Method and device for processing real-time flow signal by using artificial neural network

Abstract

An artificial neural network-based flow characteristic monitoring method according to one embodiment of the present disclosure may comprises the steps of: collecting impedance data of first fluids constituting a learning dataset in a stationary state or a flow state; generating resistance data and reactance data by normalizing a resistance component and a reactance component of the impedance data; generating at least one reservoir vector of an artificial neural network by performing recurrent mapping on the resistance data and the reactance data; comparing output data of the at least one reservoir vector with first label data corresponding to the resistance component and the reactance component, and updating a model parameter of the artificial neural network to thus generate an optimal parameter; and applying the at least one reservoir vector to impedance data of a second fluid to be analyzed on the basis of the optimal parameter to obtain a second label indicating flow characteristics of the second fluid.

Description

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

[0001] The present disclosure relates to a method and device for processing a flow signal, and more particularly, to a method and device for processing a flow signal in real time using an artificial neural network.

[0002] Typically, chemical plants mass-produce various substances through chemical reactors, which trigger chemical reactions. Specifically, chemical plants transport fluids containing various chemicals through pipes to chemical reactors, where the resulting substances can be produced or processed.

[0003] Bulk properties such as pressure, flow rate, and conductivity can be measured in a pipe that transports chemicals, but local properties of the fluid (e.g., velocity, stress, morphology, etc.) are difficult to measure.

[0004] Bulk property measurements themselves merely measure physical quantities (i.e., chaotic data) and do not provide meaningful information. Therefore, simple bulk property measurements have presented a challenge in monitoring meaningful rheological properties of fluids in real time.

[0005] The present disclosure has been made to solve the above-described problems, and the purpose of the present disclosure is to provide a method and device for processing a flow signal in real time using an artificial neural network.

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

[0007] In one embodiment of the present disclosure, a method for monitoring flow characteristics based on an artificial neural network, performed by a flow characteristic monitoring device, may include the steps of: collecting impedance data of first fluids constituting a learning dataset in a stationary state or a flowing state; performing normalization on resistance components and reactance components of the impedance data to generate resistance data and reactance data; performing recurrent mapping on the resistance data and the reactance data to generate at least one reservoir vector of an artificial neural network; comparing output data of the at least one reservoir vector with first label data corresponding to the resistance components and the reactance components to update model parameters of the artificial neural network to generate optimal parameters; and applying the at least one reservoir vector based on the optimal parameters to impedance data of a second fluid to be analyzed to obtain a second label representing a flow characteristic of the second fluid.

[0008] And, the step of collecting the impedance data may include a step of measuring the resistance component and the reactance component of the first fluid through an interval sweep that records a plurality of data points at a fixed frequency.

[0009] In addition, the fixed frequency can be selected within a range of 0.1 kHz or more and 10 kHz or less, and the recording speed of the plurality of data points can be selected within a range of 10 or more per second and 1000 or less per second.

[0010] In addition, the first fluids constituting the learning dataset may be prepared in advance to have different first labels corresponding to different flow characteristics, and each of the different first labels may correspond to the first label data.

[0011] And, the different flow characteristics may include velocity, stress and morphology, and the different flow characteristics may be determined based on the pre-treatment time, main treatment time and whether or not post-treatment is performed for the first fluids.

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

[0013] And, the first fluids and the second fluid may include an electrode slurry including an electrode active material, a conductive material, and a dispersant.

[0014] And, the step of generating the optimal parameters may include a step of performing supervised learning to minimize an error between the first label data corresponding to the resistance component and the reactance component and the output data of the at least one reservoir vector.

[0015] And, in one embodiment of the present disclosure, an artificial neural network-based flow characteristic monitoring device may include: an impedance analyzer configured to collect impedance data of a first fluid constituting a learning data set in a stationary state or a flowing state; and a controller configured to perform normalization on resistance components and reactance components of the impedance data to generate resistance data and reactance data, perform recurrent mapping on the resistance data and the reactance data to generate at least one reservoir vector of the artificial neural network, compare output data of the at least one reservoir vector with first label data corresponding to the resistance component and the reactance component to update model parameters of the artificial neural network to generate optimal parameters, and apply the at least one reservoir vector to impedance data of a second fluid to be analyzed based on the optimal parameters to obtain a second label representing flow characteristics of the second fluid.

[0016] And, as one embodiment of the present disclosure, a computer program may be stored in a computer-readable recording medium to execute the flow characteristic monitoring method in combination with an artificial neural network-based flow characteristic monitoring device.

[0017] According to one embodiment of the present disclosure, a method and device for processing a flow signal in real time using an artificial neural network can be provided.

[0018] According to one embodiment of the present disclosure, the physical properties of a fluid can be more efficiently obtained based on physical information about the fluid that can be easily measured using an artificial neural network.

[0019] The effects of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0020] FIG. 1 is a diagram illustrating a method for obtaining a real-time flow signal according to one embodiment of the present disclosure.

[0021] FIG. 2 is a block diagram briefly illustrating the configuration of a device for processing real-time flow signals using an artificial neural network according to one embodiment of the present disclosure.

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

[0023] FIG. 4 is a diagram for explaining the structure and operation of an echo state network (ESN) according to one embodiment of the present disclosure.

[0024] FIG. 5 and FIG. 6 are diagrams for explaining data for learning of ESN according to one embodiment of the present disclosure.

[0025] FIG. 7 and FIG. 8 are diagrams for explaining data output through ESN according to one embodiment of the present disclosure.

[0026] FIG. 9 is a block diagram illustrating a configuration of a flow characteristic monitoring device based on an artificial neural network according to one embodiment of the present disclosure.

[0027] FIG. 10 and FIG. 11 are drawings for explaining a specific structure of a flow characteristic monitoring device based on an artificial neural network according to one embodiment of the present disclosure.

[0028] FIGS. 12 to 15 are diagrams for explaining resistance data and reactance data generated by performing normalization on the resistance component and reactance component of impedance data according to one embodiment of the present disclosure.

[0029] FIGS. 16 to 19 are diagrams for explaining the results of model learning performed with impedance data by interval sweep according to a numerical value of a fixed frequency according to one embodiment of the present disclosure.

[0030] FIG. 20 is a flowchart illustrating steps of a method for monitoring flow characteristics based on an artificial neural network according to one embodiment of the present disclosure.

[0031] The advantages and features of the present disclosure, and methods for achieving them, will become clearer with reference to the embodiments described below in detail with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided solely to ensure that the disclosure is complete and to fully inform those skilled in the art of the scope of the present disclosure, and the present disclosure is defined solely by the scope of the claims.

[0032] The terminology used herein is for the purpose of describing embodiments and is not intended to limit the present disclosure. In this specification, singular forms also include plural forms, unless specifically stated otherwise. As used herein, the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components in addition to the components mentioned.

[0033] Throughout the specification, the same reference numerals refer to the same elements, and the term "and / or" includes each and every combination of the elements mentioned. Although terms such as "first," "second," etc. are used to describe various elements, these elements are not limited by these terms. These terms are merely used to distinguish one element from another. Accordingly, it should be understood that a first element mentioned below may also be a second element within the technical scope of the present disclosure.

[0034] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which this disclosure pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0035] Spatially relative terms such as "below," "beneath," "lower," "above," and "upper" may be used to easily describe the relationship of one component to another, as illustrated in the drawings. Spatially relative terms should be understood to include different orientations of components during use or operation in addition to the orientations illustrated in the drawings.

[0036] For example, if a component depicted in a drawing is flipped, a component described as "below" or "beneath" another component may be positioned "above" the other component. Thus, the exemplary term "below" may encompass both the above and below orientations. Components may also be oriented in other directions, and thus spatially relative terms may be interpreted based on their orientation.

[0037] In describing the present disclosure, the term "device" may include a smartphone, a laptop, a desktop, a laptop, a tablet PC, a slate PC, a server device, a wearable device, and the like. As another example, the term "device" may refer to a group of devices in which two or more types of devices are connected wirelessly / wirelessly.

[0038] Below, a method and device for processing real-time flow signals using an artificial neural network will be specifically described with reference to drawings.

[0039] FIG. 1 is a diagram illustrating a method for obtaining a real-time flow signal according to one embodiment of the present disclosure.

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

[0041] For example, as illustrated in FIG. 1, the device can acquire a flow signal of a fluid circulating in a pipe (by a pump) through a plurality of pressure gauges and flow meters. The device can acquire a flow signal of the fluid measured from the plurality of pressure gauges and flow meters in real time.

[0042] Here, the device may be electrically connected to at least one sensor. As another example, the device may be connected to at least one sensor via wireless communication over a network. The device may acquire a fluid flow signal from at least one sensor in real time using wireless communication.

[0043] Here, the network may include wired and wireless networks. For example, the network may include various networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN).

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

[0045] Here, an artificial neural network refers to a machine learning model created based on the structure of human neurons. An artificial neural network can consist of an input layer, an intermediate layer (or hidden layer), and an output layer. Various parameters (weights) can be set on the input layer, intermediate layer, and output layer, and data can be output through the output layer through calculations of input data and each weight.

[0046] And, the device can update / learn parameters (e.g., weights, etc.) included in the hidden state vector based on data output through the label and hidden state vector matched to the flow signal.

[0047] Accordingly, the device can additionally predict / calculate the label of the flow signal by inputting the flow signal acquired through at least one sensor into the learned hidden state vector. Here, the label of the fluid flow signal may include (classification) information indicating (or representing) the physical properties of the fluid.

[0048] And, in explaining the present disclosure, training an artificial neural network may mean updating a plurality of parameters used in the operation of the artificial neural network to optimal values.

[0049] The device will specifically describe how to process real-time flow signals using an artificial neural network in FIGS. 3 to 8.

[0050] FIG. 2 is a block diagram briefly illustrating the configuration of a device for processing real-time flow signals using an artificial neural network according to one embodiment of the present disclosure.

[0051] As illustrated in FIG. 2, the device (100) may include a memory (110), a communication module (120), a display (130), an input module (140), and a processor (150). However, the present invention is not limited thereto, and the device (100) may have its software and hardware configurations modified / added / omitted within a range apparent to those skilled in the art according to the required operation.

[0052] The memory (110) can store data supporting various functions of the device (100), a program for the operation of the processor (150), input / output data (e.g., physical data of a fluid, parameters included in a reservoir vector, etc.), and a plurality of application programs (or applications) run on the device, data for the operation of the device (100), and commands. At least some of these application programs can be downloaded from an external server via wireless communication.

[0053] The memory (110) may include at least one type of storage medium among a flash memory type, a hard disk type, an SSD (Solid State Disk type), an SDD (Silicon Disk Drive type), a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk.

[0054] Additionally, the memory (110) may include a database that is separate from the device but connected via wire or wirelessly. That is, the database illustrated in FIG. 1 may be implemented as a component of the memory (110).

[0055] The communication module (120) may include one or more components that enable communication with an external device, for example, the communication module (120) may receive physical data of a fluid from at least one sensor.

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

[0057] The wired communication module may include various wired communication modules such as a Local Area Network (LAN) module, a Wide Area Network (WAN) module, or a Value Added Network (VAN) module, as well as various cable communication modules such as a Universal Serial Bus (USB), a High Definition Multimedia Interface (HDMI), a Digital Visual Interface (DVI), RS-232 (recommended standard 232), power line communication, or plain old telephone service (POTS).

[0058] The wireless communication module may include a wireless communication module that supports various wireless communication methods such as GSM (global System for Mobile Communication), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access), UMTS (universal mobile telecommunications system), TDMA (Time Division Multiple Access), LTE (Long Term Evolution), 4G, 5G, and 6G, in addition to a WiFi module and a Wireless Broadband module.

[0059] The display (130) displays (outputs) information processed in the device (100) (e.g., physical data of a fluid, data output through a reservoir vector, update results of various parameters, etc.). For example, the display may display execution screen information of an application program (e.g., an application) running in the device (100), or UI (User Interface) or GUI (Graphical User Interface) information according to such execution screen information.

[0060] The input module (140) is for receiving information from a user. When information is input through the user input unit, the processor (150) can control the operation of the device (100) to correspond to the input information.

[0061] The input module (140) may include hardware physical keys (e.g., buttons, dome switches, jog wheels, jog switches, etc. located on at least one of the front, rear, and side of the device) and software touch keys. For example, the touch keys may be formed of virtual keys, soft keys, or visual keys displayed on a touchscreen type display (130) through software processing, or may be formed of touch keys placed on a part other than the touchscreen. Meanwhile, the virtual keys or visual keys may be displayed on the touchscreen in various forms, and may be formed of, for example, graphics, text, icons, videos, or a combination thereof.

[0062] The processor (150) can control the overall operation and function of the device (100). Specifically, the processor (150) may be implemented as a memory that stores data regarding an algorithm for controlling the operation of components within the device (100) or a program that reproduces the algorithm, and at least one processor (not shown) that performs the aforementioned operation using the data stored in the memory. In this case, the memory and the processor may be implemented as separate chips. Alternatively, the memory and the processor may be implemented as a single chip.

[0063] In addition, the processor (150) can control any one or a combination of the components discussed above to implement various embodiments according to the present disclosure described in FIGS. 3 to 8 below on the device (100).

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

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

[0066] At this time, the device can acquire first physical data of a first fluid moving in the pipe through at least one sensor for a plurality of time units. The time unit may be set to the time it takes for the fluid to circulate once in the pipe, but is not limited thereto, and may be set by the device.

[0067] And, at least one sensor may include a pressure sensor, a flow sensor, a conductivity sensor, or an impedance sensor. And, the first physical data may include pressure, flow rate, conductivity, or impedance of the fluid.

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

[0069] Specifically, the device can perform principal component analysis (PCA) on the first physical data to reduce its dimensionality. PCA refers to a method of transforming high-dimensional data into low-dimensional data while preserving the distribution of the original data as much as possible.

[0070] The device can obtain a trend line by performing linear regression analysis on the dimensionally reduced first physical data. Linear regression analysis is an analytical method for modeling the relationship between variables as a linear function, assuming that the independent and dependent variables are linearly related. A trend line is a line representing the distribution or trend of the first physical data.

[0071] The device can remove trends (e.g., data that are a predefined distance from the trend line) from the first physical data whose dimensionality has been reduced based on a trend line, and obtain a standard deviation from the first physical data from which the trends have been removed. The device can perform normalization on 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 a hidden state vector. ESN refers to an artificial neural network generated based on a recurrent neural network (RNN).

[0074] RNN refers to a sequence model that processes input and output data in sequence units. In other words, RNNs can simultaneously transmit the output values ​​produced through activation functions at hidden layer nodes toward the output layer, and then transmit these results as inputs for the next calculation at the hidden layer nodes. Circular mapping can refer to performing operations between the first physical data and the hidden layer of the RNN.

[0075] An ESN is a network that randomly generates RNNs and then trains only the parameters (or weights) between nodes in the output layer and hidden layers. In other words, the weights between nodes in the input layer and nodes in the hidden layer, as well as the weights between nodes in the hidden layer, may not be trained.

[0076] Referring to Figure 4, non-linear data in the input layer can be operated on with the reservoir vector. The output data can be linear by linear mapping of the reservoir vector.

[0077] The device stores the reservoir vector for the first physical data ( ) can be used to perform circular mapping.

[0078] Type and size of input ( ) can be determined based on the type and size of the first physical data output through at least one sensor. The size of the reservoir vector ( ) is the total time step (or unit time). If defined as, must be satisfied.

[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 close to the upper limit of the device's memory or computational capacity.

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

[0081] [Mathematical Formula 1]

[0082]

[0083] [Equation 2]

[0084]

[0085] And, the value output from the reservoir vector (W out ) can be implemented as in mathematical expression 3.

[0086] [Equation 3]

[0087]

[0088] And, the loss function (or objective function) in the reservoir vector can be implemented as in mathematical expression 4.

[0089] [Equation 4]

[0090]

[0091] The device can update parameters between at least one reservoir vector and output data (S340).

[0092] The device can update a parameter between the at least one reservoir vector and the output data based on the data output through the at least one reservoir vector and first label data corresponding to the first physical data of the first fluid.

[0093] At this time, learning is not performed during the operation phase of the reservoir vector, and learning can only be performed during the phase where the reservoir vector is converted into output data. In other words, learning / updating can only be performed for parameters between the node corresponding to the reservoir vector and the node corresponding to the output node.

[0094] Specifically, the device may preset first label data for first physical data. Here, the first label may include information related to the physical properties of the first fluid (i.e., information that can represent the physical properties of the first fluid). The device may perform one-hot encoding on the first label to obtain the first label data.

[0095] Here, one-hot encoding refers to a method of expressing words in vector form by setting the size of the classification set to the dimension of the vector, assigning 1 to the index of the word to be expressed, and assigning 0 to other indices.

[0096] The device can calculate an error between data output through at least one reservoir vector and first label data corresponding to first physical data of the first fluid. The data output through the reservoir vector is also data in the form of a vector that has undergone one-hot encoding, and the device can calculate the error by calculating the distance / difference between the two vectors. The device can update parameters between the at least one reservoir vector and the output data so that the calculated error is reduced (i.e., minimized).

[0097] That is, the device can train an artificial neural network including a reservoir vector to output label data corresponding to physical data through physical data by updating parameters between at least one reservoir vector and output data in a supervised learning manner.

[0098] And, for example, based on the definition that a nonlinear operation is performed on at least one reservoir vector, the device can update the parameters based on the Stochastic Gradient Descent (SGD) algorithm or the Adaptive Moment Estimation (ADAM) algorithm. However, this is only one embodiment, and the device can update the parameters by performing a direct inverse matrix operation on at least one reservoir vector and / or output data.

[0099] As another example, a matrix used for linear mapping from reservoir vectors, based on the definition that linear operations are performed on at least one reservoir vector. can be implemented as Equation 5 with hyperparameter β.

[0100] [Equation 5]

[0101]

[0102] Based on β, the reservoir vector can be classified into training data and test data. The device can learn parameters through training data and then perform inference operations through test data.

[0103] The type and number of input data (i.e., first physical data) ( ) can be optimized by testing multiple combinations. And, all hyperparameters required to generate the reservoir vector ( ) can be used to derive the optimal value, such as random search, grid search, or Bayesian optimization.

[0104] As an example, FIG. 5 illustrates training data. For example, the device may acquire a first physical data set for a first fluid over a seven-day period, and each class may represent a first physical data set measured daily. The first physical data set can be easily distinguished by its mean and variance. In other words, the first physical data set may exhibit trends over time. FIG. 6 illustrates a case where normalization is performed on the training data in FIG. 5.

[0105] Figures 7 and 8 show the output results obtained by applying the reservoir vector to the learning data.

[0106] For example, the input data can be any combination of physical data (e.g., pressure and velocity of the first fluid, etc.). And, the size of the reservoir vector ( ) is 10 4 and W in The input scaling a can be 1, and the spectral radius of w ( )silver , and the leaking ratio (α) can be 1. The output data can be implemented in the form of binary values ​​of the first or last day or multi-class classification information.

[0107] The device can obtain a second label of the second physical data by applying at least one reservoir vector to the second physical data of the second fluid (S350). Here, the second fluid may mean a fluid different from the first fluid, i.e., a fluid not used for learning.

[0108] Based on acquiring second physical data of the fluid through at least one sensor, the device can input the second physical data into at least one reservoir vector to acquire a second label of the second physical data.

[0109] Specifically, the device can input second physical data of a second fluid into an artificial neural network including at least one reservoir vector, wherein parameters associated with at least one reservoir vector are optimized through learning, and thereby obtain second label data for the second physical data. By decoding the second label data, the device can obtain information related to the properties of the second physical data (i.e., information representing the properties of the second fluid).

[0110] As another example, the device may acquire second physical data of a second fluid for a plurality of time units via at least one sensor. The device may acquire a plurality of output data by applying a reservoir vector to each of the second physical data acquired for each of the plurality of time units. The device may acquire second label data, which is an average of the plurality of output data, and may decode at least one of the second label data or the reservoir vector to acquire information related to the properties of the second physical data.

[0111] Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing computer-executable instructions. The instructions may be stored in the form of program code, and when executed by a processor, may generate program modules to perform the operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

[0112] Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disks, flash memory, and optical data storage devices.

[0113] FIG. 9 is a block diagram illustrating a configuration of a flow characteristic monitoring device based on an artificial neural network according to one embodiment of the present disclosure.

[0114] Referring to FIG. 9, the artificial neural network-based flow characteristic monitoring device (900) may include an impedance analyzer (910) and a controller (920). However, the present invention is not limited thereto, and some components may be omitted from the flow characteristic monitoring device (900), or other general-purpose components may be further included in the flow characteristic monitoring device (900).

[0115] The impedance analyzer (910) may be at least one type of sensor described in FIGS. 1 to 9. The at least one sensor may include a pressure sensor, a flow sensor, a conductivity sensor, an impedance sensor, etc., and among them, the impedance sensor may be the impedance analyzer (910). For a specific structure of the impedance analyzer (910), reference may be made to FIGS. 10 and 11, which will be described later.

[0116] The controller (920) can perform processing steps for monitoring flow characteristics based on an artificial neural network. For example, the controller (920) can correspond to the device (100) described in FIGS. 1 to 9.

[0117] The impedance analyzer (910) may be configured to collect impedance data of first fluids constituting a learning dataset in a stationary or flowing state. For example, the first fluids may be electrode slurries in various states. The first fluids may be in a stationary or flowing state by a stationary vessel and a flowing vessel of the impedance analyzer (910). In this regard, the transducer body of the impedance analyzer (910) may measure impedance data by applying AC power to the first fluids.

[0118] The controller (920) may be configured to perform normalization on the resistance and reactance components of the impedance data to generate resistance data and reactance data. The impedance data may include a real part and an imaginary part, and the real part may correspond to the resistance component of the first fluids, and the imaginary part may correspond to the reactance component of the first fluids. As described in FIGS. 1 to 9, the normalization may be performed using the mean and standard deviation of the resistance / reactance components. Meanwhile, as described in FIGS. 1 to 9, principal component analysis and linear regression analysis may be performed during the normalization process.

[0119] The controller (920) may be configured to perform recurrent mapping on resistance data and reactance data to generate at least one reservoir vector of an artificial neural network. Recurrent mapping may mean performing an operation on a layer of an artificial neural network such as an RNN. An artificial neural network (ANN) may be an artificial intelligence model trained to perform a given function, and at least one reservoir vector may include a state vector of a hidden layer. At least one reservoir vector may receive resistance data and reactance data as input and output label data, and the label data may represent a flow characteristic or a physical property of a target fluid.

[0120] The controller (920) may be configured to update model parameters of an artificial neural network by comparing output data of at least one reservoir vector with first label data corresponding to resistance components and reactance components to generate optimal parameters. The first label data may be prepared in advance to correspond to flow characteristics or physical properties of first fluids. The model parameters of the artificial neural network may be updated in a direction that outputs the first label data when resistance / reactance data is input. When the artificial neural network is trained through parameter updates, at least one reservoir vector may classify / infer flow characteristics or physical properties of the fluid based on the input data.

[0121] The controller (920) may be configured to obtain a second label representing the flow characteristics of the second fluid by applying at least one reservoir vector based on optimal parameters to the impedance data of the second fluid to be analyzed. When the model parameters are updated and the learning of the artificial neural network is completed, the at least one reservoir vector may classify the flow characteristics or physical properties of the fluid based on the impedance information of the fluid. The second fluid may be a fluid that has not been used for updating / learning and may be a target fluid that the user wishes to identify. Once the second label is obtained, the flow characteristics or physical properties of the target fluid may be identified from it.

[0122] According to an embodiment, the controller (920) may be configured to measure the resistance and reactance components of the first fluid through an interval sweep that records a plurality of data points at a fixed frequency. The interval sweep may include a technique of recording a large number of data points through impedance sweeping at a fixed frequency, such as 1 kHz, 50 kHz, or 500 kHz. Once the impedance of the target fluid is measured through the interval sweep, the impedance may be separated into a real part and an imaginary part, and the resistance and reactance components may be measured. Alternatively, a frequency sweep method that uses a variable frequency may be utilized instead of an interval sweep that uses a fixed frequency.

[0123] According to an embodiment, the fixed frequency may be selected within a range of 0.1 kHz to 10 kHz, and the recording speed of multiple data points may be selected within a range of 10 to 1000 data points per second. The numerical ranges of the fixed frequency and recording speed may be selected based on the performance of classifying the second label of the second fluid to be analyzed. Preferably, the fixed frequency may be 1 kHz, and the recording speed may be 100 data points per second. Referring to FIGS. 12 to 19, which will be described later, this can be described in detail.

[0124] According to an embodiment, the first fluids constituting the learning dataset may be prepared in advance to have different first labels corresponding to different flow characteristics, and each of the different first labels may correspond to first label data. For example, if the first fluid is an electrode slurry, the first fluid may have different flow characteristics or physical properties depending on the type and duration of mixing of the electrode slurry. The flow characteristics or physical properties may include local characteristics such as fluid velocity, fluid strength, and fluid shape. To generate the learning dataset, after preparing first fluids having different flow characteristics, first labels may be assigned to them, and the impedance characteristics they exhibit may be analyzed.

[0125] In an embodiment, the different flow characteristics include velocity, stress, and morphology, and the different flow characteristics can be determined based on the pre-treatment time, main treatment time, and whether or not a post-treatment is performed for the first fluids. For example, the pre-treatment, main treatment, and post-treatment for the first fluids can be pre-treatment, main treatment, and post-treatment for the mixing process for the electrode slurry. The pre-treatment, main treatment, and post-treatment can be performed as different types of mixing operations. Meanwhile, the physical properties / flow characteristics of the electrode slurry can also vary depending on how long the mixing process is performed. When the first fluids undergo different types of mixing processes, they can have different combinations of fluid velocities, fluid stresses, and fluid morphologies.

[0126] 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), and the at least one hidden state vector may be configured to output output data linearly. As described above, even if non-linear data is input, the artificial neural network may generate linear output data by linear mapping of the at least one reservoir vector.

[0127] According to an embodiment, the first fluids and the second fluid may include an electrode slurry including an electrode active material, a conductive agent, and a dispersant. For example, in the first fluids 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 agent 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 exemplary, and specific values ​​may vary depending on changes in electrode design or other specifications.

[0128] According to an embodiment, the controller (920) may be configured to perform supervised learning to minimize the error between the first label data corresponding to the resistance component and the reactance component and the output data of at least one reservoir vector. In the process of updating the model parameters of the artificial neural network, supervised learning may be performed using the relationship between the properties of the first fluids prepared in advance and the first label data as training data. To this end, the difference / error between the output data of at least one reservoir vector and the first label data may be set as a loss function, and machine learning may be performed in a direction to minimize the loss function.

[0129] FIG. 10 and FIG. 11 are drawings for explaining a specific structure of a flow characteristic monitoring device based on an artificial neural network according to one embodiment of the present disclosure.

[0130] Referring to FIG. 10, a transducer body (1000) of an impedance analyzer (910) may be illustrated. For example, the transducer body (1000) may be a product of HIOKI, but is not limited thereto.

[0131] The transducer body (1000) can apply an input signal for impedance measurement to a target fluid in a stationary or flowing state, and can obtain impedance information by sensing changes in the electrical characteristics of the target fluid.

[0132] Referring to FIG. 11, a stationary vessel (1110) configured to maintain a target fluid in a stationary state and a flowing vessel (1120) configured to maintain a target fluid in a flowing state can be illustrated.

[0133] The stationary vessel (1110) can accommodate 2 to 3 ml of a target fluid, such as first fluids or second fluids. The signal input / output ports of the transducer body (1000) can be connected to the stationary vessel (1110), through which impedance information of the stationary fluid can be collected. The flowing vessel (1120) can be installed on a pipe through which the target fluid flows, as in FIG. 1.

[0134] FIGS. 12 to 15 are diagrams for explaining resistance data and reactance data generated by performing normalization on the resistance component and reactance component of impedance data according to one embodiment of the present disclosure.

[0135] Referring to FIG. 12, resistance data (1210) and reactance data (1220) may be illustrated for a case where an interval sweep is performed at a fixed frequency of 1 kHz. The resistance data (1210) and reactance data (1220) may be expressed in the form of a graph showing the resistance / reactance value on the vertical axis versus the number of sweeps on the horizontal axis.

[0136] Resistance data (1210) and reactance data (1220) can be generated for five different first fluids. The five first fluids can be electrode slurries containing an active material, a conductive material, a dispersant, etc. Depending on the specific aspect of performing the slurry mixing process for the electrode slurry, the five first fluids can be distinguished, and the flow characteristics or physical properties of the five first fluids can vary.

[0137] Since the five first fluids have different flow characteristics or properties, the resistance data (1210) and reactance data (1220) may appear in different patterns as illustrated. An artificial neural network implemented with an ESN, etc., can learn data patterns such as the resistance data (1210) and reactance data (1220) and reflect them in model parameters. When the fluid starts to flow, both the resistance and reactance decrease rapidly, and the range of variation in the resistance and reactance may increase during the flow. If the flow continues for a long time, the range of variation may decrease, and when the flow stops, the resistance and reactance values ​​may recover.

[0138] Typically, data patterns exhibit chaotic vibration patterns, and machine learning can learn the unique features of these patterns. After parameter updates are complete, an artificial neural network can receive resistance or reactance patterns and classify / infer the flow characteristics or physical properties of the fluid. Experimental results show that reactance data (1220) exhibits more distinct patterns than resistance data (1210), and that learning / inference performance is also higher.

[0139] Referring to FIG. 13, resistance data (1310) and reactance data (1320) may be illustrated for a case where an interval sweep is performed at a fixed frequency of 50 kHz, and referring to FIG. 14, resistance data (1410) and reactance data (1420) may be illustrated for a case where an interval sweep is performed at a fixed frequency of 500 kHz, and referring to FIG. 15, resistance data (1510) and reactance data (1520) may be illustrated for a case where an interval sweep is performed at a fixed frequency of 8 MHz.

[0140] Looking at the resistance data and reactance data of FIGS. 13 to 15, it can be confirmed that they are more irregular and difficult to distinguish than the resistance data (1210) and reactance data (1220) of FIG. 12. Based on this, it can be confirmed that the highest learning / inference performance is shown at a fixed frequency of 0.1 kHz to 10 kHz, and preferably, a fixed frequency of 1 kHz can be utilized. In the case of 8 MHz of FIG. 15, a recording speed of 70 times per second was utilized, and in addition, the data of FIGS. 12 to 14 were measured at a recording speed of 100 times per second.

[0141] FIGS. 16 to 19 are diagrams for explaining the results of model learning performed with impedance data by interval sweep according to a numerical value of a fixed frequency according to one embodiment of the present disclosure.

[0142] Referring to FIG. 16, a graph (1600) may be illustrated showing the classification accuracy of an artificial neural network when an interval sweep is performed at a fixed frequency of 1 kHz. Similarly, FIGS. 17 to 19 may illustrate graphs (1700, 1800, 1900) for 50 kHz, 500 kHz, and 8 MHz, respectively.

[0143] Referring to the graph (1600), the number of training data (n train : Model performance may vary depending on the number of training data (n) (250, 500, 1000, 2000, 4000). For example, the number of training data (n) train ) can correspond to the number of first fluids. Meanwhile, the number of identical learning data (n train ), the average number (n) avg ) can be confirmed that the higher the model performance, the higher the performance.

[0144] Unlike graph (1600), it can be confirmed that it is difficult to show stable performance in graphs (1700, 1800, 1900). That is, among the fixed frequencies of the interval sweep, the section including 1 kHz shows the highest performance, and based on this, the fixed frequency can be selected within the range of 0.1 kHz or more and 10 kHz or less.

[0145] FIG. 20 is a flowchart illustrating steps of a method for monitoring flow characteristics based on an artificial neural network according to one embodiment of the present disclosure.

[0146] Referring to FIG. 20, the artificial neural network-based flow characteristic monitoring method (2000) may include steps (2010) to (2050). However, the present invention is not limited thereto, and some steps may be omitted from the flow characteristic monitoring method (2000), or other general steps may be further included in the flow characteristic monitoring method (2000), and steps (2010) to (2050) may be performed in a different order than the illustrated order.

[0147] The artificial neural network-based flow characteristic monitoring method (2000) may be composed of steps that are processed in a time-series manner by the artificial neural network-based flow characteristic monitoring device (900). Therefore, the contents described above for the flow characteristic monitoring device (900) may also be equally applied to the flow characteristic monitoring method (2000) described below.

[0148] The artificial neural network-based flow characteristic monitoring method (2000) can be performed by the impedance analyzer (910) and controller (920) of the artificial neural network-based flow characteristic monitoring device (900).

[0149] In step (2010), the flow characteristic monitoring device (900) can perform a step of collecting impedance data of first fluids constituting a learning data set in a stationary state or a flowing state.

[0150] In step (2020), the flow characteristic monitoring device (900) may perform a step of generating resistance data and reactance data by performing normalization on the resistance component and reactance component of the impedance data.

[0151] In step (2030), the flow characteristic monitoring device (900) may perform a step of generating at least one reservoir vector of an artificial neural network by performing recurrent mapping on resistance data and reactance data.

[0152] In step (2040), the flow characteristic monitoring device (900) can perform a step of generating optimal parameters by comparing the output data of at least one reservoir vector with first label data corresponding to the resistance component and the reactance component and updating the model parameters of the artificial neural network.

[0153] In step (2050), the flow characteristic monitoring device (900) can perform a step of obtaining a second label representing the flow characteristic of the second fluid by applying at least one reservoir vector based on an optimal parameter to the impedance data of the second fluid to be analyzed.

[0154] The method for monitoring flow characteristics based on an artificial neural network (2000) may be implemented in the form of a recording medium storing computer-executable instructions. The instructions may be stored in the form of program code, and when executed by a processor, may generate program modules to perform the operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

[0155] Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disks, flash memory, and optical data storage devices.

[0156] The method for monitoring flow characteristics based on an artificial neural network (2000) can be implemented in the form of a computer program. The computer program can be stored in a computer-readable recording medium in combination with an artificial neural network-based flow characteristic monitoring device (900) to execute the method for monitoring flow characteristics based on an artificial neural network (2000).

[0157] The disclosed embodiments have been described with reference to the attached drawings as described above. Those skilled in the art will understand that the present disclosure can be implemented in forms other than the disclosed embodiments without altering the technical spirit or essential features of the present disclosure. The disclosed embodiments are illustrative and should not be construed as limiting.

[0158] [Explanation of symbols]

[0159] 100: Device

[0160] 110: Memory

[0161] 120: Communication module

[0162] 130: Display

[0163] 140: Processor

[0164] 900: Artificial neural network-based flow characteristic monitoring device

Claims

1. In a method for monitoring flow characteristics based on an artificial neural network, performed by a flow characteristic monitoring device, A step of collecting impedance data of first fluids constituting a learning dataset in a stationary or flowing state; A step of generating resistance data and reactance data by performing normalization on the resistance component and reactance component of the above impedance data; A step of generating at least one reservoir vector of an artificial neural network by performing recurrent mapping on the resistance data and the reactance data; A step of generating optimal parameters by comparing the output data of the at least one reservoir vector with first label data corresponding to the resistance component and the reactance component and updating the model parameters of the artificial neural network; and A method for monitoring flow characteristics based on an artificial neural network, comprising: obtaining a second label representing the flow characteristics of the second fluid by applying at least one reservoir vector based on the optimal parameter to the impedance data of the second fluid to be analyzed; 2. In paragraph 1, The steps of collecting the above impedance data are: An artificial neural network-based flow characteristic monitoring method, comprising the step of measuring the resistance component and the reactance component of the first fluid through an interval sweep that records a plurality of data points at a fixed frequency.

3. In paragraph 2, The above fixed frequency is selected within the range of 0.1 kHz to 10 kHz, A method for monitoring flow characteristics based on an artificial neural network, wherein the recording speed of the above-mentioned plurality of data points is selected within a range of 10 per second or more and 1000 per second or less.

4. In paragraph 1, A method for monitoring flow characteristics based on an artificial neural network, wherein the first fluids constituting the learning dataset are prepared in advance to have different first labels corresponding to different flow characteristics, and each of the different first labels corresponds to the first label data.

5. In paragraph 4, The above different flow characteristics include velocity, stress and morphology, An artificial neural network-based flow characteristic monitoring method, wherein the above different flow characteristics are determined based on the pre-processing time, main processing time, and whether or not post-processing is performed for the first fluids.

6. In paragraph 1, The above artificial neural network includes an echo state network (ESN), wherein said at least one reservoir vector comprises at least one hidden state vector of said echo state network (ESN), A method for monitoring flow characteristics based on an artificial neural network, wherein at least one of the hidden state vectors is configured to linearly output the output data.

7. In paragraph 1, An artificial neural network-based method for monitoring flow characteristics, wherein the first fluids and the second fluid include an electrode slurry containing an electrode active material, a conductive agent, and a dispersant.

8. In paragraph 1, The step of generating the above optimal parameters is: A method for monitoring flow characteristics based on an artificial neural network, comprising a step of performing supervised learning to minimize an error between first label data corresponding to the resistance component and the reactance component and output data of the at least one reservoir vector.

9. In paragraph 1, An artificial neural network-based method for monitoring flow characteristics, comprising a step of further collecting pressure data of first fluids.

10. In paragraph 9, A step of performing normalization on the above pressure data and performing recurrent mapping on the normalized pressure data to generate at least one reservoir vector of an artificial neural network; A step of generating optimal parameters by updating model parameters of the artificial neural network by comparing data linearly output through at least one reservoir vector and first label data corresponding to pressure data of the first fluid; and A method for monitoring flow characteristics based on an artificial neural network, comprising: obtaining a second label representing the flow characteristics of the second fluid by applying at least one reservoir vector to the pressure data of the second fluid to be analyzed based on the optimal parameter; 11. In a flow characteristic monitoring device based on an artificial neural network, An impedance analyzer configured to collect impedance data of a first fluid constituting a learning dataset in a stationary or flowing state; and Perform normalization on the resistance and reactance components of the above impedance data to generate resistance data and reactance data, Performing recurrent mapping on the resistance data and the reactance data to generate at least one reservoir vector of an artificial neural network, Generating optimal parameters by comparing the output data of at least one reservoir vector with first label data corresponding to the resistance component and the reactance component and updating the model parameters of the artificial neural network, A flow characteristic monitoring device based on an artificial neural network, comprising: a controller configured to obtain a second label representing the flow characteristic of the second fluid by applying the at least one reservoir vector based on the optimal parameter to the impedance data of the second fluid to be analyzed; 12. A computer program stored in a computer-readable recording medium to execute a flow characteristic monitoring method according to any one of claims 1 to 10, in combination with a flow characteristic monitoring device based on an artificial neural network.

Images