KR-102963499-B1 - MACHINE LEARNING-BASED DISPENSER CONTROL METHOD BASED ON REAL-TIME MEASUREMENT OF FLOW RATE AND FLUID PROPERTIES

KR102963499B1KR 102963499 B1KR102963499 B1KR 102963499B1KR-102963499-B1

Abstract

A method for controlling a dispenser for dispensing fluid according to an embodiment of the present invention comprises: a step of generating a first artificial intelligence model by machine learning that receives sensor data including the flow rate, physical property information, pressure, humidity, and temperature of the fluid and outputs control data including the discharge pressure and position of the dispenser - wherein the physical property information includes viscosity, density, flow resistance, and shear sensitivity -; a step of inputting the first sensor data to the first artificial intelligence model to verify the output first control data; and a step of controlling the dispenser based on the first control data, wherein the objective function of the first artificial intelligence model may include a term for evaluating quantitative dispensing ability, a term for evaluating shape conformity, and a term for evaluating dispensing operation speed.

Inventors

홍주완
김기만

Assignees

주식회사 더다옴

Dates

Publication Date: 20260513
Application Date: 20260120

Claims (5)

In a method for controlling a fluid dispensing dispenser: A step of generating a first artificial intelligence model by machine learning that receives sensor data including the flow rate, physical property information, pressure, humidity, and temperature of the above fluid and outputs control data including the discharge pressure and position of the above dispenser - the physical property information includes viscosity, density, flow resistance, and shear sensitivity - ; A step of inputting first sensor data into the first artificial intelligence model and verifying the output first control data; A step of controlling the dispenser based on the first control data. Includes, A method for controlling a fluid dispensing dispenser, wherein the objective function of the first artificial intelligence model above includes a term for evaluating quantitative application, a term for evaluating shape matching, and a term for evaluating the application speed.
In paragraph 1, The above flow resistance is the value obtained by dividing the above pressure by the above flow rate, and A method for controlling a fluid dispensing dispenser, wherein the shear sensitivity is the ratio of the change in flow rate to the change in pressure.
In paragraph 1, the above method is: A step of collecting second sensor data and second control data corresponding to the work of a past skilled worker; A step of generating a first probability function that defines a state-action probability distribution by learning the second sensor data and the second control data; A step of generating a second probability function based on third sensor data and third control data output by the first artificial intelligence model that received the third sensor data; A step of determining a compensation function of the first artificial intelligence model as a weighted sum of a coating quantitative evaluation term, a shape matching evaluation term, a fluid waste evaluation term, a control stability evaluation term, and a path consistency evaluation term for a coating result based on the third control data; A step of training the first artificial intelligence model by performing reinforcement learning based on the above reward function; Includes more, A method for controlling a fluid dispensing dispenser, wherein the step of determining the above compensation function includes determining weights corresponding to a dispensing quantity evaluation term, a shape matching evaluation term, a fluid waste evaluation term, a control stability evaluation term, and a path consistency evaluation term by at least one of Maximum Entropy Inverse Reinforcement Learning, Gradient Descent, and Reinforcement Learning with Reward Shaping, and wherein the similarity between the first probability function and the second probability function and the predicted value of the compensation function are included in the objective function.
In paragraph 3, The weight of the above fluid waste evaluation term is determined to be higher as the absorption rate of the fluid increases, and The ratio of the weight of the coating quantitative evaluation term to the weight of the shape matching evaluation term is determined to be higher as the viscosity of the fluid increases, and The weight of the above path consistency evaluation term is determined to be higher as the hardness of the fluid increases, and A method for controlling a dispenser that applies fluid, wherein the weight of the shape matching evaluation term is determined to be higher as the electrical conductivity of the fluid increases.
In paragraph 1, the above method is: A step of determining the initial state and update rules of learning parameters for the first artificial intelligence model by learning sensor data, control data, and quality data for coating operations performed under multiple different fluid conditions - the quality data includes a coating quantity evaluation score, a shape matching evaluation score, a fluid waste evaluation score, a control stability evaluation score, and a path consistency evaluation score -; A step of updating the initial state and update rule of the above learning parameters through at least one technique among Model-Agnostic Meta-Learning (MAML), Reptile Meta-Learning, or Meta-SGD Meta-Learning; and A step of adapting the learning parameters for the first artificial intelligence model using the initial state and update rule of the updated learning parameters under new fluid conditions. Includes, The step of inputting first sensor data into the above-mentioned first artificial intelligence model to verify the output first control data is: A step of identifying a potential vector based on the first sensor data above; and Step of transforming the above potential vector; and It includes the step of verifying the first control data output based on the converted potential vector, and The physical property information of the above fluid includes viscosity, The step of transforming the above latent vector is: When the magnitude of the error in the amount of fluid applied by the dispenser controlled based on the first control data is greater than or equal to a threshold value, the latent variable z visc corresponding to viscosity among the plurality of latent variables included in the latent vector It includes a step of updating according to the rule z visc ←clip(z visc +η p ·(m act -m ref )+η i ·Em, z min , z max ), and m act is the actual application amount of the above fluid, and m ref is the ideal application amount of the above fluid, and Em is the cumulative value of the fluid application amount error, defined as Em(t)=λλ·(m act (t)-m ref (t)) at time t, and λ, η p , η i , z min , and z max are parameters determined through at least one of Model-Agnostic Meta-Learning (MAML), Reptile Meta-Learning, or Meta-SGD Meta-Learning, Method for controlling a dispenser that applies fluid.

Description

Machine Learning-Based Dispenser Control Method Based on Real-Time Measurement of Flow Rate and Fluid Properties The present invention relates to a method for controlling a dispenser that applies fluid based on machine learning, based on real-time measurement of flow rate and fluid properties. Fluid dispensers are used in various industrial fields, such as electronic devices, semiconductors, displays, and automotive parts, to precisely dispense fluids such as adhesives, sealants, and conductive pastes. The fluid dispensing process requires advanced control technology because it must simultaneously consider multiple performance factors, such as uniformity of dispensing amount, precision of the dispensing pattern shape, working speed, and productivity. Traditional dispensing operations often relied on the experience and intuition of skilled workers. While skilled workers intuitively grasped environmental conditions such as viscosity, temperature, and humidity as well as fluid properties to set optimal control signals, this method suffers from low reproducibility and is unsuitable for mass production. Furthermore, the difficulty in transferring a skilled worker's techniques to other operators leads to quality variations. While the development of automated dispenser control technology has made it possible to achieve consistent patterns and speeds, simultaneously satisfying quantitative accuracy and shape precision by reflecting various fluid characteristics remains a challenge. Since existing systems rely on pre-set control parameters, performance degradation occurs when fluid viscosity or environmental conditions change, and iterative experiments and calibration are required to respond to new conditions. Recently, attempts have been made to control fluid coating processes using artificial intelligence. Models that generate control signals based on sensor data have been studied, and methods that learn by evaluating coating results have also been proposed. However, existing approaches often target only a specific performance metric as a single objective, making it difficult to balance and optimize multiple metrics such as quantity, shape precision, and operation speed. Furthermore, their practicality is limited because new fluid conditions require large amounts of training data and long training times. FIG. 1 illustrates a fluid application system according to an embodiment of the present invention. FIG. 2 illustrates a dispenser control method performed in a fluid dispensing system according to an embodiment of the present invention. FIG. 3 illustrates the input and output variables of a first artificial intelligence model generated in a fluid application system according to an embodiment of the present invention. FIG. 4 illustrates a dispenser control method performed in a fluid dispensing system according to an embodiment of the present invention. FIG. 5 illustrates a dispenser control method performed in a fluid dispensing system according to an embodiment of the present invention. FIG. 6 illustrates a dispenser control method performed in a fluid dispensing system according to an embodiment of the present invention. As used herein, “comprises” and/or “comprising” do not exclude the presence or addition of one or more other components in addition to the components mentioned. The various embodiments of this specification and the terms used therein are not intended to limit the technical features described herein to specific embodiments and should be understood to include various modifications, equivalents, or substitutions of said embodiments. In connection with the description of the drawings, similar reference numerals may be used for similar or related components. The singular form of a noun corresponding to an item may include one or more of said items unless the relevant context clearly indicates otherwise. In this specification, each of phrases such as “A or B,” “at least one of A and B,” “at least one of A or B,” “A, B or C,” “at least one of A, B and C,” and “at least one of A, B, or C” may include any one of the items listed together in the corresponding phrase, or all possible combinations thereof. Terms such as “first,” “second,” or “first” or “second” may be used simply to distinguish a component from another component and do not limit the components in any other aspect (e.g., importance or order). Where any (e.g., first) component is referred to as “coupled” or “connected” to another (e.g., second) component, with or without the terms “functionally” or “communicationally,” it means that said component may be connected to said other component directly (e.g., wired), wirelessly, or through a third component. The term “module” as used in the various embodiments of this specification may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit, for example. A module may be a component formed integrally,