KR-20260067646-A - Method of predicting Optimal Initial Composition and Process Condition in Alloying based on Directed Energy Deposition

KR20260067646AKR 20260067646 AKR20260067646 AKR 20260067646AKR-20260067646-A

Abstract

A method is disclosed for deriving optimal initial composition and process conditions capable of predicting a target final composition to solve the problem in which the final composition of an alloy formed by a direct welding method differs from the initial composition due to the evaporation of metal. Training is performed on a machine learning model, and the performance of the training is verified using holdout validation and k-fold ensemble techniques. Randomly generated input sets through random search are input into the trained machine learning model, and optimal initial composition and process conditions capable of achieving the target composition are predicted through judgment using reference values.

Inventors

이병주
왕재민
김형섭
김은성

Assignees

포항공과대학교 산학협력단

Dates

Publication Date: 20260513
Application Date: 20241106

Claims (17)

A step of training a machine learning model using an input set consisting of an initial composition and process conditions and the final composition of a product according to the input set; and A method for predicting optimal initial composition and process conditions for alloying, comprising the step of predicting an optimal input set to derive a target final composition through random search and the above-mentioned trained machine learning model.
In paragraph 1, the step of training the machine learning model A step of collecting a data set composed of the above input set and the above final composition; A step of performing preprocessing through normalization of the above input set; and A method for predicting optimal initial composition and process conditions for alloying, comprising the step of training the machine learning model using the normalized input set.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in paragraph 2, the machine learning model is composed of a plurality of sub-learning models distinguished by algorithm.
In paragraph 3, the training of the above-mentioned one sub-learning model is A step of splitting the above normalized input sets into test sets and training sets; A step of assigning the above training sets to k folds, dividing each fold into N cases, and training the sub-learning model such that the k folds are used as a validation set at least once; and A method for predicting optimal initial composition and process conditions for alloying, characterized by including a step of verifying the performance of N×k individual learning models generated through training using the test set.
A method for predicting the optimal initial composition and process conditions for alloying, wherein, in paragraph 4, the verification of the performance of the individual learning models is characterized by obtaining an average value of the predicted values of the final composition generated from the individual learning models, setting the standard deviation of the predicted values as an error range, and comparing the average value of the predicted values with the final composition corresponding to the test set.
In claim 1, the step of predicting the optimal input set is A step of inputting whether the above initial composition and the above process conditions are fixed, a fixed value, or other options; A step of randomly generating the above input set through random search; A step of inputting the above randomly generated input set into the above-mentioned trained machine learning model to derive a predicted value of the final composition; and A method for predicting the optimal initial composition and process conditions for alloying, characterized by including the step of deriving a reference value using the following Equation 1 for the predicted final composition and selecting the optimal input set from the randomly generated input set through comparison of the reference value. [Formula 1] In the above Equation 1, Fp represents the reference value, s represents the quality threshold, μ represents the average of the predicted values, x represents the target final composition, and σ represents the standard deviation of the predicted values. Additionally, max indicates that the larger value among the quality threshold s, the average of the predicted values μ, and the difference between the target final composition x is selected. Furthermore, the units of the quality threshold s, the average of the predicted values μ, and the target final composition x are %.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in claim 6, the other options are the number of the optimal input sets and the number of random searches.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in claim 6, among the arbitrarily generated input sets, the input set with the lower reference value is selected.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in claim 6, the quality threshold is further included in the other options, and the quality threshold is a criterion for judging overfitting.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in claim 6, the step of selecting the optimal input set further includes the step of not selecting a similar input set among the selected input sets and selecting an input set with a lower reference value among the similar input sets.
A method for predicting initial composition and process conditions for alloying using a machine learning model composed of multiple said sub-learning models, wherein one algorithm constitutes a sub-learning model, A step of training the sub-learning models using a dataset consisting of an input set of initial composition and process conditions and a final composition, and validating the trained sub-learning models using holdout validation and k-fold ensemble techniques; A step of randomly generating input sets greater than the number of optimal input sets derived through random search, and inputting the randomly generated input sets into the trained sub-learning models to derive predicted values of the final composition; and The method includes the step of deriving a reference value from the predicted values of the final composition derived from multiple sub-learning models for a specific input set, and selecting the optimal input set through the comparison of reference values of different input sets. A method for predicting the optimal initial composition and process conditions for alloying, characterized in that the above reference value is derived from the mean and standard deviation of the predicted values of the final composition of the above-mentioned trained sub-learning models derived from the above-mentioned input set, and is represented by the following Equation 1. [Formula 1] In the above Equation 1, Fp represents the reference value, s represents the quality threshold, μ represents the average of the predicted values, x represents the target final composition, and σ represents the standard deviation of the predicted values. Additionally, max indicates that the larger value among the quality threshold s, the average of the predicted values μ, and the difference between the target final composition x is selected. Furthermore, the units of the quality threshold s, the average of the predicted values μ, and the target final composition x are %.
In Clause 11, prior to the arbitrary generation of the above input sets A method for predicting optimal initial composition and process conditions for alloying, characterized by further including a step of receiving input regarding whether the initial composition and process conditions are fixed, fixed values, or other options, wherein the other options are the number of optimal input sets and the number of random searches.
In paragraph 12, after selecting the optimal input set, the method further includes a step of determining whether the random search has reached the number of random searches set in the other options. A method for predicting optimal initial composition and process conditions for alloying, characterized by proceeding with random generation of an input set by random search when the number of random searches mentioned above has not been reached.
A method for predicting optimal initial composition and process conditions for alloying, characterized in that, in claim 12, the above other options further include the above quality threshold, and the above quality threshold is a criterion for judging overfitting.
A method for predicting initial composition and process conditions for alloying, wherein, in claim 11, the step of selecting the optimal input set is characterized by comparing the reference values and selecting the input set with the smaller reference value.
A method for predicting initial composition and process conditions for alloying according to claim 15, characterized in that, after comparing the reference values, the input set with the smaller corresponding reference value among the input sets having similar sizes among the selected input sets is selected, and the remaining input sets are not selected.
A method for predicting initial composition and process conditions for alloying, characterized in that, in claim 16, the determination of input sets having similar sizes follows the following Equation 2. [Equation 2] In the above Equation 2, a i is the normalized input set of predicted value A, and b i is the normalized input set of predicted value B. Also, p represents the number of inputs in the input set.

Description

Method of predicting Optimal Initial Composition and Process Condition in Alloying based on Directed Energy Deposition The present invention relates to an alloying method using a direct welding method, and more specifically, to a method for predicting compositional variations caused by the evaporation of metal powder in an environment where various types of metal powder are used, and for finding an initial composition and process conditions, which are input sets having a desired composition. Directed Energy Deposition (DED) is a 3D metal printing technology that involves melting various metal powders using an energy beam and layering them. DED is used to manufacture complex metal structures or parts. Molten metal is supplied from the printing head upon the application of energy, and various metals are deposited. Alloying occurs during the process of melting the various types of metal powders. However, there is a problem in that the composition of the final alloy differs from the initially designed alloy composition during the melting process of various types of metal powders. This is due to the fact that metal powders have different melting points and vapor pressures depending on their type. In other words, since the type and amount of metal evaporating at a specific temperature vary during the process of supplying and melting the metal powders, the final product exhibits a composition different from that of the initially input metal. To address this, attempts are made to predict changes in composition through mathematical modeling. However, predicting compositional changes via mathematical modeling fails to ensure sufficient accuracy and universality. In particular, predicting changes in composition for various types of metal powders is a very difficult task. To predict compositional changes through mathematical modeling, factors such as heat and fluid flow, molten pool dynamics, kinetics, and the Marangoni effect must all be considered. Furthermore, due to the large number of factors that must be taken into account, a problem arises regarding the reproducibility of the mathematical modeling. FIG. 1 is a flowchart illustrating a method for predicting initial composition and process conditions to manufacture a product having a final composition according to a preferred embodiment of the present invention. FIG. 2 is a flowchart illustrating the steps of training the machine learning model of FIG. 1 according to a preferred embodiment of the present invention. FIG. 3 is a schematic diagram illustrating the machine learning model of FIG. 2 according to a preferred embodiment of the present invention. FIG. 4 is a schematic diagram illustrating one sub-learning model of FIG. 3 according to a preferred embodiment of the present invention. FIG. 5 is a flowchart illustrating the steps for deriving an initial composition and process conditions predicted to have a specific composition of FIG. 1 according to a preferred embodiment of the present invention. Figure 6 is a graph comparing the final composition predicted by the input set designed according to the experimental example of the present invention with the final composition of samples actually produced using the input sets. The present invention is susceptible to various modifications and may take various forms; therefore, specific embodiments are illustrated in the drawings and described in detail in the text. However, this is not intended to limit the invention to the specific disclosed forms, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Similar reference numerals have been used for similar components in the description of each drawing. Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application. Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. Examples FIG. 1 is a flowchart illustrating a method for predicting initial composition and process conditions to manufacture a product having a final composition according to a preferred embodiment of the present invention. Referring to Figure 1, training is performed on a machine learning model. In principle, this is a process in which the machine learning model learns a pattern through data and adjusts to optimally make predictions on new data based on this. The inputs are the initial composition and process conditions, and the output is set to the final composition. That is, the final composition