maximum energy of the spectrum. These 10,000 test spectra were then processed by an identification algorithm.

[0125] Tables 2, 3, 4, and 5 are confusion matrices. The first column represents the ground truth. The first row represents the result of the identification algorithm. 0 means the radionuclide is not present, and 1 means the radionuclide is present. The matrix value corresponds to the detection rate assigned to that radionuclide. The box corresponding to row 0 and column 1 corresponds to a false positive. The box corresponding to row 1 and column 0 corresponds to a false negative. The boxes corresponding to row 1 and column 1, and row 0 and column 0, correspond to correct detections: the value determined by the identification algorithm corresponds to the ground truth.

[0126] Table 2 ( 14 C)

[0127]

[0128] Table 3 ( 36 Cl)

[0129]

[0130] Table 4 ( 90 Sr)

[0131]

[0132] Table 5 ( 137 Cs)

[0133]

[0134] refer to Figure 8B and Figure 9B The results presented, along with the confusion matrices shown in Tables 2 through 5, demonstrate the reliability of identifying radionuclides in the spectrum by applying various algorithms (each tailored to a specific radionuclide) to the β or βγ spectrum.

[0135] Based on the output of each identification algorithm, determine whether each radionuclide associated with the identification algorithm is present in the object or not.

[0136] Step 140 Deconvolution

[0137] In this step, a deconvolution algorithm is applied to the spectrum to extract the corresponding component associated with each previously identified radionuclide. An important aspect of this step is that the deconvolution is not performed blindly, but rather based on prior knowledge from the identification steps.

[0138] The deconvolution algorithm is based on a deconvolution database containing at least one detector response, i.e., a spectrum modeled for the known activity and known distribution of the radionuclide in the object for each identified radionuclide.

[0139] One possibility is that the distribution of radionuclides in an object is known: it can be assumed to be homogeneous, for example, when the object is a homogenized sample analyzed in a laboratory. When measurements are performed on an activated object, the distribution of radionuclides can be determined by modeling the neutron flux to which the object is exposed. When the object is made of a non-porous material (e.g., a metal), the activity can be assumed to be surface activity.

[0140] When an object is made of a porous material (such as concrete), various activity distributions can be considered. For example, the activity may have a gradient that decreases from the object's surface. For instance, the gradient may have an exponential function form that decreases with depth. This type of profile is typical of contaminant migration. If z is the depth within the object, the activity A(z) distributed according to depth, starting from the object's surface, can be considered to have the following form:

[0141] (1)

[0142] A(0) is the activity at the surface, and λ is an exponential shape factor. λ (whose unit is the reciprocal of the length unit) sets the depth zmax of the activity distribution in the object. If the depth zmax is defined as the activity being a fraction of the surface activity A(0), then... At the depth of time, then:

[0143] (2)

[0144] The definition of λ or zmax allows for the definition of a volume in which the activity is assumed to be concentrated.

[0145] Preferably, the database contains various modeling spectra for each identified radionuclide, corresponding to various corresponding distributions of the radionuclide in the object and the activity of the identified radionuclide. In this way, a database containing spectra representing various activity distributions within the object is obtained for various radionuclides. For example, taking into account exponential gradients, as described in (1) or (2), the database contains modeling spectra for various nuclides corresponding to various parameters λ and zmax.

[0146] The deconvolution algorithm is implemented using representative spectra corresponding to the radionuclides identified in identification step 130. It is understood that prior identification of radionuclides allows for the selection of a modeling spectrum corresponding to each identified radionuclide from the deconvolution database. Deconvolution can then be implemented using a small number of modeling spectra (limited to the identified radionuclides). This allows deconvolution errors, particularly false positives, i.e., concluding the presence of a radionuclide when it is not present. Since a finite number of radionuclides are considered, various distribution profiles can be considered for each identified radionuclide; for example, various activity depths zmax can be considered in the case of a gradient in the descending exponential form as described in (1). Therefore, deconvolution not only allows for the estimation of the activity of each selected radionuclide but also allows for the estimation of its depth of extension within the object. It should be noted that the depth associated with one radionuclide may differ from the depth associated with another.

[0147] The advantage of combining the identification step with the deconvolution step is that, in the deconvolution step, only the spectrum representing the identified radionuclide can be selected. This allows for the consideration of various distributions for each radionuclide. Without selecting a radionuclide, deconvolution considering various distributions would be riskier because too many spectra would need to be taken into account.

[0148] The deconvolution database can contain spectra modeled for various activity depths, as well as spectra obtained by interpolating between modeled spectra (e.g., interpolating between two modeled activity depths).

[0149] To perform deconvolution, the inventors implemented a method based on the definition of a likelihood function and its maximization. The estimated distribution of the activity of each radionuclide corresponds to the distribution that maximizes the likelihood function. The likelihood function can be maximized using the MLEM algorithm, as mentioned in the prior art. Deconvolution can be performed using another method, such as regression, or using a supervised learning algorithm, such as a neural network. In this case, the output of the neural network corresponds to the contribution of each radionuclide to the input spectrum.

[0150] When using the MLEM method, the spectra corresponding to various identified radionuclides in the deconvolution database, as well as the spectra whose combination best matches the input spectrum, are fitted through an iterative process. The spectra are deconvolved in multiple iterations. The iterations continue until a convergence criterion is met, which can be the minimization of a cost function representing the difference between the input spectrum and the spectrum obtained by combining the spectra for the identified radionuclides in the deconvolution database. The cost function can be computed over all energies of the spectrum or over a predetermined region of interest. The region of interest (e.g., a pre-determined region of interest) depends on the variability of the detector's response to each radionuclide with depth: see the description below. Figure 10C and Figure 10D .

[0151] The inventors modeled various spectra corresponding to various activity depths zmax between 0.1 and 500 mm, with exponential gradients as defined in (1) and (2), and activities of 1 Bq / g and solely by 90 Sr is formed. The modeling object is a concrete wall.

[0152] Figure 10A Various modeling spectra are shown. Figure 10B The modeled spectra are shown after normalization by their respective spectral integrals. A spectrum representing a uniform activity distribution across the entire wall thickness (500 mm) was also modeled. The spectrum corresponding to the uniform activity was normalized by its integral. Figure 10C It shows Figure 10B The normalized spectrum is divided by the spectrum corresponding to the uniform activity and normalized by its integral.

[0153] As can be seen, with increasing activity depth, the detector's response tends to correspond to the response of a uniform profile. Figure 10C In the diagram, two dashed lines are drawn, corresponding to deviations from the uniform profile of +5% or -5%. This deviation corresponds to the minimum acceptable deviation, i.e., the minimum deviation required to ensure a good assessment of the activity distribution. Figure 10C The profile shown deviates from the ±5% band corresponding to uniformly distributed activity up to zmax = 40 mm. Therefore, for 90 In the case of Sr, assuming a predetermined decreasing activity gradient, the shape of the β spectrum allows for the determination of the maximum depth zmax of activity in the object, up to zmax = 40 mm. Figure 11 A comparison of modeling spectra considering the following cases is shown:

[0154] - Activity decreases exponentially until the maximum depth zmax is 40 mm;

[0155] - Activity is evenly distributed across the entire wall.

[0156] The two spectra overlapped, which confirms the hypothesis based on... Figure 10C The conclusions drawn.

[0157] Figure 10D It is equivalent to Figure 10C The image, but for 137 Cs activity rather than 90 Sr activity. Figure 10D This was obtained by modeling the βγ spectrum. Figure 10D In the diagram, two dashed lines are drawn, corresponding to deviations from the uniform profile of +5% or -5%. Figure 10D The contour shown deviates from the uniform distribution. 137 The Cs activity is within a ±5% band, up to zmax = 500 mm. Therefore, for 137 In the case of Cs, assuming a predetermined decreasing activity gradient, the shape of the βγ spectrum allows the determination of the maximum depth zmax of activity in the object to be at least 500 mm.

[0158] Figure 10C and Figure 10D It allows defining a spectral region of interest for calculating the aforementioned cost function. For example, one could consider a region exhibiting high variability depending on depth.

[0159] Figures 12A to 12D The diagram shows the fitting of the spectrum (solid line) generated by the MLEM algorithm to the measured spectrum ("Mes" - dashed line) after different numbers of iterations. The measured spectrum corresponds to a surface distribution of 92%. 137 Cs and 8% 90The spectrum composed of Sr is measured to contain 10,000 counts, that is, 10,000 detected pulses. Figures 12A to 12D These correspond to the 1st, 10th, 100th, and 10000th iterations, respectively. It can be seen that with each iteration, the spectrum reconstructed by the MLEM algorithm becomes increasingly closer to the measured spectrum.

[0160] Table 6 summarizes the corresponding percentages determined for each spectrum reconstructed from MLEM for each iteration number. The second row shows the actual percentages. As can be seen, the percentages get closer to the actual values ​​as the iteration process progresses.

[0161] Table 6

[0162]

[0163] The inventors applied the MLEM deconvolution algorithm to 90 Sr and 137 Cs is used for various depths zmax, assuming an exponential descent gradient, and applied to 90 Sr and 137 Various ratios of Cs activity are listed in Table 7.

[0164] Table 7

[0165]

[0166] The β, γ, and βγ spectra were modeled for each of these configurations using MCNP, and then MLEM deconvolution was applied. The deconvolution results are summarized in Tables 8 (Configuration 1), 9 (Configuration 2), 10 (Configuration 3), and 11 (Configuration 4).

[0167] Each table provides the following items: zmax (in mm), standard deviation (in mm) associated with the determined zmax, and activity A ( 137 Cs) or A( 90 Sr) (in Bq), and the standard deviation (in Bq) associated with the determined activity.

[0168] Table 8

[0169]

[0170] Table 9

[0171]

[0172] Table 10

[0173]

[0174] Table 11

[0175]

[0176] The results presented in Tables 8 to 11 demonstrate the reliability of the algorithm, especially when considering... 137 Cs considers the βγ spectrum and when for 90 When considering the β spectrum, Sr is used in configuration 4 (for...). 90 Sr's zmax = 400mm corresponds to an activity depth exceeding the target range. 90 Sr defines a maximum depth of 40mm, as per reference. Figure 10C The above describes the quantization using the β spectrum. 137 Cs activity depth or activity, when 90 Sr activity is at least 137 A Cs activity that is 3 times the normal activity (as is the case with configurations 2 and 4) may cause an error.

[0177] Therefore, in 90 Sr and 137 In the case of a Cs mixture, considering that both radionuclides are fission products and frequently encounter each other in spent fuel processing plants or in cases of radioactive contamination related to spent fuel, it seems optimal to include them in the βγ spectrum.

[0178] Combined experiments have confirmed the invention’s ability to quantify the activity of identified radionuclides and estimate the activity depth of each of these radionuclides.

[0179] Comparison with the case where no recognition algorithm is implemented

[0180] Figure 8A and Figure 9A The spectrum shown was used as the input spectrum for the deconvolution algorithm, without prior identification of radionuclides, i.e., no identification algorithm was implemented. Regarding the corresponding... 90 The spectrum of Sr activity (see) Figure 8A The MLEM algorithm yielded the following activity percentage: 0% 14 C, 4% 36 Cl, 90% 90 Sr, 6% 137 Cs. For the corresponding 14 C and 36 The spectrum of Cl activity (see) Figure 9A The MLEM algorithm yielded the following activity percentage: 5%. 14 C, 90% 36 Cl, 0% 90 Sr, 5% 137 Cs.

[0181] Experimental test

[0182] Steps 110 through 140 are performed on graphite samples S, which are collected by coring in a transverse direction to the horizontal channel CH, used for loading and unloading dye from the gas-cooled graphite-moderated reactor core. Each sample extends between a side F1, called the "channel side" (adjacent to the fuel channel and facing the interior of the fuel channel), and an opposite side F2, called the "core side" (facing the graphite moderator of the reactor). Figure 13 The diagram schematically shows the channel CH and the location of the core sample taken from which sample S was extracted.

[0183] Each graphite sample is cylindrical, with a diameter of 15 mm and a thickness of 20 mm. The detector is placed 25 mm from one side of each sample.

[0184] A training spectrum and deconvolution database were created, taking into account the main radionuclides that might be measured:

[0185] Activated products: 14 C 36 Cl、 60 Co、 133 Ba、 152 Eu、 154 Eu;

[0186] Possible fission products: 137 Cs and 90 Sr.

[0187] The purpose of the analysis was to verify the presence of fission products indicating possible shell rupture. To assess potential contamination from the fission products, four recognition neural networks were parameterized to process them. 137 Cs and 90 Different minimum activity levels of Sr. The minimum activity is equal to 0.1 Bq.g. -1 1 Bq.g -1 10 Bq.g -1 and 100 Bq.g -1 In the following text, the neural networks are denoted as A, B, C, and D, respectively.

[0188] These four networks were trained using 50,000 modeling spectra, where 50% of the spectra were considered to contain contributions from fission products, and 50% were considered not to contain contributions from fission products. For spectra containing contributions from fission products, 137 Cs / 90The Sr ratio was randomly selected between 0.25 and 4. The distribution depth was also randomly selected, such that zmax ranged from 0.1 mm to 5 mm. Shallow depths are reasonable given that contamination from fission products is assumed to be dry contamination. The presence of gamma background noise was also modeled, with its level set based on a reference measurement considered to represent the background noise during the measurement period. The measurement time was defined for each training spectrum. Random background noise contributions were added to the modeling spectrum, taking into account the measurement time, to simulate statistical fluctuations in the background noise. Activation products ( 14 C 36 Cl、 60 Co、 133 Ba、 152 Eu、 154 The activity of Eu was determined based on graphite activation calculations.

[0189] 80% of the spectrum was used for training. 10% of the spectrum was used for validation to fit the model. 10% of the spectrum was used for testing.

[0190] Several samples collected along two different fuel channels were analyzed. Each spectrum measured on the graphite sample was analyzed one hundred times using four neural networks. Analysis was performed on both the core side and the channel side for each sample.

[0191] Figures 14A to 14D This shows the results for various samples when used on the core side (side F2) ( Figure 14A and Figure 14C ) or the passageway side (side F1) ( Figure 14B and Figure 14D When measuring the spectrum, the outputs from four neural networks are used as a function of the distance from the center of the reactor. The y-axis corresponds to the probability of contamination (between 0 and 1), which is calculated by considering 100 analyses of each neural network. Figures 14A to 14D It shows 137 Cs and 90 The probability of Sr's existence is derived from those neural network outputs that achieved a significant probability out of 100 iterations. The arrows indicate the neural networks that achieved a significant probability.

[0192] The x-axis corresponds to the position of the sample relative to the origin (corresponding to the middle of the reactor) (unit: cm). Figure 14A and Figure 14B The probability obtained for a sample acquired from the first channel (labeled 36-17C) is shown as a function of the outputs of four sample neural networks (each used 100 times). Figure 14C and Figure 14D The probability obtained for samples acquired from the second channel (labeled 19-13C) is shown as a function of the outputs of four neural networks.

[0193] As can be seen, only one sample collected from channel 36-17C has a possible but low probability of being positive. 137 Cs+ 90 Sr activity. In this channel, neural networks C and D conclude that there is no activity.

[0194] Analysis of most of the spectra measured on the channel side in channels 19-13C suggests the presence of... 137 Cs+ 90 Sr's conclusion.

[0195] On one sample, taken at a distance of 15 cm from channel 36-17C, on the channel side, detection was performed using two neural networks (A and B). 137 Cs+ 90 Sr activity. The spectrum was deconvolved, considering both the output of the recognition neural network and the output without considering it. Table 12 summarizes the results with and without considering the output of the recognition neural network. 137 Cs and 90 The result obtained by performing the deconvolution algorithm when Sr is present.

[0196] Table 12 presents the evaluation of various radionuclide activities A (units Bq) and the associated relative uncertainty (Σ) when the deconvolution algorithm is implemented without (I) and with (II) prior implementation of the recognition neural network. On this sample, neural networks A and B are activated 100 times, yielding results at 50% and 70% accuracy, respectively. 137 Cs and 90 Sr exists (see...) Figure 14B The conclusions are as follows. The last column shows the relative differences in activities estimated in (I) and (II). 137 Cs and 90 The inclusion of Sr allows for adjustments to the estimated activity value for activated products.

[0197] Table 12

[0198]

[0199] The deconvolution algorithm was implemented on samples taken at a distance of 120 cm along channel 19-13C, on the channel side. On this sample, neural networks A through D were activated 100 times, and the results were obtained under 100% accuracy. 137 Cs and 90 Sr exists respectively (see) Figure 14D The conclusion is as follows.

[0200] The deconvolution algorithm minimizes the cost function, corresponding to:

[0201] - 90 Sr and 137Cs activities were 246±46 Bq and 296±31 Bq, respectively.

[0202] - 90 Sr and 137 The activity depths of Cs are less than 200 μm and 500 μm, respectively, which confirms the hypothesis of surface contamination.

[0203] The sample was characterized using high-resolution gamma-ray spectroscopy, which is considered a reference method. The measured values... 137 The Cs activity is 358 Bq with a relative uncertainty of 30%.

[0204] Considering the error range, the βγ energy dispersive spectroscopy method is consistent with the reference method.

[0205] Figure 15 The measured βγ spectrum is shown, and 137 Cs and 90 Sr and the activation product (act) on the input spectrum Sp in The contributions stem from deconvolution algorithms.

[0206] As previously stated, the present invention can be implemented in the β spectrum under strong γ radiation. For this purpose, the following measurements were performed:

[0207] - The βγ spectrum of the object;

[0208] - The gamma spectrum in the case where the screen is between the detector and the object.

[0209] The β spectrum is obtained by subtracting the γ spectrum from the βγ spectrum, possibly taking into account differences during the acquisition process.

[0210] Figures 16A to 16C This illustrates this possibility. Figure 16A The measured βγ spectrum is shown. Figure 16B The measured gamma spectrum is shown. Figure 16C The β spectrum is shown, calculated by extracting the difference between the βγ spectrum and the γ spectrum.

[0211] Although described in relation to spectra acquired in a non-destructive manner, this method can be generalized to the analysis of spectra with non-negligible beta components. It then pertains to laboratory measurement methods, such as liquid scintillation. In this case, the object is a sample placed in front of a beta spectrometer.

Claims

1. A method for characterizing an object, said object comprising at least one radioactive nuclide emitting beta radiation, said method comprising: - a) A detector is placed facing the object, the detector being configured to acquire a spectrum representing the distribution of energy released in the detector by radiation emitted by the object; - b) During the acquisition, the radiation emitted by the object is detected by the detector, and the spectrum of the detected radiation is acquired; - c) Based on the spectrum of radiation detected by the detector, an input spectrum is formed, the input spectrum containing a β component, which corresponds to the distribution of energy released by the β radiation in the detector; - d) Apply an identification algorithm associated with a radionuclide to the input spectrum, the identification algorithm being configured to determine the presence of the radionuclide associated with the identification algorithm in the object, and repeat step d) by implementing different identification algorithms for various radionuclides; - e) Identify each radionuclide contained in the object according to d); - f) Apply the input spectrum deconvolution algorithm to estimate the contribution of each radionuclide identified in e) to the input spectrum; - g) For each radionuclide identified in e), estimate the activity of the radionuclide and / or the depth to which the radionuclide extends in the object, based on the contribution of the radionuclide to the input spectrum estimated in f). Steps d) to g) are performed by the processing unit based on the input spectrum.

2. The method according to claim 1, wherein, In step d), each identification algorithm is an artificial intelligence identification algorithm associated with each radionuclide, and at least two different radionuclides are associated with two different corresponding identification algorithms.

3. The method of claim 2, wherein each recognition algorithm is a neural network.

4. The method of claim 1, wherein the deconvolution algorithm is based on a deconvolution database containing at least one spectrum representing each radionuclide identified in e).

5. The method according to claim 4, wherein - For a given radionuclide, the deconvolution database contains various spectra representing the various distributions of the radionuclide in the object; Step g) includes determining the distribution of the radionuclide in the object.

6. The method according to claim 5, wherein: - For a given radionuclide, the deconvolution database contains various spectra representing the radionuclide at various depths in the object, from the surface of the object facing the detector; Step g) includes determining the depth to which the radionuclide extends in the object.

7. The method of claim 1, wherein at least one radionuclide associated with the identification algorithm is a pure beta emitter.

8. The method of claim 1, wherein the detector comprises an organic scintillator material for detecting radiation emitted by the object.

9. The method of claim 1, wherein the detector comprises a volume of an inorganic scintillator or semiconductor with a thickness of less than 10 mm, which is placed facing the object, the thickness being measured in a direction perpendicular to the object.

10. The method of claim 1, wherein the object has natural activity, step c) comprising: - Estimate the natural activity spectrum of the object; - Subtract the natural activity spectrum of the object from the spectrum obtained in step b) to form the input spectrum.

11. The method according to any one of the preceding claims, wherein the detector includes a movable screen configured between the detector and the object, the method comprising: - Obtain the background spectrum when the screen is positioned between the detector and the object; - In step c), the background spectrum is subtracted from the spectrum obtained in step b) to form the input spectrum.

12. A detection device comprising a detector configured to acquire a spectrum of beta radiation emitted by an object, the spectrum representing the distribution of energy released in the detector during ionizing radiation interaction, the device comprising a processing unit configured to implement steps d) to f) of the method according to any one of the preceding claims.

13. The detection device of claim 12, wherein the detector comprises an organic scintillator material for detecting radiation emitted by the object.

14. The detection device according to claim 12, wherein the detector comprises an inorganic scintillator or semiconductor with a thickness of less than 10 mm.

15. The detection device according to any one of claims 12, wherein the detector includes a movable screen configured between the detector and the object.

Images