US-12626203-B2 - Method for generating prediction model for supply lead time of parts

US12626203B2US 12626203 B2US12626203 B2US 12626203B2US-12626203-B2

Abstract

Provided is a method for generating a lead time prediction model including: receiving input data from a user; identifying a final part corresponding to the input data; identifying one or more components that constitute the final part; classifying the identified one or more components into in-house production parts, which are produced in-house, and ordered parts, which are procured from suppliers; determining a lead time for the in-house production parts; obtaining first data including at least price data and historical lead time data, for each of the ordered parts; generating a model for generating a predicted lead time for at least one of the final part and the ordered parts, based on the obtained first data; and optimizing the model based on at least one of fixed costs, inventory costs, ordering costs, backlog costs, and demand loss costs.

Inventors

Seock Kyu YOO
Cherl Hong PARK
KyungSik LEE
Sungwon HONG
Hojin JUNG

Assignees

VMS Solutions Co. Ltd.

Dates

Publication Date: 20260512
Application Date: 20241101
Priority Date: 20231108

Claims (7)

1 . A method for implementing a model for generating a predicted lead time, comprising: receiving input data from a user; identifying a final part corresponding to the input data; identifying one or more components that constitute the final part; classifying the identified one or more components into in-house production parts, which are produced in-house, and ordered parts, which are procured from suppliers; determining a lead time for the in-house production parts; obtaining, for each of the ordered parts, first data comprising at least price data and historical lead time data by: (i) operating an artificial intelligence (AI)-based knowledge mining robot including a trained classification model and a crawler to classify web resources as related or unrelated to the ordered part based on request information including at least a part identifier and data to be acquired, (ii) crawling and indexing data from the web resources classified as related and storing crawled unstructured data and index information in a raw data archive, and (iii) structuring, by a data analysis engine, the unstructured data into part-related fields including at least time-based price data and historical lead time data and storing the structured data in a parts knowledge database (KDB); generating candidate feature expressions by combining the price data with arithmetic operators; generating second data by removing an expression in which operational units do not match among the candidate feature expressions; generating a model for generating the predicted lead time for at least one of the final part and the ordered parts by training, using a training dataset including the first data and the second data, a regression model comprising a genetic-programming-based decision tree, and iteratively updating the second data by: determining a mean square error (MSE) improvement attributable to each candidate feature expression, retaining a candidate feature expression when the MSE improvement exceeds a reference value, and otherwise generating third data using mutation and/or crossover operations and selecting surviving features through a tournament; and optimizing the model for generating the predicted lead time based on at least one of fixed costs, inventory costs, ordering costs, backlog costs, and demand loss costs.
2 . The method of claim 1 , wherein the optimizing the model comprises: optimizing the model by calculating a solution minimizing the following Equation: ∑ i ∈ ℐ ∑ t ∈ ℋ ∑ u ∈ 𝒰 i f iu ⁢ p itu + ∑ s ∈ 𝒮 p s ( ∑ i ∈ ℐ ∑ t ∈ ℋ ( h i ⁢ I it s ⁢ ∑ u ∈ 𝒰 i υ iu ⁢ Q itu ) + ( ∑ t = 1 T - 1 b i · B t s + e i · B T s ) ) wherein the first term (f iu p itu ) refers to a fixed cost component, and comprises the fixed cost (f iu ) when product i is procured from supplier u, and the contract initiation status (p itu ) with supplier u in period t regarding product i, wherein the second term (h i l it ) refers to an inventory cost component, and comprises the unit inventory cost (h i ) for product i and the inventory level (I it ) of product i in period t under scenario s, wherein the third term (v iu Q itu ) refers to an ordering cost component, and comprises the unit production/order cost (v iu ) when product i is procured from supplier u, and the order quantity (Q itu ) of product i from supplier u in period t, wherein the fourth term·(b i ·B t ) refers to the backlog cost component, and comprises the unit backlog cost (b i ) for finished product i and the backlog amount (B t ) of finished product i in period t under scenario s, wherein the fifth term (e i ·B T ) refers to the demand loss cost component, and comprises the unit demand loss cost (e i ) for finished product i and the backlog amount (B T ) of finished product i in period T under scenario s.
3 . The method of claim 2 , wherein the optimizing the model comprises: minimizing the Equation considering at least one of the following constraints: number of suppliers, supply quantity, minimum contract duration, maximum contract duration, outsourced component inventory balance, produced component inventory balance, end-item inventory balance, production and ordering capacity, initial inventory, initial backlog, supplier selection, and non-negativity constraints.
4 . The method of claim 1 , further comprising: displaying the predicted lead time generated by the model on a user interface.
5 . The method of claim 4 , wherein the displaying the predicted lead time on the user interface comprises: presenting the predicted lead time for the ordered parts over a predetermined period in the form of a graph; and in response to receiving a user selection of the predicted lead time displayed on the graph, displaying a probability of the predicted lead time as a pop-up in the form of a graph.
6 . The method of claim 1 , wherein the generating a model for generating the predicted lead time for at least one of the final part and the ordered parts, comprises: generating the model by using Sample Average Approximation (SAA) for a lead time scenario in the model, wherein the lead time scenario comprises a vector representing lead time values for all suppliers of each ordered part.
7 . A non-transitory computer-readable recording medium, wherein the non-transitory computer-readable recording medium comprises computer-executable instructions, and the instructions, when executed by a processor, perform operations comprising: receiving input data from a user; identifying a final part corresponding to the input data; identifying one or more components that constitute the final part; classifying the identified one or more components into in-house production parts, which are produced in-house, and ordered parts, which are procured from suppliers; determining a lead time for the in-house production parts; obtaining, for each of the ordered parts, first data comprising at least price data and historical lead time data by: (i) operating an artificial intelligence (AI)-based knowledge mining robot including a trained classification model and a crawler to classify web resources as related or unrelated to the ordered part based on request information including at least a part identifier and data to be acquired, (ii) crawling and indexing data from the web resources classified as related and storing crawled unstructured data and index information in a raw data archive, and (iii) structuring, by a data analysis engine, the unstructured data into part-related fields including at least time-based price data and historical lead time data and storing the structured data in a parts knowledge database (KDB); generating candidate feature expressions by combining the price data with arithmetic operators; generating second data by removing an expression in which operational units do not match among the candidate feature expressions; generating a model for generating the predicted lead time for at least one of the final part and the ordered parts by training, using a training dataset including the first data and the second data, a regression model comprising a genetic-programming-based decision tree, and iteratively updating the second data by: determining a mean square error (MSE) improvement attributable to each candidate feature expression, retaining a candidate feature expression when the MSE improvement exceeds a reference value, and otherwise generating third data using mutation and/or crossover operations and selecting surviving features through a tournament; and optimizing the model for generating the predicted lead time based on at least one of fixed costs, inventory costs, ordering costs, backlog costs, and demand loss costs.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to and the benefit thereof under 35 U.S.C. § 119 to Korean Patent Application No. 10-2023-0153761 filed in the Korean Intellectual Property Office on Nov. 8, 2023, and Korean Patent Application, the entire contents of which are incorporated herein by reference. BACKGROUND (a) Field The disclosure relates to a method for generating a supply lead time prediction model for parts, and more specifically, to a method for generating a supply lead time prediction model for a final part, considering factors such as price fluctuations of one or more component parts constituting the final part. (b) Description of the Related Art A value chain refers to the process in which value is added by combining resources such as raw materials, labor, and capital to produce goods or services. The global value chain is an extension of the traditional value chain with the concept of globalization, and in the modern society where globalization is rapidly progressing, companies are expanding their value chains globally to bypass trade barriers or reduce manufacturing costs, making it difficult for any company to independently produce goods and services. In particular, most companies require multiple parts, such as raw materials or components, to manufacture a finished product (i.e., a final product manufactured by the company). Parts can be broadly classified into in-house produced parts, which are manufactured directly by the company, and ordered parts, which must be procured through purchasing or outsourcing from suppliers or subcontractors. While in-house produced parts have relatively fixed manufacturing times because the company manufactures them directly, the lead time (the time from ordering to production) required to supply ordered parts from suppliers or subcontractors is independent for each company. This is because suppliers or subcontractors that manufacture and supply the same or similar parts may have different component ratios for the component parts such as raw materials or components constituting the final part, or there may be differences in the prices required to purchase or manufacture the component parts. These price fluctuations of the component parts may affect the supply lead time of the final part. As a result, for companies that manufacture finished products, there is an increasing need for a method to predict the lead times of parts from each supplier or subcontractor and to derive an optimized production plan accordingly. SUMMARY Some embodiments may provide a method for generating a supply lead time prediction model for parts. According to an aspect of an embodiment, a method for generating a lead time prediction model may include: receiving input data from a user; identifying a final part corresponding to the input data; identifying one or more components that constitute the final part; classifying the identified one or more components into in-house production parts, which are produced in-house, and ordered parts, which are procured from suppliers; determining a lead time for the in-house production parts; obtaining first data including at least price data and historical lead time data, for each of the ordered parts; generating a model for generating a predicted lead time for at least one of the final part and the ordered parts, based on the obtained first data; and optimizing the model based on at least one of fixed costs, inventory costs, ordering costs, backlog costs, and demand loss costs. In some embodiments, the optimizing the model may include: optimizing the model by calculating a solution minimizing the following Equation: ∑i∈ℐ∑t∈ℋ∑u∈𝒰ifiu⁢pitu+∑s∈𝒮ps⁢(∑i∈ℐ∑t∈ℋ(hi⁢Iits+∑u∈𝒰iυiu⁢Qitu)+(∑t=1T-1bi·Bts+ei·BTs)), wherein the first term (fiupitu) refers to the fixed cost component, and includes the fixed cost (fiu) when product i is procured from supplier u, and the contract initiation status (pitu) with supplier u in period t regarding product i, wherein the second term (hilit) refers to the inventory cost component, and includes the unit inventory cost (hi) for product i and the inventory level (lit) of product i in period t under scenario s, wherein the third term (viuQitu) refers to the ordering cost component, and includes the unit production/order cost (viu) when product i is procured from supplier u, and the order quantity (Qitu) of product i from supplier u in period t, wherein the fourth term·(bi·Bt) refers to the backlog cost component, and includes the unit backlog cost (bi) for finished product i and the backlog amount (Bt) of finished product i in period t under scenario s, wherein the fifth term (ei·BT) refers to the demand loss cost component, and includes the unit demand loss cost (ei) for finished product i and the backlog amount (BT) of finished product i in period T under scenario s. In some embodiments, the optimizing the model may include: minimizing the Equation considering at least one of the following cons