Patent Application
20210341400
The present application generally relates to systems and methods for interpreting high-energy interactions on a sample. In particular, this application relates to the quantification and/or classification of the input from a detector as is produced from analytical instrumentation, in particular the generation of quantification and/or classification models generated via machine learning techniques.
The miniaturization of analytical instrumentation has enabled non-destructive analysis of objects, often by way of using portable systems. This shift contrasts with the past use of this type of equipment in which analytes were destroyed during the analysis process, in some instances by being prepared into matrices such as solutions and fused glass beads. Destructive sample preparation enabled quantification based on models which used the intensities of diagnostic fluorescent peaks, reflectively scattered angles, or emitted radiation following ionization. These signals are defined and interpreted via empirical calibration or estimation through physical and geometric parameters. This approach has largely been adopted for non-or minimally-destructive analyzers, though this type of analysis is greatly complicated by sample heterogeneity, chemical transformation, and uneven surfaces. Performance of non-destructive analyzers is thus generally inferior relative to their destructive counterparts.
These analytical instruments typically have the following three components: a high energy light or particle source, a geometric relationship with an analyte, and a detector capable of measuring the light or particle's interaction with the analyte. These detectors function differently and have different requirements, though each is capable of producing a spectrum with an energy/wavelength/geometry/degrees and response. These detector inputs are by necessity acquired via different instruments, components, or geometries.
This application generally describes systems and methods for interpreting high-energy interactions on a sample. In particular, this application describes an analysis method that comprises impinging radiation from a source on an analyte, detecting the energy interactions resulting from the impinging radiation using a radiation detector, adjusting the signal from the radiation detector using a machine learning module to emphasize specific parts of the detector signal, training the machine learning module in a supervised or unsupervised manner, producing quantitative and qualitative models using the machine leaning module, and then applying the machine learning module to additional energy interactions. The signal received by the detector can be preprocessed to emphasize specific parts of the detector signal, which is then mapped to a machine learning module for training in a supervised or unsupervised manner. The quantitative and qualitative models derived from this training can be applied to new detector inputs from the same or similar instruments.
The following description can be better understood in light of the Figures which show various embodiments and configurations of the systems and methods for interpreting high-energy interactions on a sample.
Together with the following description, the Figures demonstrate and explain the principles of the structures, methods, and principles described herein. In the drawings, the thickness and size of components may be exaggerated or otherwise modified for clarity. The same reference numerals in different drawings represent the same element, and thus their descriptions will not be repeated. Furthermore, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the described devices.
The following description supplies specific details in order to provide a thorough understanding. Nevertheless, the skilled artisan will understand that the described electronic devices can be implemented and used without employing these specific details. Indeed, the described systems and methods for using touchpads with multiple surfaces can be placed into practice by modifying the described systems and methods and can be used in conjunction with any other apparatus and techniques conventionally used in the industry. For example, while the description below focuses on x-ray fluorescence, they could be used with other devices and systems, including atomic emission spectroscopy. As well, while this description focuses on photon energy interactions, the systems and methods described herein could be used with electron or proton energy interactions.
In addition, as the terms on, disposed on, attached to, connected to, or coupled to, etc. are used herein, one object (e.g., a material, element, structure, member, etc.) can be on, disposed on, attached to, connected to, or coupled to another object—regardless of whether the one object is directly on, attached, connected, or coupled to the other object or whether there are one or more intervening objects between the one object and the other object. Also, directions (e.g., on top of, below, above, top, bottom, side, up, down, under, over, upper, lower, lateral, orbital, horizontal, etc.), if provided, are relative and provided solely by way of example and for ease of illustration and discussion and not by way of limitation. Where reference is made to a list of elements (e.g., elements a, b, c), such reference is intended to include any one of the listed elements by itself, any combination of less than all of the listed elements, and/or a combination of all of the listed elements. Furthermore, as used herein, the terms a, an, and one may each be interchangeable with the terms at least one and one or more.
In the systems and methods described herein, the radiation used can be any ionizing radiation that will create photon-electron interactions. In some embodiments, the radiation can be electromagnetic radiation, including radio waves, microwaves, infrared, ultraviolet, x-rays, and gamma rays. In other embodiments, the radiation can be particle radiation, such as alpha radiation, beta radiation, electrons, protons, and neutron radiation.
The radiation source 5 and filter/collimator module 8 can vary in the geometry 9 so that the radiation beam 7 has a desired intensity and impingement angle on the analyte 11. The radiation beam 7 emerging from the radiation source 5 can pass through a barrier 10 that may or may not attenuate or otherwise influence the beam 7. An analyte (or sample) 11 is irradiated by the radiation beam 7 and produces a post-analyte beam of radiation 12 from the consequences of ionization within it. This post-analyte beam 12 passes through a barrier 10 that may or may not attenuate or otherwise influence the post-analyte beam 12. The post-analyte beam 12 enters the radiation detector (or detector) 13 through a detector window 14 that may or may not attenuate it. The detector 13 can have variable geometry 15, which influences the angle and intensity of the incoming post-analyte beam 12. The detector 13 is powered by a power source 1 through an electrical connection 16. The power source 1 may or may not be shared with the radiation power generator 3 and/or radiation source 5. The detector 13 produces a signal 17 which is passed on to a machine learning module 18, which interprets the detector signal 17 and can send quantitative or qualitative results 20 via connection 19.
In some configurations, the radiation source 5 can contain any source that generates and emits X-rays, including a stationary anode X-ray source, a micro-focus x-ray source, a rotating anode x-ray source, and/or a fluoroscopic X-ray source. In other configurations, the radiation source 5 can emit particles such as electrons or protons. In these configurations, the detector 13 can contain any detector that detects X-rays, including a silicon-drift detector, silicon pin diode, silicon lithium detector, germanium detector, cadmium-telluride detector, monochromater, polychromater, photomultiplier, charged-doubled device, image intensifier, a CMOS camera and/or a digital flat panel detector. In some configurations, the X-ray source and x-ray detector can be made modular so that different sizes and types of X-ray sources and x-ray detectors can be used in the x-ray device 100. In other configurations, the system can contain multiple x-ray sources and/or multiple x-ray detectors.
Other embodiments of the systems described herein are shown in
Yet other embodiments of the systems described herein as shown in
The systems and methods described herein can be implemented in connection with any electronics, including the computer system illustrated in
Computer system 300 includes a processor 301 (or multiple processors) such as a central processing unit (CPU), graphical processing unit (GPU), application specific integrated circuit (ASIC), or a field programmable gate array (FPGA). The computer system 300 may also contain a memory component (or memory) 303 and a storage component (storage) 308 that communicate with each other and with other components via a bus 340. The bus 340 may also link one or more displays 332, one or more input devices 333 (e.g., a keypad, a keyboard, a mouse, a stylus, etc.), one or more output devices 334, one or more storage devices 335, and various non-transitory, tangible computer-readable storage media 336 with each other and with the processor 301, the memory 303, and the storage 308. All of these components can communicate directly or via one or more interfaces or adaptors to the bus 340. For example, the various non-transitory, tangible computer-readable storage media 336 can interface with the bus 340 via storage medium interface 326.
Processor(s) 301 optionally contains a cache memory unit 302 for temporary local storage of instructions or data. Processor(s) 301 can also execute computer-readable instructions stored on at least one non-transitory, tangible computer-readable storage medium. Computer system 300, as a result of the processor(s) 301, may also execute software embodied in one or more non-transitory, tangible computer-readable storage media, such as memory 303, storage 308, storage devices 335, storage medium 336 (i.e., read only memory or ROM), or the machine learning module 60. The non-transitory, tangible computer-readable storage media may store software that implements particular embodiments, and processor(s) 301 may execute the software.
Memory 303 may implement and/or execute the software from one or more other non-transitory, tangible computer-readable storage media (such as mass storage device(s) 335, 336) or from one or more other sources through any interface, such as network interface 320. The software may cause processor(s) 301 to carry the process(es) or step(s) of any process described herein. Executing such processes or steps may include defining data structures stored in memory 303 and modifying the data structures as directed by the software. In some embodiments, an FPGA can store instructions for carrying out the functionality while in other embodiments, firmware includes instructions for carrying out any functionality described herein.
The memory 303 may include various components (e.g., non-transitory, tangible computer-readable storage media) including random access memory component (e.g., RAM 304 whether static or dynamic RAM), a read-only component (e.g., ROM 305), and any combinations thereof ROM 305 may communicate data and instructions unidirectionally to processor(s) 301, and RAM 304 may act to communicate data and instructions bidirectionally with processor(s) 301. ROM 305 and RAM 304 may include any suitable non-transitory, tangible computer-readable storage media. In some instances, ROM 305 and RAM 304 may include non-transitory, tangible computer-readable storage media for carrying out the methods described herein. A basic input/output system 306 (BIOS), including basic routines to transfer information between elements within computer system 300, may be stored in the memory 303.
Fixed storage 308 can be connected to processor(s) 301, optionally through storage control unit 307. Fixed storage 308 provides data storage capacity and may also include any suitable non-transitory, tangible computer-readable media described herein. Storage 308 may be used to store operating system 309, executable commands (EXEC) 310, data 311, API applications 312 (application programs), and the like. For example, multiple instances of the storage 308 could be used for storage by the machine learning module described herein. In some configurations, storage 308 can be a secondary storage medium (i.e., a hard disk) that is slower than primary storage (i.e., memory 303). Storage 308 can also include an optical disk drive, a solid-state memory device (e.g., flash-based systems), or a combination thereof. Information in storage 308 may also be incorporated as virtual memory in memory 303.
In some configurations, storage device(s) 335 may be removably interfaced with computer system 300 (e.g., via an external port connector (not shown)) via a storage device interface 325. Thus, storage device(s) 335 and an associated machine-readable medium may provide nonvolatile and/or volatile storage of machine-readable instructions, data structures, program modules, and/or other data for the computer system 300. For example, software may reside completely or partially within a machine-readable medium on storage device(s) 335. In another example, software may reside, completely or partially, within processor(s) 301.
Bus 340 connects a wide variety of subsystems and/or components in the computer system 300. Bus 340 may encompass one or more digital signal lines serving a common function. Bus 340 may also comprise any type of bus structures including a memory bus, a memory controller, a peripheral bus, a local bus, and any combinations thereof, using any of a variety of bus architectures. Such architectures include an Industry Standard Architecture (ISA) bus, an Enhanced ISA (EISA) bus, a Micro Channel Architecture (MCA) bus, a Video Electronics Standards Association local bus (VLB), a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, an Accelerated Graphics Port (AGP) bus, HyperTransport (HTX) bus, serial advanced technology attachment (SATA) bus, and/or any combinations thereof.
Computer system 300 may also include an input device(s) 333. A user of computer system 300 may enter commands and/or other data into computer system 300 via input device(s) 333. Examples of an input device(s) 333 include an alpha-numeric input device (keyboard), a tracking device (mouse), a touchpad, touchscreen, a joystick, a gamepad, an audio input device (microphone), an optical scanner, a video or still image capture device (e.g., a camera), and any combinations thereof. Input device(s) 333 may interface with bus 340 via any number of input interfaces 323 including serial, parallel, game port, USB, FIREWIRE, or any combination thereof.
When computer system 300 is connected to a network 330, the computer system 300 may communicate with other electronic devices, such as mobile devices and enterprise systems, that are connected to network 330. Communications to and from computer system 300 may be sent through a network interface 320 which may receive incoming communications in the form of one or more packets (such as Internet Protocol (IP) packets) from network 330. Computer system 300 may then store the incoming communications in memory 303 for processing. Computer system 300 may also store outgoing communications in the form of one or more packets in memory 303 and communicate them to network 330 via network interface 320. Examples of the network interface 320 include a network interface card, a modem, and any combination thereof. Examples of a network 330 (or network segment 330) include a wide area network (WAN) (e.g., the Internet, an enterprise network), a local area network (LAN), a telephone network, virtual private network (VPN), a direct connection between two computing devices, and any combinations thereof. The network 330 may employ any wired and/or a wireless mode of communication.
Information and data can be displayed through a display(s) 332. Examples of a display 332 include a liquid crystal display (LCD), an organic liquid crystal display (OLED), a cathode ray tube (CRT), a plasma display, and any combinations thereof. The display 332 can interface to the processor(s) 301, memory 303, fixed storage 308, as well as other devices (i.e., input device(s) 333) via the bus 340. The display 332 can be linked to the bus 340 via a video interface 322, and transport of data between the display 332 and the bus 340 can be controlled via graphics controller 321. The results presented by the machine learning module 60 may also be displayed by the display 332.
The computer system 300 may include one or more other peripheral output devices 634 including an audio speaker, a printer, and any combinations thereof. Such peripheral output devices may be connected to the bus 340 via an output interface 324. Examples of an output interface 324 include a serial port, a parallel connection, a USB port, a FIREWIRE port, a THUNDERBOLT port, and any combinations thereof.
Computer system 300 may also provide functionality as a result of logic hardwired or otherwise embodied in a circuit, which may operate in place of or together with software to execute the process(es) or step(s) described herein. Software in the computer system 300 may encompass logic, and reference to logic may encompass software. As well, the non-transitory, tangible computer-readable medium may encompass a circuit (such as an IC) storing software for execution, a circuit embodying logic for execution, or both.
Within computer system 300, the same reference characters are used to refer to terminals, signal lines, wires, etc. and their corresponding signals. The terms signal and wire can represent one or more signals, e.g., the conveyance of a single bit through a single wire or the conveyance of multiple parallel bits through multiple parallel wires. And each wire or signal may represent unidirectional or bidirectional communication between two or more components connected by a signal or wire.
The various logical blocks, modules, and circuits described herein may be implemented or performed with a general purpose processor, a digital signal processor (DSP), ASIC, GPU, a FPGA or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor or may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other similar configuration.
In some configurations, the detector input can be converted to an output using the embodiments illustrated in
Other embodiments of the systems and methods described herein are illustrated in
Other embodiments of the systems and methods described herein are illustrated in
Other embodiments of the systems and methods described herein are illustrated in
The systems and methods described herein can determine the chemistry of a sample by measuring the fluorescent (or other high energy) x-ray emitted from it when excited or affected by primary x-rays. Each of the elements present in the sample produce a set of characteristic fluorescent x-rays (a fingerprint) that is unique for that specific element. But the system must be calibrated in order to ensure the results are accurate because each instrument, when manufactured, produces a different photon flux in addition to small differences in geometry that affect the final output. The intent of the calibration is to a) force coherence onto a population of machines to produce similar and reliable results and b) to convert the spectral data into actionable information that can be relied on by the operator of the instrument.
There have been numerous methods used for XRF instrument calibration. Some of these methods include comparing the XRF results against a known standard. Other methods have compared the XRF results with the results from other forms of analytical chemistry. But these methods were deficient since they are highly sensitive to matrix types, and are limited to values which are directly manifested in the spectrum, such as the K-alpha peak of Iron (Fe), the L-beta peak of Lead (Pb). In these methods, the measurement of peak intensity or amplitude are related directly to concentration of the value within the analyte. But these methods are subject to important limitations, including a) detection limit, b) a quantification protocol capable of accurately representing the composition of the analyte, and c) limitation to variables with diagnostic fluorescence or reflectance. Because of these limitations, these calibration techniques are specific to a given instrument. Any other instrument could not use the same calibration technique, but would rather require a separate calibration technique to be performed to produce useful information. In other words, these calibration techniques are instrument specific and can't be used across multiple instruments even if those instruments are similar or are even if they are of the same model type.
Indeed, instrument-specific calibration approaches encounter additional limitations besides those mentioned in the previous paragraph. The first limitation is that for empirical analysis, standards must be manually analyzed for every intended material type for each instrument. Second, the definitions for each variable must be predetermined (e.g., defined externally from the data). And third, any algorithmic corrections risk introducing covariance, in which relationships between the predictor variables bias the result in a manner difficult to detect. The systems and methods described herein improve on all three of these limitations. First, the machine learning model can be applied to more than one instrument, making it a scalable approach. Second, definitions for variables can be determined from all detector inputs, leading to optimal results. And third, randomized cross validation may be used to eliminate or significantly reduce the influence of covariance.
An additional limitation of calibration is that the detector input to be analyzed can be varied and complex. Examples of the detector inputs which could be used include ionizing radiation with fluorescence peaks of inner electron orbitals, outer electron orbitals, electron orbital changes, coherent scattering, Bremsstrahlung scattering, Compton (inelastic) scattering, Rayleigh (elastic scattering), sum peaks, or even escape peaks from the detector. Some of these detector inputs 103 are shown in
One of the primary challenges of working with energy spectra as a detector input is their complexity. Specific geometries and energies of one part of the energy spectra are representative of a given electron energy/alignment, but they are convoluted with other parts of the energy spectra. In XRF, the energy peaks for an element like Dysprosium (Dy) can be obscured by the energy peaks for Iron (Fe). Iron (Fe) has 2 K emissions in XRF which overlap with Dysprosium (Dy) L emissions, but up to 18,000 in atomic emission spectroscopy (AES) which overlap with many other elemental lines, such as 402 in
The advent of machine learning methods helps meet this primary challenge since they have the ability to completely analyze the spectrum produced by detectors. Machine learning techniques such as forest regressions, forest classification, gradient boosting, and neural networks, among others, present methods to impartially analyze the whole spectrum produced by detectors and integrate it with other existing information. However, machine learning algorithms cannot improve upon the original data presented to them. Pre-processing of data has the potential to emphasize key features that machine learning can use to draw generalizable rules for interpreting data. The systems and methods described herein in connection to
These machine learning techniques are also deficient in other ways besides the power requirements. First, these machine learning techniques can be much more vulnerable to outliers, even though machine learning cross validation iterations can reduce this risk. Second, these machine learning techniques can also be constrained by co-variance, where predictors in a model themselves are correlated and lead to inflated model metrics and overfitting. Third, these machine learning techniques can be constrained by a-priori definitions of how a given independent variable relates to the input data from a detector; whereas the machine learning module described herein can derive its own definitions from the data itself. This leads to much higher accuracy and flexibility in machine learning approaches. The machine learning module described herein also uses data preprocessing which can prepare the data for the analysis, increasing the efficiency of the machine learning process and decreasing the time and computer power needed. Examples of the data preprocessing includes normalizing by time of acquisition, proportional representation of a given detector input relative to all other detector inputs, or normalization to a region of interest in the spectrum based on the properties of the target measurement, as illustrated in
The machine learning module described herein presents a different relationship with traditional implementations for a given variable. An example of a traditional use of quantification algorithms in in x-ray spectrometry is as follows:
C
i
=r
0
+I
i(ri+Σrrn*Irn)
where Ci represents the concentration of element, r0 is the intercept/empirical constant for element i, ri is the slope/empirical coefficient for intensity of element i, rn is the slope/empirical constant for effect of element n on element i, Ii is the net intensity of element I, and In is the net intensity of element n. This is known as the Lucas-Tooth and Price equation and an assumption underlying this equation is that the independent variables are pre-determined (Ii . . . n) in their definition with a specified energy range. For example, Iron (Fe) K-alpha fluoresces at 6.4 keV, its peak definition (IFeKα1) could be 6.4±0.2 keV, or based on a 68% area underneath a Gaussian distribution at 6.4 keV.
The machine learning modules differs from this equation and does not use a pre-determined definition for a variable [I] like Fe. Instead, the machine learning module uses in some configurations an equation for a given iteration that resembles:
C
i
=re
i
×re
i+1
× . . . re
i+n
where ei represents a given energy which could be calculated from detector output and r represents a given weight or series of weights and/or other modifications for a particular energy (e). This equation can be iterated using randomly-selected known values to generate a definition for a given element, such as Fe. However, random selection does not require that variables are not positional, e.g., that their adjacent points in the spectrum are not part of the information transmitted to the model. The definition of Fe can then be formed relative to the nature of a specified data set with an iterative approach, rather than selected a-priori. And the interaction of high energy photons with the analyte to produce (ei) is not restricted to just fluorescence. The photon-analyte interactions that can be analyzed by the machine learning modules described herein can include Compton (inelastic) scattering, Rayleigh (elastic) scattering, Bragg (coherent) scattering, random scattering indirectly related to Bremsstrahlung radiation from the source, and other interactions not directly produced due to fluorescence. Indeed, some embodiments can use a regression framework to evaluate variable (ei) importance. Other embodiments, however, could use variants that include decision trees, support vector machines (SVMs), k-nearest neighbor classifiers, and other ensemble machine learning techniques related to detector output and existing information about a given material. The object of these embodiments is similar to some conventional quantification and/or classification methods that create generalizable models that can be employed on materials of a given matrix.
In the systems and methods described herein, the matrix of an analyte material can be a gas, liquid, or solid material. As well, the interaction of photons can be variable within and between atoms, diagnostic to either molecules or elements respectively depending on the detector employed. The integration of multiple detector inputs in the systems and methods described herein allows for automated serialized or parallel determination of qualitative or quantitative values for the detector output, depending on the particular procedure employed. In some configurations, though, the systems and methods described herein do not require either serialization or parallel analysis.
In some configurations, the machine learning module described herein can use training of algorithms on a particular data set. This training requires detector input from samples with known or estimated composition or qualities related to the final output values obtained using the machine learning module. As an example, a family of standards determined via a method such as inductively-coupled plasma spectroscopy or determined via an institution such as the National Institute of Standards and Technology could be analyzed given a known instrumental method(s). The detector inputs could then be systematically analyzed datum by datum of raw, transformed, or otherwise preprocessed analytic output from given sensor/sensors. This training can use multiple decision trees informed by randomized cross-validation against given raw analytic output against the known values, a single tree and/or linear model that is updated through gradient boosting or another iterative technique, and/or third a series of weights to adjust the combination of variables. This procedure used in the training could be replicated tens, hundreds, thousands, or even millions of times. The raw analytic detector input can be weighted so that the portions of the detector input with the highest predictive power of the known value can be prioritized in future estimations from unknown detector inputs in the machine learning modules described herein.
Some methods to obtain the quantitative output 90 from the machine leaning module as shown in
The results from the machine learning module can include a diagram of detector input influence 403, as illustrated in
The machine learning module identifies both direct and indirect signals of a given value in the XRF instrument(s). As such, a quantification model for particular element [i.e., Sodium (Na)] output from detector input could include a direct proxy for that element, such as the Na K-alpha emission alongside indirect measurements of other class I elements such as the K-alpha and K-beta emissions of Potassium (K) and Rubidium (Rb) as well as a change in the Compton (inelastic) scattering of x-rays of the analyte, even though it is also possible that an element like Na can be successfully infected without regard to its K-line fluorescence. The particular method used for this quantification model will be determined by the machine-learning module, in either a supervised or unsupervised mode, where supervised machine learning includes instances where external information is provided (known or estimated values) to guide analysis (e.g., standards) while unsupervised machine learning includes pattern recognition in the data.
The machine learning module can be deployed either locally on a computer connected directly with the XRF equipment, such as a computer 300 shown in
The machine learning module can also be connected to any type of electronic device with a wired or a wireless connection. In these embodiments, the machine learning module can transmit data to the desired electronic device, such as a computer 300 shown in
Thus, the systems and methods described herein interpret high energy photon interactions in a sample via a detector and use that information as an input for a machine-learning module which can produce as its output a variety of qualitative and quantitative models. The high energy photon light interacts with the sample, producing a number of interactions including Rayleigh (elastic) scattering, Compton (inelastic) scattering, Bragg (coherent) scattering, fluorescence, and other photon-fermion interactions. These interactions are detected by the detector which produces a series of counts that are interpreted by the machine learning module. This module can, with existing data about the materials in the sample, produce qualitative classes or quantitative units as an output. The existing data can comprise compositional data about the analyte, membership of the analyte in a class, or any other qualitative or quantitative attribute. The counts are determined by a counting mechanism within the detector that provides data about each element in the sample in a continuous interval, which can be used either in a full or summarized form by the module. The high energy photon interactions are determined by the machine learning module to identify relationships in the detector output that include random forests, gradient boosting, linear support vector machines, neural networks, and k-nearest neighbors, among others. The fluorescent peaks shown in the output form the module are not pre-determined, but rather guided by relationships within the input from the detector. The manifestation of the target variable output through fluorescence is not required.
The geometric relationship between the radiation source, analyte, and detector in the XRF systems does not depend on any one particular configuration and can be varied from instrument to instrument. A key difficulty in analyzing with any XRF instrument, but particularly for those which engage in non-destructive analysis, is the calibration of multiple units where the configuration between these three components varies from instrument to instrument. A manufacturer may produce thousands of XRF instruments, each of which requires manual calibration to a number of standards. Each instrument then needs a set investment in time and labor to provide it effective analytical capabilities.
The machine learning module described herein improves on this instrument-by-instrument calibration process. The machine learning module described herein can be used with multiple instruments of a given class to provide a functional model capable of being applied by any instrument of that class. And the machine learning module addresses this inter-instrument reproducibility in a scalable manner. If a small subset of instruments, which represent the range of variation for that instrument class, are calibrated using the machine learning module described herein, it can allow for the transfer of those resulting calibrations to other instruments by differentiating between diagnostic and instrument-dependent signals in the data input from the detector. This ability allows for a given set of analysis to scale without manual calibration on each, while maintaining integrity of the analysis and its ability to be reproduced by other devices or equipment.
An example of why the calibration using the machine learning module described herein is advantageous can be illustrated in the following hypothetical example. The L-alpha and -beta lines for lead (Pb) fluoresce at 10.5 and 12.5 keV, respectively. In a silicate material, these x-ray energies will penetrate through 900 and 1600 μm, respectively through a silicon dioxide (SiO2) matrix. However, a lead-oxide material would have much smaller penetration depths of 44 and 70 μm, respectively. A set of known standards used to quantify Pb ranging from pure SiO2 to PbO would be unable to rely on the fluorescence lines, as with higher concentrations of Pb the Pb lines themselves would be reabsorbed. As such, a naive regression model which assumes increasing Pb fluorescence corresponds to increasing Pb concentration would be unable to function. The machine learning module described herein would allow Pb to be estimated by, in one hypothetical implementation, automatically identifying regions of the spectrum which correspond to density shifts, including but not limited to Compton peaks, Rayleigh peaks, or secondary effects of Bremsstrahlung radiation to infer density and, by proxy, the concentration of Pb. Alternatively, a classification system would match the entire spectrum to its newest match qualitatively. In either implementation, the algorithm would evaluate the entire spectrum and various forms of scattering in addition to x-ray fluorescence.
Described herein:
A1. An analysis method, comprising:
A2. The method of claim A1, wherein the existing measurements comprise compositional data about the analyte though empirical measurements or estimated values, membership of the analyte in a class, or any other quantitative or quantitative attribute.
A3. The method of claim A1, wherein the detector signals are determined by a counting mechanism within the detector that provides data in a continuous interval, which can be used either in a full or summarized form by the module.
A4. The method of claim A1, wherein normalization or another data pre-processing step may take place to adjust detector counts as a normalization process or method to highlight distinctive features for input mapping into the machine learning module.
A5. The method of claim A1, wherein machine-learning algorithms are used to identify relationships in detector output with or without normalization including, but not limited to, random forests, gradient boosting, support vector machines, neural networks, and k-nearest neighbors, among others.
A6. The method of claim A1, wherein the use of fluorescent peaks is not necessarily pre-determined, but can also be guided by relationships within the input from the detector.
A7. The method of claim A1, wherein the geometric relationship between the radiation source, analyte, and detector is not dependent on any one particular configuration, but could be optimized for any variable output.
A8. The method of claim 1A, wherein information from other instruments can be used as a part of supervised training.
A9. The method of claim A1, wherein the photons comprise those with potential ionizing effects on a sample.
A10. The method of claim A9, wherein the detector input into the machine learning module may include photon-fermion interactions that are not the product of ionization.
A11. The method of claim A1, wherein the geometric relationship or component composition could vary within a set of limits, and machine learning acts within the range of these differences.
A12. The method of claim A1, further comprising applying the quantitative and qualitative models during analysis of the radiation.
A13. The method of claim A1, wherein charged particles from a source could include photons, electrons, or protons with the ability to ionize an analyte.
B1. An analysis method, comprising:
B2. An analysis method, comprising:
B3. The method of claim B2, further comprising applying the machine learning module to additional energy interactions.
C1. An analysis system, comprising:
C2. The system of claim C1, wherein the machine leaning module is further configured to apply the quantitative and qualitative models to additional energy interactions.
D1. An analysis method, comprising:
E1. An analysis method, comprising:
F1. An analysis method, comprising:
G1. An analysis method, comprising:
H1. An analysis method, comprising:
J1. An analysis method, comprising:
In addition to any previously indicated modification, numerous other variations and alternative arrangements may be devised by those skilled in the art without departing from the spirit and scope of this description, and appended claims are intended to cover such modifications and arrangements. Thus, while the information has been described above with particularity and detail in connection with what is presently deemed to be the most practical and preferred aspects, it will be apparent to those of ordinary skill in the art that numerous modifications, including, but not limited to, form, function, manner of operation and use may be made without departing from the principles and concepts set forth herein. Also, as used herein, the examples and embodiments, in all respects, are meant to be illustrative only and should not be construed to be limiting in any manner.
This patent application claims priority of U.S. Provisional Application Ser. No. 62/741,231, filed on Oct. 4, 2018, the entire disclosure of which is hereby incorporated by reference.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2019/054741 | 10/4/2019 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62741231 | Oct 2018 | US |
