Method and apparatus for processing real-time flow signals using artificial neural networks

By processing fluid impedance data through artificial neural networks and generating reservoir vectors, the real-time problem of monitoring fluid flow characteristics in chemical plants has been solved, enabling accurate analysis of fluid properties.

CN123125478APending Publication Date: 2026-11-17COATING SOLUTIONS LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202580028414.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-04-15
Filing Date
2025-03-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When chemical plants measure the flow characteristics of fluids, it is difficult to achieve real-time monitoring of meaningful information related to the fluid based solely on overall characteristic measurements. Existing technologies cannot effectively grasp the local characteristics of the fluid.

Method used

A flow characteristic monitoring method based on artificial neural networks is adopted. By collecting fluid impedance data, normalizing and cyclic mapping are performed to generate reservoir vectors, updating model parameters, obtaining fluid flow characteristic labels, and using echo state networks (ESN) for flow characteristic analysis.

Benefits of technology

This enables a more effective understanding of fluid properties based on fluid physical information, real-time monitoring of fluid flow characteristics, and improves the accuracy and efficiency of flow characteristic monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123125478A_ABST
    Figure CN123125478A_ABST
Patent Text Reader

Abstract

An embodiment of the flow characteristic monitoring method based on an artificial neural network disclosed herein may include the following steps: collecting impedance data of a first fluid constituting a learning dataset in a static or flowing state; normalizing the resistive and reactive components of the impedance data to generate resistive and reactive data; performing recurrent mapping on the resistive and reactive data to generate at least one reservoir vector of the artificial neural network; comparing the output data of the at least one reservoir vector with first label data corresponding to the resistive and reactive components, and updating the model parameters of the artificial neural network to generate optimal parameters; and applying the at least one reservoir vector to the impedance data of a second fluid as the object of analysis, based on the optimal parameters, to obtain a second label representing the flow characteristics of the second fluid.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to a method and apparatus for processing mobile signals, and more specifically, to a method and apparatus for processing mobile signals in real time using an artificial neural network. Background Technology

[0002] Typically, chemical plants produce a wide variety of substances in large quantities through chemical reactors that initiate chemical reactions. Specifically, chemical plants move fluids containing various chemical substances through pipelines to chemical reactors, where substances are generated or processed.

[0003] In pipelines carrying moving chemicals, it is possible to measure fluid properties (e.g., bulk properties such as pressure, flow rate, and conductivity), but it is difficult to measure local properties of the fluid (e.g., velocity, stress, and morphology).

[0004] Overall characteristic measurements are merely the measurement of simple physical values ​​(i.e., chaotic data), and cannot yield meaningful information. Therefore, relying solely on simple overall characteristic measurements makes it difficult to monitor meaningful rheological properties related to fluids in real time. Summary of the Invention

[0005] Technical issues This disclosure is made to solve the above-mentioned problems, and the purpose of this disclosure is to provide a method and apparatus for real-time processing of flowing signals using an artificial neural network.

[0006] The problems to be solved by this disclosure are not limited to those mentioned above, and other unmentioned problems can be clearly understood by those skilled in the art from the following description.

[0007] means for solving problems As in one embodiment of this disclosure, a flow characteristic monitoring method based on an artificial neural network, executed by a flow characteristic monitoring device, may include the following steps: collecting impedance data of a first fluid constituting a learning dataset in a static or flowing state; normalizing the resistive and reactive components of the impedance data to generate resistive and reactive data; performing recurrent mapping on the resistive and reactive data to generate at least one reservoir vector of the artificial neural network; comparing the output data of the at least one reservoir vector with first label data corresponding to the resistive and reactive components, and updating the model parameters of the artificial neural network to generate optimal parameters; and applying the at least one reservoir vector to the impedance data of a second fluid as the object of analysis, based on the optimal parameters, to obtain a second label representing the flow characteristics of the second fluid.

[0008] Furthermore, the step of collecting the impedance data may include the following steps: measuring the resistive component and the reactive component of the first fluid by recording multiple data points at a fixed frequency using an interval sweep.

[0009] Furthermore, the aforementioned fixed frequency can be selected from a range of 0.1kHz to 10kHz, and the recording speed of the aforementioned multiple data points can be selected from a range of 10 to 1000 per second.

[0010] Furthermore, the first fluid constituting the aforementioned learning dataset can be prepared in advance with different first labels corresponding to different flow characteristics, and the different first labels can each correspond to the aforementioned first label data.

[0011] Furthermore, the aforementioned different flow characteristics may include velocity, stress, and morphology, which can be determined based on the pretreatment time, main treatment time, and whether post-treatment is performed for the aforementioned first fluid.

[0012] Furthermore, the aforementioned artificial neural network may include an echo state network (ESN), and the aforementioned at least one reservoir vector may include at least one hidden state vector of the aforementioned echo state network (ESN), and the aforementioned at least one hidden state vector may be configured to linearly output the aforementioned output data.

[0013] Furthermore, the first fluid and the second fluid may include an electrode slurry, which includes an electrode active material, a conductive material, and a dispersant.

[0014] Furthermore, the step of generating the above-mentioned optimal parameters may include the following steps: performing supervised learning to minimize the error between the first tag data corresponding to the above-mentioned resistivity component and the above-mentioned reactance component and the output data of the above-mentioned at least one reservoir vector.

[0015] Furthermore, as in one embodiment of this disclosure, the flow characteristic monitoring device based on an artificial neural network may include: an impedance analyzer configured to collect impedance data of a first fluid constituting a learning dataset in a static or flowing state; and a controller configured to normalize the resistive and reactive components of the impedance data to generate resistive and reactive data, perform recurrent mapping on the resistive and reactive data to generate at least one reservoir vector of the artificial neural network, compare the output data of the at least one reservoir vector with first label data corresponding to the resistive and reactive components, update the model parameters of the artificial neural network to generate optimal parameters, and apply the at least one reservoir vector to the impedance data of a second fluid as the object of analysis based on the optimal parameters to obtain a second label representing the flow characteristics of the second fluid.

[0016] Furthermore, as in one embodiment of this disclosure, the computer program can be stored in a computer-readable recording medium to be combined with a flow characteristic monitoring device based on an artificial neural network to execute the above-described flow characteristic monitoring method.

[0017] Invention Effects According to one embodiment of the present disclosure, a method and apparatus for real-time processing of flowing signals using an artificial neural network can be provided.

[0018] According to one embodiment of this disclosure, by utilizing artificial neural networks, fluid-related physical information can be grasped more effectively based on easily measurable fluid-related physical information.

[0019] The effects of this disclosure are not limited to those mentioned above, and those skilled in the art can clearly understand other effects not mentioned from the following description. Attached Figure Description

[0020] Figure 1 This is a diagram illustrating a method for acquiring real-time flow signals according to an embodiment of the present disclosure.

[0021] Figure 2 A block diagram illustrating the structure of an apparatus for processing real-time flowing signals using an artificial neural network according to an embodiment of the present disclosure is provided.

[0022] Figure 3 This is a flowchart illustrating a method for processing real-time flowing signals using an artificial neural network according to an embodiment of the present disclosure.

[0023] Figure 4 This diagram illustrates the structure and operation of an echo state network (ESN) according to an embodiment of the present disclosure.

[0024] Figure 5 and Figure 6 A diagram illustrating data used for ESN learning in an embodiment of this disclosure.

[0025] Figure 7 and Figure 8 This is a diagram illustrating data output via ESN in an embodiment of this disclosure.

[0026] Figure 9 A block diagram illustrating the structure of a flow characteristic monitoring device based on an artificial neural network according to an embodiment of this disclosure.

[0027] Figure 10 and Figure 11 This diagram illustrates the specific structure of a flow characteristic monitoring device based on an artificial neural network according to an embodiment of this disclosure.

[0028] Figures 12 to 15 A graph illustrating resistance and reactance data generated by normalizing the resistance and reactance components of impedance data according to an embodiment of this disclosure.

[0029] Figures 16 to 19 This is a diagram illustrating the result of model learning using impedance data obtained through interval scanning based on a fixed frequency value, according to an embodiment of the present disclosure.

[0030] Figure 20 The flowchart illustrates the steps of a flow characteristic monitoring method based on an artificial neural network according to an embodiment of the present disclosure. Detailed Implementation

[0031] The advantages, features, and methods of achieving these advantages and features of this disclosure will become clear from the accompanying drawings and detailed embodiments described below. However, this disclosure is not limited to the embodiments disclosed below, but can be implemented in many different forms. These embodiments are provided only to make the disclosure more complete and to fully inform those skilled in the art of the scope of this disclosure. This disclosure is defined only by the scope of the claims.

[0032] The terminology used in this specification is for illustrative purposes and does not limit this disclosure. In this specification, the singular includes the plural unless specifically mentioned herein. The terms "comprises" and / or "comprising" as used in this specification do not exclude the presence or addition of more than one component in addition to those mentioned.

[0033] Throughout this specification, the same reference numerals refer to the same components, and "and / or" includes each and all combinations of more than one of the mentioned components. While terms such as "first," "second," etc., are used to describe various components, these components are not limited to these terms. These terms are used only to distinguish one component from another. Therefore, the first component mentioned below can also be a second component within the technical concept of this disclosure.

[0034] Unless otherwise defined, all terms used in this specification (including technical and scientific terms) are intended to be used in the sense that is commonly understood by one of ordinary skill in the art to which this disclosure pertains. Furthermore, unless expressly defined otherwise, terms as defined by commonly used dictionaries should not be idealized or over-interpreted.

[0035] Spatial relative terms such as "below," "below," "lower," "above," and "upper," as shown in the figure, are used to simply describe the relationship between a component and other components. Spatial relative terms should be understood as including, in addition to the directions shown in the figure, the different directions of components during use or operation.

[0036] For example, in the case of components shown in the flip diagram, a component described as "below" or "below" other components may be located "above" other components. Therefore, the illustrative term "below" can include both up and down directions. Components can also be oriented along different directions, thus, spatially relative terms can be interpreted according to orientation.

[0037] When describing this disclosure, "device" may include smartphones, laptops, desktop computers, laptops, tablets, Slate PCs, server devices, wearable devices, etc. As another example, "device" may refer to a group of devices consisting of two or more types of devices connected wirelessly / wired.

[0038] The method and apparatus for processing real-time flowing signals using artificial neural networks will be described in detail below with reference to the accompanying drawings.

[0039] Figure 1 This is a diagram illustrating a method for acquiring real-time flow signals according to an embodiment of the present disclosure.

[0040] The device can acquire flow signals (or physical data) of fluid moving in real time through a pipe using at least one sensor. For example, the at least one sensor may include a pressure gauge, flow meter, conductivity meter, or impedance sensor. The at least one sensor may be installed at a specific location on the pipe through which the fluid moves.

[0041] For example, such as Figure 1 As shown, the device can acquire flow signals of fluid circulating in a pipeline (via a pump) through multiple pressure gauges and flow meters. The device can acquire the flow signals of the fluid measured by multiple pressure gauges and flow meters in real time.

[0042] The device can be electrically connected to at least one sensor. Alternatively, the device can be connected to at least one sensor via a network or wireless communication. The device can acquire fluid flow signals from at least one sensor in real time using wireless communication.

[0043] The network can include wired and wireless networks. For example, the network can include various types of networks such as Local Area Network (LAN), Metropolitan Area Network (MAN), and Wide Area Network (WAN).

[0044] The device normalizes the received fluid flow signal and can generate a hidden state vector (or a vector on the hidden layer) of an artificial neural network model based on the normalized flow signal.

[0045] Artificial neural networks (ANNs) are machine learning models generated based on the structure of human neurons. An ANN can consist of an input layer, intermediate layers (or hidden layers), and an output layer. Various parameters (weights) can be set on the input, intermediate, and output layers. Data is output through the output layer by performing operations on the input data and the weights.

[0046] Furthermore, the device can update / learn the parameters (e.g., weights, etc.) contained in the hidden state vector based on the tag matched with the flow signal and the data output through the hidden state vector.

[0047] Therefore, the device can also input flow signals acquired by at least one sensor into a trained hidden state vector to predict / calculate labels for the flow signals. The labels for the fluid flow signals may include (classification) information representing (or indicating) the physical properties of the fluid.

[0048] Furthermore, when describing this disclosure, training an artificial neural network can refer to updating multiple parameters used in the operation of the artificial neural network to optimal values.

[0049] For the device in Figures 3 to 8 The method of using artificial neural networks to process real-time flowing signals is explained in detail.

[0050] Figure 2 A block diagram illustrating the structure of an apparatus for processing real-time flowing signals using an artificial neural network according to an embodiment of the present disclosure is provided.

[0051] like Figure 2 As shown, device 100 may include memory 110, communication module 120, display 130, input module 140, and processor 150. However, it is not limited to this, and device 100 may modify / add / omit software and hardware structures as needed, within the scope that is obvious to a person skilled in the art.

[0052] The memory 110 can store data supporting various functions of the device 100, programs for the operation of the processor 150, and input / output data (e.g., physical data of the fluid, parameters included in the reservoir vector, etc.). It can also store multiple application programs (or applications) driven by the device, data for the operation of the device 100, and instructions. At least a portion of such applications can be downloaded from an external server via wireless communication.

[0053] Such a memory 110 may include at least one type of storage medium selected from flash memory type, hard disk type, solid state disk type (SSD), silicon disk drive type (SDD), multimedia card micro type, card type memory (e.g., SD or XD memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk.

[0054] Furthermore, the memory 110 may include a database that is separate from this device but connected via a wired or wireless means. That is, Figure 1 The database shown can be implemented as a component of the memory 110.

[0055] The communication module 120 may include one or more components capable of communicating with external devices. For example, the communication module 120 may receive physical data of the fluid from at least one sensor.

[0056] For example, the communication module 120 may include at least one of a broadcast receiving module, a wired communication module, a wireless communication module, a short-range communication module, and a location information module.

[0057] Wired communication modules can include not only various wired communication modules such as Local Area Network (LAN) modules, Wide Area Network (WAN) modules, or Value Added Network (VAN) modules, but also various cable communication modules such as Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), Digital Visual Interface (DVI), recommended standard 232 (RS-232), power line communication, or plain old telephone service (POTS).

[0058] In addition to Wi-Fi modules and wireless broadband modules, wireless communication modules may also include wireless communication modules that support various wireless communication methods such as Global System for Mobile Communication (GSM), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Universal Mobile Telecommunications System (UMTS), Time Division Multiple Access (TDMA), Long Term Evolution (LTE), 4G, 5G, and 6G.

[0059] Display 130 displays (outputs) information processed in this device 100 (e.g., fluid physical data, data output through reservoir vectors, update results of various parameters, etc.). For example, the display may display runtime screen information of an application (e.g., a program) driven in this device 100, or user interface (UI) or graphical user interface (GUI) information based on such runtime screen information.

[0060] The input module 140 is used to receive information input from the user. When information is input through the user input section, the processor 150 can control the operation of the device 100 to correspond to the input information.

[0061] This input module 140 may include hardware physical keys (e.g., buttons, dome switches, microwheels, microswitches, etc. located on at least one of the front, back, and sides of the device) and software touch keys. For example, the touch keys may consist of virtual keys, soft keys, or visual keys that are processed by software and displayed on a touchscreen-type display 130, or they may consist of touch keys located outside the touchscreen. Furthermore, the aforementioned virtual keys or visual keys may be displayed on the touchscreen simultaneously in various forms, such as graphics, text, icons, videos, or combinations thereof.

[0062] The processor 150 can control the overall operation and function of the device 100. Specifically, the processor 150 can be implemented as: a memory storing data of an algorithm for controlling the operation of components within the device 100 or a program for reproducing the algorithm, and at least one processor (not shown) that uses the data stored in the memory to perform the aforementioned operations. In this case, the memory and the processor can be implemented by separate chips. Alternatively, the memory and the processor can also be implemented by a single chip.

[0063] Furthermore, in order to achieve the following on this device 100 Figures 3 to 8 In the various embodiments of this disclosure described herein, processor 150 can perform combined control of any one or more of the components examined above.

[0064] Figure 3 This is a flowchart illustrating a method for processing real-time flowing signals using an artificial neural network according to an embodiment of the present disclosure.

[0065] The device can acquire first physical data (or flow signal) of the first fluid moving in the pipe through at least one sensor (S310).

[0066] At this point, the device can acquire initial physical data of the first fluid moving within the pipe through at least one sensor over multiple time units. The time unit can be set to the time it takes for the fluid to circulate once within the pipe, but it is not limited to this and can also be set by the device.

[0067] Furthermore, at least one sensor may include a pressure gauge, flow meter, conductivity meter, or impedance sensor. And the first physical data may include the fluid's pressure, flow rate, conductivity, or impedance.

[0068] The device can normalize the first physical data of the first fluid (S320).

[0069] Specifically, the device can perform principal component analysis on the first physical data to reduce the dimensionality of the first physical data. Principal component analysis refers to the method of transforming high-dimensional data into low-dimensional data while preserving the distribution of existing data to the greatest extent possible.

[0070] The device can perform linear regression analysis on the reduced-dimensional first physical data to obtain a trend line. Linear regression analysis is an analytical method that models the relationship between variables as a linear function, assuming that the independent and dependent variables are linearly correlated. The trend line is a line representing the distribution or tendency of the first physical data.

[0071] The device can remove the trend (e.g., data at a predefined distance from the trend line) from the first physical data with reduced dimensionality based on a trend line, and obtain the standard deviation from the trend-removed first physical data. The device can normalize the first physical data by dividing the first physical data by the standard deviation.

[0072] The device can perform recurrent mapping on the normalized first physical data to generate at least one reservoir vector of the artificial neural network (S330).

[0073] When the artificial neural network is an echo state network (ESN), the reservoir vector can refer to the hidden state vector. ESN refers to an artificial neural network generated based on a recurrent neural network (RNN).

[0074] RNN stands for Sequence Neural Network, which processes input and output data in sequence units. That is, while passing the results calculated by activation functions at hidden layer nodes to the output layer, an RNN can also pass these results back as input for the next calculation at a hidden layer node. A cyclic mapping can refer to operations performed between the first physical data and the hidden layers of the RNN.

[0075] ESN refers to a network that, after generating any RNN, only learns the parameters (or weights) between the nodes in the output layer and the nodes in the hidden layer. That is, the weights from the input layer nodes to the hidden layer nodes and the weights between the hidden layer nodes are not learned.

[0076] Reference Figure 4Nonlinear data in the input layer can be processed with reservoir vectors. Through a linear mapping of the reservoir vectors, the output data can be linear.

[0077] The device can perform cyclic mapping on the first physical data to calculate the reservoir vector (x(n=1,2,…T)).

[0078] Input type and size (N) u The size of the reservoir vector (N) can be determined based on the type and magnitude of the first physical data output by at least one sensor. x When the entire time step (or unit time) is defined as T, it should satisfy (T>1+N). u +N x ).

[0079] The larger the size of the reservoir vector, the better the computational performance of the ESN. Therefore, the size of the reservoir vector can be increased to near the upper limit of the device's memory or computing power.

[0080] The operations performed in the reservoir vector can be implemented as shown in mathematical expressions 1 and 2. Here, α can be 0 or 1.

[0081] [Mathematical Expression 1]

[0082] [Mathematical Expression 2]

[0083] Furthermore, the value output from the reservoir vector (W) out It can be implemented as mathematical expression 3.

[0084] [Mathematical Expression 3]

[0085]

[0086] Furthermore, the loss function (or objective function) in the reservoir vector can be implemented as shown in Equation 4.

[0087] [Mathematical Expression 4]

[0088] The device can update the parameters between at least one reservoir vector and the output data (S340).

[0089] The device can update the parameters between at least one reservoir vector and the output data based on the data output through at least one reservoir vector and the first label data corresponding to the first physical data of the first fluid.

[0090] In this case, learning can be omitted from the reservoir vector calculation steps and only performed in the step of transforming the reservoir vector into output data. That is, the parameters between the nodes corresponding to the reservoir vector and the nodes corresponding to the output nodes can be learned / updated only.

[0091] Specifically, the device can preset first tag data for the first physical data. The first tag may include information related to the physical properties of the first fluid (i.e., information that represents the physical properties of the first fluid). The device can perform one-hot encoding on the first tag to obtain the first tag data.

[0092] One-hot encoding refers to setting the size of the classification set to the dimension of a vector, assigning 1 to the index of the word to be represented, and assigning 0 to other indices, thereby representing the word in vector form.

[0093] The apparatus can calculate the error between data output through at least one reservoir vector and first tag data corresponding to first physical data of a first fluid. The data output through the reservoir vector is also in vector form with one-thermal encoding, and the apparatus can calculate the error by calculating the distance / difference between the two vectors. The apparatus can update the parameters between at least one reservoir vector and the output data to reduce (i.e., minimize) the calculated error.

[0094] That is, the device updates the parameters between at least one reservoir vector and the output data through supervised learning, thereby enabling it to learn by using an artificial neural network including the reservoir vector to output label data corresponding to the physical data through the physical data output.

[0095] Furthermore, as an example, based on nonlinear operations defined in at least one reservoir vector, the device can update parameters using either a stochastic gradient descent (SGD) algorithm or an adaptive moment estimation (ADAM) algorithm. However, this is merely one embodiment; the device can also update parameters by performing a direct inverse matrix operation on at least one reservoir vector and / or the output data.

[0096] As another example, based on the definition of performing linear operations in at least one reservoir vector, the matrix in the reservoir vector used for linear mapping ( It can be implemented together with the hyperparameter β as mathematical expression 5.

[0097] [Mathematical Expression 5]

[0098] Based on β, reservoir vectors can be classified into learning data and test data. The device can learn parameters using the learning data and then perform inference actions using the test data.

[0099] Types and quantity (N) of input data (i.e., first physical data) u The optimal value can be calculated through various combinations of tests. Furthermore, in order to calculate all the hyperparameters (c, a, ρ, α, N) required to generate the reservoir vector... x The optimal value of can be found using random search, grid search, or Bayesian optimization techniques.

[0100] For example, Figure 5 The example illustrates the learning data. For instance, the device can acquire a first physical dataset for a first fluid over 7 days, where each category can refer to the first physical dataset measured on a daily basis. The first physical dataset can be easily distinguished by its mean and variance. That is, the first physical dataset can have a time trend. Figure 6 Examples Figure 5 The normalization of learning data in the process.

[0101] Figure 7 and Figure 8 This shows the result of applying reservoir vectors to the learning data and the output.

[0102] For example, the input data can be any combination of physical data (e.g., the pressure and velocity of the first fluid, etc.). Furthermore, the size of the reservoir vector (N) x ) is 10 4 w in The input scaling factor a can be 1, and the spectral radius (ρ(W)) of w can be... The leakage ratio (α) can be 1. The output data can be implemented as binary values ​​of the first or last day or in the form of multi-class classification information.

[0103] The device can apply at least one reservoir vector to the second physical data of the second fluid to obtain a second label for the second physical data (S350). The second fluid may refer to a fluid different from the first fluid, i.e., a fluid not used for learning.

[0104] Based on second physical data of the fluid acquired through at least one sensor, the device can acquire a second tag of the second physical data by inputting the second physical data into at least one reservoir vector.

[0105] Specifically, the parameters associated with at least one reservoir vector are optimized through learning. The device can obtain second label data about the second physical data by inputting the second physical data of the second fluid into an artificial neural network including at least one reservoir vector. The device can obtain information related to the physical properties of the second physical data (i.e., information representing the physical properties of the second fluid) by decoding the second label data.

[0106] As another example, the device can acquire second physical data of a second fluid over multiple time units using at least one sensor. The device can apply a reservoir vector to the second physical data acquired over the multiple time units to obtain multiple output data. The device can acquire second tag data as the average of the multiple output data, and decode at least one of the second tag data or the reservoir vector to obtain information related to the physical properties of the second physical data.

[0107] Furthermore, the disclosed embodiments can be implemented in the form of a recording medium storing computer-executable instructions. The instructions can be stored as program code, which, when executed by a processor, can generate program modules to perform the actions of the disclosed embodiments. The recording medium can be implemented as a computer-readable recording medium.

[0108] Computer-readable recording media include all types of recording media that store computer-readable instructions. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disk, flash memory, optical data storage devices, etc.

[0109] Figure 9 A block diagram illustrating the structure of a flow characteristic monitoring device based on an artificial neural network according to an embodiment of this disclosure.

[0110] Reference Figure 9 The flow characteristic monitoring device 900 based on artificial neural networks may include an impedance analyzer 910 and a controller 920. However, it is not limited to this; some structures may be omitted from the flow characteristic monitoring device 900, or the flow characteristic monitoring device 900 may also include other general structures.

[0111] Impedance analyzer 910 can be in Figures 1 to 9 This describes at least one type of sensor. The at least one sensor may include a pressure sensor, a flow sensor, a conductivity sensor, an impedance sensor, etc., wherein the impedance sensor may be an impedance analyzer 910. For the specific structure of the impedance analyzer 910, please refer to the description below. Figure 10 and Figure 11 .

[0112] The controller 920 can perform processing procedures for monitoring flow characteristics based on artificial neural networks. For example, the controller 920 can be used with... Figures 1 to 9 The device 100 described herein corresponds to this.

[0113] Impedance analyzer 910 can be configured to collect impedance data of a first fluid constituting a learning dataset in either a static or flowing state. For example, the first fluid can be an electrode slurry in various states. The first fluid can be in a static or flowing state through both the static and flowing containers of impedance analyzer 910. In this regard, the transducer body of impedance analyzer 910 can apply alternating current to the first fluid to measure the impedance data.

[0114] The controller 920 can be configured to normalize the resistive and reactive components of the impedance data to generate resistance and reactance data. The impedance data may include a real part and an imaginary part; the real part may correspond to the resistive component of the first fluid, and the imaginary part may correspond to the reactive component of the first fluid. For example... Figures 1 to 9 As explained, normalization can be performed using the average and standard deviation of the resistive / reactive components. Additionally, as... Figures 1 to 9 As explained in the paper, the normalization process can be accompanied by principal component analysis and linear regression analysis.

[0115] The controller 920 can be configured to perform recurrent mapping on resistance and reactance data to generate at least one reservoir vector of an artificial neural network. Recurrent mapping can refer to performing operations related to the layers of an artificial neural network such as an RNN. An artificial neural network (ANN) can be an artificial intelligence model that learns in a manner that performs given functions, and the at least one reservoir vector may include the state vectors of the hidden layers. The at least one reservoir vector can receive resistance and reactance data as input and output labeled data, which may represent the flow characteristics or physical properties of the target fluid.

[0116] The controller 920 is configurable to generate optimal parameters by updating the model parameters of an artificial neural network by comparing the output data of at least one reservoir vector with first label data corresponding to the resistivity and reactance components. The first label data can be pre-prepared in a manner corresponding to the flow characteristics or physical properties of a first fluid. The model parameters of the artificial neural network can be updated in the direction of the output first label data when the input resistance / reactance data is received. After training the artificial neural network through parameter updates, at least one reservoir vector can classify / infer the flow characteristics or physical properties of the fluid based on the input data.

[0117] The controller 920 can be configured to apply at least one reservoir vector based on optimal parameters to obtain a second label representing the flow characteristics of the second fluid, given the impedance data of the second fluid being analyzed. When the model parameters are updated and the artificial neural network completes its learning, the at least one reservoir vector can classify the flow characteristics or physical properties of the fluid based on its impedance information. The second fluid can be a fluid not used for updating / learning, or it can be the target fluid that the user wants to understand. When the second label is obtained, the flow characteristics or physical properties of the target fluid can be confirmed.

[0118] According to an embodiment, the controller 920 can be configured to determine the resistive and reactive components of a first fluid by recording multiple data points at a fixed frequency using an interval sweep. The interval sweep can include techniques for recording a large number of data points by impedance scanning at fixed frequencies such as 1 kHz, 50 kHz, and 500 kHz. When measuring the impedance of the target fluid by interval sweep, it can be separated into real and imaginary parts to determine the resistive and reactive components. Alternatively, a frequency sweep using a varying frequency can be used instead of a fixed-frequency interval sweep.

[0119] According to an embodiment, the fixed frequency can be selected from a range of 0.1 kHz to 10 kHz, and the recording speed for multiple data points can be selected from a range of 10 to 1000 data points per second. The numerical ranges of the fixed frequency and recording speed described above can be selected based on the performance of the second label classifying the second fluid as the object of analysis. Preferably, the fixed frequency can be 1 kHz, and the recording speed can be 100 data points per second. For this, please refer to the following description. Figures 12 to 19 .

[0120] According to an embodiment, the first fluid constituting the learning dataset can be prepared in advance with distinct first labels corresponding to distinct flow characteristics, and each distinct first label can correspond to a first label data. For example, when the first fluid is an electrode slurry, it can have different flow characteristics or physical properties depending on how and for how long the electrode slurry is mixed. Flow characteristics or physical properties may include local characteristics such as fluid velocity, fluid stress, and fluid morphology. To generate the learning dataset, after preparing the first fluids with distinct flow characteristics, assigning them first labels, and analyzing their impedance characteristics, the dataset can be analyzed.

[0121] According to embodiments, the different flow characteristics include velocity, stress, and morphology, which can be determined based on the pretreatment time, main treatment time, and whether post-treatment is performed on the first fluid. For example, the pretreatment, main treatment, and post-treatment of the first fluid can be the pretreatment, main treatment, and post-treatment of a mixing process for the electrode slurry. The pretreatment, main treatment, and post-treatment can be performed in different mixing operations. Furthermore, the physical / flow characteristics of the electrode slurry will vary depending on the duration of the mixing process. When subjected to different forms of mixing processes, the first fluid can have different combinations of fluid velocity, fluid stress, and fluid morphology.

[0122] According to an embodiment, the artificial neural network may include the aforementioned echo state network (ESN), and at least one reservoir vector may include at least one hidden state vector of the echo state network (ESN), the at least one hidden state vector being configured to linearly output data. As previously stated, even with nonlinear input data, the artificial neural network can generate linear output data through a linear mapping of at least one reservoir vector.

[0123] According to an embodiment, the first fluid and the second fluid may include an electrode slurry, which includes an electrode active material, a conductive material, and a dispersant. For example, in the first fluid and the second fluid, the composition ratio of the electrode active material may be 35.84 wt% (20.14 vol.%), the composition ratio of the conductive material may be 0.80 wt% (0.44 vol.%), and the composition ratio of the dispersant may be 0.80 wt%. However, such composition ratios are merely illustrative, and specific values ​​may vary depending on changes in electrode design or other specifications.

[0124] According to an embodiment, the controller 920 can be configured to perform supervised learning to minimize the error between the first label data corresponding to the resistivity and reactance components and the output data of at least one reservoir vector. During the updating of the model parameters of the artificial neural network, supervised learning can be performed using a pre-prepared relationship between the physical properties of the first fluid and the first label data as training data. For this purpose, the difference / error between the output data of at least one reservoir vector and the first label data can be set as a loss function, enabling machine learning to be performed in the direction that minimizes the loss function.

[0125] Figure 10 and Figure 11 This diagram illustrates the specific structure of a flow characteristic monitoring device based on an artificial neural network according to an embodiment of this disclosure.

[0126] Reference Figure 10The transducer body 1000 of the impedance analyzer 910 can be shown. For example, the transducer body 1000 can be a product of HIOKI Corporation, but is not limited to this.

[0127] The transducer body 1000 can apply an impedance measurement input signal to a fluid object that is in a static or flowing state, and can obtain impedance information by sensing changes in the electrical characteristics of the fluid object.

[0128] Reference Figure 11 The diagram shows a stationary container 1110 configured to keep the target fluid stationary and a flowing container 1120 configured to keep the target fluid flowing.

[0129] The stationary container 1110 can hold 2 ml to 3 ml of the target fluid, such as a first fluid or a second fluid. The signal input / output port of the transducer body 1000 can be connected to the stationary container 1110, thereby collecting impedance information of the stationary fluid. The flowing container 1120 can be disposed in, for example... Figure 1 The object is on the pipe through which the fluid flows.

[0130] Figures 12 to 15 A graph illustrating resistance and reactance data generated by normalizing the resistance and reactance components of impedance data according to an embodiment of this disclosure.

[0131] Reference Figure 12 The resistance data 1210 and reactance data 1220 can be displayed when scanning at fixed intervals of 1 kHz. The resistance data 1210 and reactance data 1220 can be represented in a graphical form with the resistance / reactance value on the vertical axis representing the number of scans on the horizontal axis.

[0132] Resistance data 1210 and reactance data 1220 can be generated for five different first fluids. The five first fluids can be electrode slurries including active substances, conductive materials, dispersants, etc. The five first fluids can be distinguished according to the specific method of performing the slurry mixing process on the electrode slurry, and the flow characteristics or physical properties of the five first fluids will become different.

[0133] The five primary fluids possess distinct flow characteristics or physical properties; therefore, as shown in the figure, resistance data 1210 and reactance data 1220 can exhibit different patterns. Artificial neural networks, such as ESN, can learn these data patterns and incorporate them into the model parameters. When fluid flow begins, both resistance and reactance decrease rapidly; during flow, the amplitude of their fluctuations increases. As flow continues for a longer period, the amplitude of these fluctuations decreases; and when flow terminates, the resistance and reactance values ​​recover.

[0134] Typically, data patterns exhibit irregular (chaotic) vibrational modes, and their unique characteristics can be learned through machine learning. After parameter updates, the artificial neural network can receive input from resistance or reactance patterns to classify / infer the flow characteristics or physical properties of the fluid. Experimental results confirm that, compared to resistance data 1210, patterns can be distinguished more clearly in reactance data 1220, and the learning / inference performance is also higher.

[0135] Reference Figure 13 The resistance data 1310 and reactance data 1320 can be displayed when the data is scanned at a fixed frequency of 50 kHz. (Refer to...) Figure 14 The resistance data 1410 and reactance data 1420 can be shown when the data is scanned at a fixed frequency of 500 kHz. (Refer to...) Figure 15 The resistance data 1510 and reactance data 1520 can be shown when the interval is scanned at a fixed frequency of 8 MHz.

[0136] observe Figures 13 to 15 Based on the resistance and reactance data, it can be confirmed that compared to Figure 12 The resistance data 1210 and reactance data 1220 are more irregular and difficult to distinguish. Based on this, it can be confirmed that the highest learning / inference performance is exhibited in a fixed frequency range of 0.1kHz to 10kHz, and preferably, a fixed frequency of 1kHz can be used. Figure 15 At 8MHz, a recording speed of 70 times per second was utilized; in addition, Figures 12 to 14 The data was measured at a recording rate of 100 times per second.

[0137] Figures 16 to 19 This is a diagram illustrating the result of model learning using impedance data obtained through interval scanning based on a fixed frequency value, according to an embodiment of the present disclosure.

[0138] Reference Figure 16 A graph 1600 shows the classification accuracy of an artificial neural network when scanning at fixed intervals of 1 kHz. Similarly, Figures 17 to 19 Charts 1700, 1800, and 1900 can be displayed for 50kHz, 500kHz, and 8MHz respectively.

[0139] Referring to Table 1600, model performance can be adjusted based on the amount of training data (n). train The numbers 250, 500, 1000, 2000, and 4000 become different. For example, the amount of training data (n) changes. train This corresponds to the number of the first fluid. Additionally, it can be confirmed that for the same number of training data (n)... train ), average number (n)avg The higher the value, the better the model performance.

[0140] Unlike Figure 1600, it can be confirmed that stable performance is difficult to achieve in the cases of Figures 1700, 1800, and 1900. That is, the highest performance is observed in the range of 1 kHz within the fixed frequency of the interval scan. Based on this, the fixed frequency can be selected from the range of 0.1 kHz to 10 kHz.

[0141] Figure 20 The flowchart illustrates the steps of a flow characteristic monitoring method based on an artificial neural network according to an embodiment of the present disclosure.

[0142] Reference Figure 20 The flow characteristic monitoring method 2000 based on artificial neural networks may include steps 2010 to 2050. However, it is not limited to this. Some steps may be omitted in the flow characteristic monitoring method 2000, or the flow characteristic monitoring method 2000 may also include other general steps. Steps 2010 to 2050 may be performed in a different order than that shown.

[0143] The flow characteristic monitoring method 2000 based on artificial neural networks can be composed of steps processed sequentially by the flow characteristic monitoring device 900 based on artificial neural networks. Therefore, the content described above for the flow characteristic monitoring device 900 is equally applicable to the flow characteristic monitoring method 2000.

[0144] The flow characteristic monitoring method 2000 based on artificial neural networks can be executed by the impedance analyzer 910 and controller 920 of the flow characteristic monitoring device 900 based on artificial neural networks.

[0145] In step 2010, the flow characteristic monitoring device 900 may perform the following steps: collect impedance data of a first fluid constituting a learning dataset in a static or flowing state.

[0146] In step 2020, the flow characteristic monitoring device 900 may perform the following steps: normalize the resistive and reactive components of the impedance data to generate resistance and reactance data.

[0147] In step 2030, the flow characteristic monitoring device 900 may perform the following steps: perform recurrent mapping on the resistance data and reactance data to generate at least one reservoir vector of the artificial neural network.

[0148] In step 2040, the flow characteristic monitoring device 900 may perform the following steps: compare the output data of at least one reservoir vector with the first tag data corresponding to the resistivity component and the reactance component, and update the model parameters of the artificial neural network to generate optimal parameters.

[0149] In step 2050, the flow characteristic monitoring device 900 may perform the following steps: applying at least one reservoir vector based on the optimal parameters to the impedance data of the second fluid as the object of analysis, in order to obtain a second label representing the flow characteristics of the second fluid.

[0150] The flow characteristic monitoring method 2000 based on artificial neural networks can be implemented as a recording medium storing computer-executable instructions. The instructions can be stored as program code, which, when executed by a processor, can generate program modules to perform the actions of the disclosed embodiments. The recording medium can be implemented as a computer-readable recording medium.

[0151] Computer-readable recording media include all types of recording media that store computer-readable instructions. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disk, flash memory, optical data storage devices, etc.

[0152] The flow characteristic monitoring method 2000 based on artificial neural networks can be implemented as a computer program. The computer program can be combined with the flow characteristic monitoring device 900 based on artificial neural networks and stored in a computer-readable recording medium to execute the flow characteristic monitoring method 2000 based on artificial neural networks.

[0153] As described above, the disclosed embodiments have been illustrated with reference to the accompanying drawings. It is understood that those skilled in the art to which this disclosure pertains can implement this disclosure in ways different from the disclosed embodiments without altering the technical concept or essential features of this disclosure. The disclosed embodiments are illustrative and should not be construed as restrictive.

[0154] Explanation of reference numerals in the attached figures 100: Device 110: Memory 120: Communication Module 130: Monitor 140: Processor 900: Flow characteristic monitoring device based on artificial neural network

Claims

1. A flow characteristic monitoring method based on an artificial neural network, executed by a flow characteristic monitoring device, comprising the following steps: Collect impedance data of the first fluid that constitutes the learning dataset in either a static or flowing state; The resistance and reactance components of the above impedance data are normalized to generate resistance and reactance data. The resistance data and reactance data are cyclically mapped to generate at least one reservoir vector of the artificial neural network. The output data of at least one reservoir vector is compared with the first label data corresponding to the resistivity and reactance components, and the model parameters of the artificial neural network are updated to generate optimal parameters; and For the impedance data of the second fluid being analyzed, at least one reservoir vector is applied based on the above-mentioned optimal parameters to obtain a second label representing the flow characteristics of the second fluid.

2. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, The steps for collecting the above impedance data include the following: The resistivity and reactance components of the first fluid are measured by recording multiple data points at intervals at a fixed frequency.

3. The flow characteristic monitoring method based on artificial neural networks according to claim 2, wherein, The aforementioned fixed frequency is selected from the range of 0.1 kHz to 10 kHz, and the recording speed of the aforementioned multiple data points is selected from the range of 10 to 1000 per second.

4. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, The first fluid constituting the above learning dataset is prepared in advance with different first labels corresponding to different flow characteristics, and the different first labels correspond to the first label data respectively.

5. The flow characteristic monitoring method based on artificial neural networks according to claim 4, wherein, The aforementioned different flow characteristics include velocity, stress, and morphology. The aforementioned different flow characteristics are determined based on the pretreatment time, main treatment time, and whether post-treatment is performed for the first fluid.

6. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, The aforementioned artificial neural networks include echo-state networks. The aforementioned at least one reservoir vector includes at least one hidden state vector of the aforementioned echo state network. At least one of the aforementioned hidden state vectors is configured to linearly output the aforementioned output data.

7. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, The first fluid and the second fluid mentioned above include electrode slurry, and the electrode slurry includes electrode active material, conductive material and dispersant.

8. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, The steps to generate the above optimal parameters include the following: Supervised learning is performed to minimize the error between the first tag data corresponding to the aforementioned resistivity and reactance components and the output data of the aforementioned at least one reservoir vector.

9. The flow characteristic monitoring method based on artificial neural networks according to claim 1, wherein, This includes the step of collecting pressure data for the first fluid.

10. The flow characteristic monitoring method based on artificial neural networks according to claim 9, wherein, The steps include the following: The pressure data is normalized, and the normalized pressure data is cyclically mapped to generate at least one reservoir vector of the artificial neural network. The data linearly output through at least one reservoir vector is compared with the first label data corresponding to the pressure data of the first fluid, and the model parameters of the artificial neural network are updated to generate optimal parameters; and For the pressure data of the second fluid being analyzed, at least one reservoir vector is applied based on the above-mentioned optimal parameters to obtain a second label representing the flow characteristics of the second fluid.

11. A flow characteristic monitoring device based on an artificial neural network, comprising: Impedance analyzer, configured to collect impedance data of a first fluid constituting a learning dataset in either a static or flowing state; as well as The controller is configured to normalize the resistive and reactance components of the aforementioned impedance data to generate resistance and reactance data. The aforementioned resistance and reactance data are cyclically mapped to generate at least one reservoir vector for an artificial neural network. The output data of at least one reservoir vector is compared with the first label data corresponding to the resistivity component and the reactance component, and the model parameters of the artificial neural network are updated to generate optimal parameters. The impedance data of the second fluid, which is the object of analysis, is then applied to the at least one reservoir vector based on the optimal parameters to obtain a second label representing the flow characteristics of the second fluid.

12. A computer program stored in a computer-readable recording medium for use in conjunction with a flow characteristic monitoring device based on an artificial neural network to perform the flow characteristic monitoring method of any one of claims 1 to 10.