Ai-based manufacturing supply chain management method and system using the same

Abstract

The present invention relates to an AI-based manufacturing supply chain management method and a system utilizing the same. The method comprises a process in which a communication unit of a main server receives process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data from a plurality of external devices; a processor preprocesses this data to calculate the probability of process defects; integrates the defect probability with material-related information to calculate the supply risk for each material; structures field record data to analyze unstructured data; and transmits these analysis results to a manager terminal. Through the present invention, proactive manufacturing supply chain management is possible by integrating real-time defect prediction, material supply risk assessment, and unstructured data analysis, going beyond the simple reporting functions of existing manufacturing information systems. This enables the achievement of effects such as minimizing production stoppages, optimizing material utilization, and improving the speed of decision-making.

Description

Technology Field

[0001] The present invention relates to an AI-based manufacturing supply chain management method and a system utilizing the same. More specifically, it relates to a technology that predicts defects that may occur in high-precision manufacturing processes, such as semiconductors, secondary batteries, and precision assembly, in advance, and provides information necessary for determining material supply and demand based on the prediction results. The present invention includes a method and a system that collect data from manufacturing information systems (MES, WMS, ERP), perform AI-based analysis, and visually provide predicted defect risks to a manager. In particular, it belongs to the field of artificial intelligence technology that performs comprehensive manufacturing supply chain management by integrating time-series-based process data analysis, categorical material data analysis, and unstructured text data analysis. Background Technology

[0002] Recently, manufacturing information systems such as MES, WMS, and ERP are being widely implemented in the manufacturing industry to improve production efficiency and ensure quality stability; however, existing systems have several fundamental limitations. First, most systems are utilized only for simple reporting or post-hoc analysis, resulting in a significant lack of capabilities for real-time process status analysis and defect prediction. Second, a linkage system between defect prediction and material management strategies has not been established at all; consequently, predicted defect information is not utilized for practical operational decision-making, such as inventory management, order planning, and the review of substitute materials. Third, unstructured information generated on-site, such as equipment alarms and worker comments, is not systematically analyzed, leading to frequent instances where critical quality signals are missed. Fourth, in fields extremely sensitive to yield fluctuations, such as the semiconductor industry, even minor process issues have a massive impact on overall production schedules and material supply plans; however, there is currently a lack of an integrated management system capable of predicting and responding to these issues in advance. The problem to be solved

[0004] The problem that this invention aims to solve is to overcome the fundamental limitations of existing manufacturing information systems and provide practical manufacturing supply chain management capabilities. First, it seeks to implement a preemptive defect prediction system that utilizes AI analysis technology to detect quality risks by process at an early stage and predict them quantitatively. Second, it aims to provide an integrated material supply management system that can comprehensively assess risks such as inventory shortages, urgent orders, and irreplaceable materials by directly linking defect prediction results with material operation strategies. Third, it aims to establish comprehensive analytical capabilities capable of capturing quality signals that are easily missed by existing systems by structuring and analyzing unstructured text data generated in the field. Fourth, it aims to support rapid and accurate decision-making by providing managers with integrated prediction figures and related material information. Fifth, it aims to ensure stable implementation and operation through a non-intrusive system structure that can be linked with existing MES, ERP, and WMS without conflict.

[0005] The problems of the present invention are not limited to those mentioned above, and other unmentioned problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0006] The AI-based manufacturing supply chain management method of the present invention for solving the above problem comprises the following steps. A communication unit of a main server performs the step of receiving at least one of process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data from a plurality of external devices including at least one of a production management device, a material management device, a quality inspection device, and a field recording device. Subsequently, a processor of the main server preprocesses the received data, aligns it with process-specific analysis units, calculates the probability of defects in the process, and calculates the supply risk for each material by considering at least one of defect probability, connection information between the process and the material, inventory, order, lead time, and substitutability information. At the same time, the processor structures the field recording data to analyze the unstructured data, and finally, the communication unit transmits the defect probability, the supply risk for each material, and the unstructured data analysis results to a manager terminal. Through this integrated approach, it is possible to provide the effect of simultaneously enhancing the predictability of the manufacturing process and the responsiveness of the supply chain.

[0007] Specific details of other embodiments are included in the detailed description and drawings. Effects of the invention

[0008] According to the present invention as described above, it has various effects as follows.

[0009] The AI-based manufacturing supply chain management method and system according to the present invention provides the following significant effects. First, through a multi-layered approach integrating time-series analysis, categorical material analysis, and unstructured text analysis, it is possible to predict defects with significantly improved accuracy and reliability compared to existing single-model-based predictions. Second, by directly linking defect prediction results with material supply strategies, optimized material management is realized to prevent production stoppages caused by inventory shortages in advance and minimize capital inefficiency resulting from excessive inventory holding. Third, through the systematic analysis of unstructured information such as equipment alerts, worker comments, and customer claims, potential risk factors that are difficult to capture using process data alone can be detected early. Fourth, through non-intrusive integration with existing manufacturing information systems, it provides a practical solution that allows the utilization of advanced AI-based analysis functions while protecting system investments. Fifth, through intuitive visualization and the provision of systematized information, it significantly improves the speed and accuracy of managerial decision-making, thereby simultaneously ensuring the stability and efficiency of overall manufacturing operations.

[0010] The effects according to the present invention are not limited to those exemplified above, and various other effects are included in this specification. Brief explanation of the drawing

[0011] FIG. 1 is a block diagram illustrating the schematic configuration of an AI-based manufacturing supply chain management system according to one embodiment of the present invention. FIG. 2 is a diagram illustrating a processor of an AI-based manufacturing supply chain management system according to one embodiment of the present invention. FIG. 3 is a block diagram illustrating an AI-based manufacturing supply chain management method according to an embodiment of the present invention. Specific details for implementing the invention

[0012] Various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The advantages and features of the present invention, and the methods for achieving them, will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims.

[0013] Shapes, sizes, ratios, angles, numbers, etc. disclosed in the drawings for explaining embodiments of the present invention are exemplary, and therefore the present invention is not limited to the depicted details. Throughout the specification, the same reference numerals refer to the same components. Furthermore, in describing the present invention, if it is determined that a detailed description of related prior art may unnecessarily obscure the essence of the present invention, such detailed description is omitted. Where terms such as "includes," "has," or "is made up" are used in this specification, other parts may be added unless "only" is used. Where a component is expressed in the singular, it includes cases where it is included in the plural unless specifically stated otherwise.

[0014] In interpreting the components, they are interpreted to include a margin of error even in the absence of a separate explicit statement.

[0015] In the case of describing a positional relationship, for example, when the positional relationship between two parts is described using expressions such as 'on,' 'upper,' 'lower,' or 'next to,' one or more other parts may be located between the two parts unless 'immediately' or 'directly' is used.

[0016] When an element or layer is referred to as "on" another element or layer, it includes cases where another layer or element is placed directly on top of or in between.

[0017] Although terms such as "first," "second," etc., are used to describe various components, these components are not limited by these terms. These terms are used merely to distinguish one component from another. Accordingly, the first component mentioned below may be the second component within the technical scope of the present invention.

[0018] Throughout the specification, the same reference numerals refer to the same components.

[0019] The size and thickness of each component shown in the drawings are illustrated for convenience of explanation, and the present invention is not necessarily limited to the size and thickness of the illustrated components.

[0020] The features of each of the various embodiments of the present invention may be combined or combined with one another, either partially or wholly, and as will be fully understood by those skilled in the art, various technical interlocking and operation are possible, and each embodiment may be implemented independently of one another or together in an interlocking relationship.

[0022] FIG. 1 is a block diagram illustrating the schematic configuration of an AI-based manufacturing supply chain management system according to an embodiment of the present invention. FIG. 2 is a diagram illustrating a processor of an AI-based manufacturing supply chain management system according to an embodiment of the present invention. FIG. 3 is a block diagram illustrating an AI-based manufacturing supply chain management method according to an embodiment of the present invention.

[0023] Referring to FIGS. 1 to 3, the present invention relates to an AI-based manufacturing supply chain management system (1000) that predicts in advance quality defects that may occur in high-precision manufacturing processes such as semiconductors, secondary batteries, and precision assembly, and systematically provides information necessary for determining material supply and demand based on the prediction results. In modern manufacturing, with the spread of smart factory environments, the demand for production efficiency and quality stability is rapidly increasing, and especially in fields sensitive to yield changes, such as the semiconductor industry, even small process issues can have a massive impact on the overall production schedule and material supply plans.

[0024] This system goes beyond existing manufacturing information systems that rely on simple post-analysis or reports, enabling real-time monitoring of process status and quantitative prediction of defect probability. Furthermore, by directly linking predicted defect risks to material management strategies, it supports decision-making regarding inventory shortages, urgent orders, and the review of substitutes. Through this, manufacturers can minimize production stoppages caused by quality issues, ensure material supply stability, and maximize overall operational efficiency. Moreover, this invention can be implemented without system conflicts through non-intrusive integration with existing MES, WMS, and ERP systems, thereby making a substantial contribution to the digital transformation of the manufacturing industry.

[0026] As mentioned earlier, while systems such as MES, WMS, and ERP are widely implemented in the manufacturing industry, the reality is that most of these systems are utilized only for simple reporting or post-mortem analysis. The biggest limitation of existing systems is the lack of real-time process status analysis and defect prediction capabilities; in particular, there is a problem in that a linkage system between defect prediction and material management strategies has not been established at all.

[0027] Due to these technical limitations, in fields extremely sensitive to yield fluctuations, such as the semiconductor industry, the entire production schedule is repeatedly disrupted and material supply plans collapse whenever unexpected process issues arise. Furthermore, because unstructured information generated on-site, such as equipment alarms and worker comments, is not systematically analyzed, critical quality signals are frequently missed.

[0028] The AI-based manufacturing supply chain management method and system (1000) of the present invention was developed to overcome the fundamental limitations of such existing technologies. The system is designed to enable early detection of quality risks by process using AI analysis technology and to enable preemptive response by directly linking them to material management strategies. Furthermore, by structuring and analyzing even unstructured text data, it is possible to capture quality signals that are easily missed in existing systems.

[0030] Accordingly, with reference to FIGS. 1 to 3, an AI-based manufacturing supply chain management system (1000) according to one embodiment of the present invention includes a main server (100), an administrator terminal (200), and a plurality of external devices (300). Specifically, the main server (100) of the AI-based manufacturing supply chain management system (1000) is composed of a communication unit (110), a processor (120), and a storage unit (130), and the processor (120) may be configured to include a process defect rate calculation unit (121), an unstructured text analysis unit (122), and a material supply risk evaluation unit (123).

[0031] The main server (100) is a core component of the system and can perform the role of comprehensively analyzing and processing various data collected from external devices. The communication unit (110) smoothly performs data transmission and reception between the external device (300) and the administrator terminal (200) and can act as an intermediary for real-time data collection and the delivery of analysis results. The process defect rate calculation unit (121) within the processor (120) can quantitatively calculate and predict the probability of defect occurrence based on the collected process data and material data.

[0032] The unstructured text analysis unit (122) can systematically analyze unstructured data, such as inspection records or worker memos generated on-site, to extract and structure defect types. The material supply risk evaluation unit (123) can comprehensively evaluate the supply risk for each material by linking the previously calculated defect rate with material flow information. The storage unit (130) can safely store the data and results generated during all these analysis processes. The manager terminal (200) can support the manager's rapid decision-making by visually displaying the results analyzed by the main server (100), and multiple external devices (300) can perform the role of providing various manufacturing information in real time, such as production management, material management, quality inspection, and on-site records.

[0033]

[0034] Each of the previously introduced components is organically connected to perform comprehensive manufacturing supply chain management functions. First, the communication unit (110) of the main server (100) is responsible for transmitting and receiving data between multiple external devices (300), such as a production management device (310), a material management device (320), a quality inspection device (330), and a field recording device (340), and the manager terminal (200), and can collect process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data in real time. At this time, the communication unit (110) supports various communication protocols to ensure compatibility with existing manufacturing information systems and can perform stable data transmission while maintaining data integrity and security.

[0035] Next, the processor (120) serves as the core computing unit of the main server (100) and can play a pivotal role in comprehensively analyzing and processing various data collected through the communication unit (110). The processor (120) is configured to include a high-performance multi-core CPU and a GPU to enable complex AI algorithms and large-scale data processing, and can provide parallel processing capabilities to meet real-time analysis requirements. Additionally, the processor (120) is internally composed of a process defect rate calculation unit (121), an unstructured text analysis unit (122), and a material supply risk evaluation unit (123), each capable of performing specialized analysis functions.

[0036] Specifically, the process defect rate calculation unit (121) can preprocess the collected data and align it to process-specific analysis units, and then calculate the probability of process defects through a multi-layered analysis method. Through time-series-based analysis, multivariate time-series data such as temperature, pressure, and flow rate matched to the process time window can be processed by a recurrent neural network-based analysis module to derive time-series-based defect probabilities. At the same time, categorical material-based analysis can be performed using product type, process code, equipment type, work shift, material lot attribute, and inventory indicator as inputs, and categorical material-based defect probabilities can be calculated through a decision tree-based ensemble classifier. Additionally, an outlier detection algorithm can be applied to calculate an outlier score that quantifies the difference from the normal pattern, and this can be converted into a correction value normalized according to the field distribution.

[0037] Next, the unstructured text analysis unit (122) can systematically process unstructured data such as equipment alarm messages, worker comments, and customer claims collected from the field recording device (340). First, the text data can be aligned with the process time window and converted into normalized text through preprocessing such as unit and symbol standardization, abbreviation interpretation, and typo correction. Subsequently, texts with similar meanings can be grouped through TF-IDF vectorization and K-means clustering, and for each cluster, the frequency of occurrence, severity, duration of occurrence, and sensor consistency can be quantified to calculate a text risk score.

[0038] The material supply risk evaluation unit (123) can comprehensively evaluate the supply risk for each material by combining the defect probability derived from the process defect rate calculation unit (121) with the material requirements, inventory, lead time, and substitutability information. The expected value of the additional requirement for each material can be calculated through a policy function that reflects the corrected defect probability and the standard scrap rate, and the total requirement can be calculated by adding it to the basic requirement. Available inventory can be calculated based on the current inventory and the quantity scheduled to be received within the lead time, and the possibility of a shortage can be derived by comparing this with the total requirement and safety stock. In addition, a lead time risk considering the procurement lead time and a substitutability penalty reflecting the existence of substitute materials and substitution efficiency can be calculated, and then the comprehensive supply risk index for each material can be calculated by weighted combination.

[0039] Meanwhile, the storage unit (130) systematically stores collected process and material data, analysis results, and policy information, and can utilize process history, equipment status, inspection results, material inflow and outflow and movement paths, unstructured text data, and analysis results for post-verification and auditing by version control. The administrator terminal (200) visualizes and displays the defect rate, risk grade, and material risk level analyzed by the main server (100), and can support the administrator in making quick decisions such as order adjustment, material substitution, and production schedule change.

[0040] Finally, multiple external devices (300) can play a key role as data sources for the system. The production management device (310) provides production history, equipment operating status, work order details, etc., in real time, and can transmit detailed process data including time-series parameters such as LOT identification information, process stage, equipment identification information, start and end times, temperature, pressure, and flow rate. The material management device (320) manages material inflow and outflow, inventory status, material movement paths, etc., and can provide material-related information such as material production date, supplier history, material LOT elapsed days, inventory level, and ratio to safety stock. The quality inspection device (330) provides quality-related data such as inspection results and defect judgment records, and the field recording device (340) can collect and transmit unstructured data such as worker comments, alarm messages, and customer claims. These external devices can each operate independently while being connected to the main server (100) through the communication unit (110) to form an integrated data ecosystem.

[0041] In this way, all components are closely interconnected to implement an integrated system that simultaneously enhances manufacturing process predictability and supply chain responsiveness.

[0043] Referring to FIG. 3, the basic operation process of the AI-based manufacturing supply chain management system (1000) configured as described above can be examined. First, the communication unit (110) of the main server (100) can comprehensively collect process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data from the production management device (310), material management device (320), quality inspection device (330), and field recording device (340) (S110). At this stage, the communication unit (110) can convert data of different forms transmitted from each external device (300) into a standardized format, verify the integrity of the data, and then systematically store it in the storage unit (130). Time series parameters such as LOT identification information, process stage, equipment identification information, start and end times, temperature, pressure, and flow rate are collected from the production management device (310), and material input data such as material production date, supplier history, material LOT elapsed days, and inventory level can be obtained from the material management device (320).

[0044] Subsequently, the processor (120) of the main server (100) can preprocess the previously collected data and align it into process-specific analysis units, and then calculate the probability of process defects (S120). Specifically, the process defect rate calculation unit (121) can receive the collected time-series-based process data and material flow data as input and perform a comprehensive analysis. At this time, the time-series data is converted into a form suitable for analysis through preprocessing processes such as missing value interpolation, outlier removal, and normalization, and the material flow data can be aligned by reflecting the material requirements and quality impact of each process. The process defect rate calculation unit (121) can determine whether the process deviates from the normal range by analyzing the change patterns of process parameters such as temperature, pressure, and flow rate, and can predict the probability of defects in the current situation by learning the history of defect occurrences under similar past conditions. The calculated defect rate value can be classified and output as a process risk grade along with a reliability range.

[0045] In the next step, the processor (120) of the main server (100) can calculate the supply risk for each material by considering at least one of the previously calculated defect probability, process and material connection information, inventory, order, lead time, and substitutability information (S130). The material supply risk evaluation unit (123) can identify materials associated with processes with high defect rates based on defect rate information received from the process defect rate calculation unit (121), and calculate the supply risk by comprehensively considering the current inventory level of the material, procurement lead time, and availability of substitute materials. For example, if the defect rate increases in a specific process, the consumption of major materials input into that process is expected to increase, and the shortage risk can be calculated by comparing the current inventory amount with the expected consumption amount. Finally, the supply risk for each material can be classified into grades such as safety, caution, warning, and danger and output.

[0046] At the same time, the processor (120) of the main server (100) can perform the process of structuring field record data to analyze unstructured data (S140). The unstructured text analysis unit (122) can convert unstructured text data, such as quality inspection records, alarm messages, and worker comments collected from the field record device (340), into meaningful information using natural language processing technology. In this process, the text data is first converted into a standardized form through a normalization process, and then texts with similar meanings can be grouped through vectorization and clustering algorithms. The unstructured text analysis unit (122) can quantify the frequency of occurrence, severity, duration, and sensor consistency for each cluster to extract frequently mentioned problem types or concerns regarding specific equipment. These analysis results can be used as supplementary information to improve overall prediction accuracy by providing potential risk signals that are difficult to capture with process data alone.

[0047] Finally, the communication unit (110) of the main server (100) can transmit the previously analyzed defect probability, material supply risk, and unstructured data analysis results to the manager terminal (200) (S150). The communication unit (110) can integrate the analysis results derived from the process defect rate calculation unit (121), the material supply risk evaluation unit (123), and the unstructured text analysis unit (122), respectively, and transmit them to the manager terminal (200). On the manager terminal (200), these analysis results are visualized and displayed in the form of an intuitive dashboard, allowing the manager to grasp the defect rate status by process, the risk grade by material, and the types of recurring problems based on unstructured data at a glance. Through this, the manager can quickly make preemptive decisions such as identifying materials requiring urgent ordering, reviewing alternative materials, and adjusting production schedules, and can establish specific response measures such as order adjustment, material substitution, and production schedule change.

[0049] According to a manufacturing supply chain management method performed by an AI-based manufacturing supply chain management system (1000) according to one embodiment of the present invention, a processor (120) of a main server (100) can preprocess previously collected data and align it with process-specific analysis units, and then calculate the probability of process defects. Specifically, the process defect rate calculation unit (121) of the processor (120) within the main server (100) can calculate the probability of process defects more precisely and reliably.

[0050] In this process, the process defect rate calculation unit (121) of the processor (120) may further include the step of performing a time-series-based analysis using time-series data matched to a process time window as input and calculating a time-series-based defect probability; the process defect rate calculation unit (121) may further include the step of performing a category material-based analysis using product type, process code, equipment type, work shift, material lot attribute, and inventory indicator as input and calculating a category material-based defect probability; and the process defect rate calculation unit (121) may further include the step of performing outlier detection on a dataset integrating the time-series-based defect probability and the category material-based defect probability to calculate an outlier score, and converting the said outlier score into a correction value normalized according to the field distribution. Additionally, the process defect rate calculation unit (121) may include the step of calculating a combined defect probability using the time-series-based defect probability, the category material-based defect probability, the normalized correction value, and the text risk score calculated from the unstructured data analysis result.

[0051] To examine this more specifically, first, during the data input and preprocessing stage, the communication unit (110) can collect data from an external device and record it in the storage unit (130). From the production management device (310), process history data including LOT identification information, process stage, equipment identification information, start and end times, and time series parameters can be collected, and from the material management device (320), material input data including material production date, supplier history, material LOT elapsed days, inventory level, and ratio to safety stock can be obtained. From the quality inspection device (330), quality inspection data can be collected, and from the field recording device (340), unstructured text data such as equipment alarm logs, worker comments, and customer claims can be collected. The storage unit (130) can systematically store all of this data and transmit it to the processor (120).

[0052] Next, in the time series characteristic calculation step, the process defect rate calculation unit (121) can generate indicators such as a representative value, degree of dispersion, ratio of deviation from an allowable range, change trend between start and end, and energy distribution of a specific frequency band by analyzing time series parameters such as temperature, pressure, and flow rate at each process stage. This allows for the quantitative characterization of process stability, abnormal signs, and periodic variability.

[0053] In the subsequent categorical and material characteristic calculation stage, the process defect rate calculation unit (121) can analyze categorical data such as product type, process code, equipment type, and work shift together with material-related attributes such as production date, supplier history, and inventory level. Through this, external factors affecting the occurrence of defects can be reflected.

[0054] At the same time, in the unstructured text reflection stage, the unstructured text analysis unit (122) can cluster logs, comments, and claims collected from the field recording device (340) and evaluate severity, duration, and sensor event consistency to generate a text risk score (stxt). This value can be transmitted to the process defect rate calculation unit (121) and reflected complementarily.

[0055] Next, in the model-specific defect probability calculation step, the process defect rate calculation unit (121) can derive multiple defect probabilities and correction values ​​by performing different analysis procedures for each type of collected data. First, the time-series-based defect probability (pseq) can take multivariate time-series data, such as temperature, pressure, and flow rate recorded at each process stage, as input. These data are arranged in chronological order after correcting for missing values ​​or outliers, and can be fed into a recurrent neural network-based analysis module capable of learning the trend of change and variability over time. Through this process, the probability of a defect occurring within a specific time window can be calculated as a probability value.

[0056] Next, the categorical material-based defect probability (pcat) can take as input categorical information such as product type, process code, equipment type, and work shift, as well as material-related information such as material LOT attributes, material inventory levels, and ratios relative to safety stock. These features are processed by a decision tree-based ensemble classifier, which can calculate the likelihood of defect occurrence as another probability value by considering the combinations and interactions of each variable.

[0057] Additionally, the anomaly score (a) can be calculated by inputting a dataset integrating time series features, categories, and material features into an unsupervised learning-based anomaly detection algorithm. This is a numerical value representing how much individual data differs from a normal pattern, and a higher value indicates greater abnormality. This score is not used as is, but is converted into a range between 0 and 1 based on the distribution of field data or a policy-set threshold, and can be utilized as a normalized correction value (g(a)). This allows for adjusting the reliability of the defect prediction results; for example, a high anomaly score reflects greater risk, while a low score can conservatively mitigate it.

[0058] Finally, the text risk score (stxt) transmitted from the unstructured text analysis unit can be used as an additional input. This is a value obtained by structuring repetitive defect types from text data such as equipment alarms, worker comments, and customer claims, and can complement risk signals that are difficult to capture with process data alone.

[0059] Finally, the above three prediction results can be applied to the following mathematical formula 1 to calculate the probability of a failed connection (P).

[0061] [Mathematical Formula 1]

[0062]

[0064] Here, logit may be a function that converts probability into log-likelihood. The process defect rate calculation unit (121) can derive the joint defect probability (P) by comprehensively combining the values ​​obtained in this way. The final joint defect probability is calculated as P = 1 / (1+e^(-z)), and the reliability of the prediction probability can be improved by applying a probability correction technique (flat correction or isotonic regression) using a verification dataset. In addition, a confidence interval that a manager can refer to can be provided by analyzing the prediction error distribution based on cross-validation. For example, for a specific LOT or process stage, the calculated value and uncertainty can be displayed simultaneously, such as "defect probability 0.32 (±0.07)".

[0065] This embodiment is technically differentiated in that it goes beyond simply applying an artificial intelligence model to integrating time-series stability indicators of process data, material-related risk factors, and recurring defect information based on unstructured text, and provides quantitative and reliable predictions of defect probability through fusion formulas, corrections, and confidence interval calculation procedures.

[0067] Based on the quantitative and reliable defect probability prediction value calculated in another embodiment of the present invention described above, according to another embodiment of the present invention, a material supply risk evaluation unit (123) within the processor (120) of the main server (100) can quantify the supply risk for each material and present a priority order for action by combining the defect probability and confidence interval with material requirements, inventory, lead time, and substitutability information. This module can operate in a non-intrusive structure that provides judgment information to an administrator without directly linking with ERP / MES to automatically execute orders.

[0068] Specifically, the processor (120) of the main server (100) can calculate the supply risk for each material by considering at least one of the previously calculated defect probability, process and material connection information, inventory, order, lead time, and substitutability information. In this process, the material supply risk evaluation unit (123) of the processor (120) may include the step of calculating the expected value of the additional requirement for each material based on the process and material connection information based on a policy function reflecting the corrected defect probability and the standard scrap rate; the step of the material supply risk evaluation unit (123) calculating the basic requirement as the product of the unit requirement coefficient and the planned production volume; and the step of the material supply risk evaluation unit (123) calculating the total requirement for each material by period by summing the basic requirement and the expected value of the additional requirement.

[0069] Additionally, the method may include the steps of calculating available inventory by summing the current inventory and the quantity scheduled to be received within the lead time, deriving a shortage margin by comparing the available inventory with the total required quantity and safety stock, calculating the probability of a shortage using the shortage margin and past demand variability, calculating the lead time risk using the lead time required for procurement in the event of a shortage, and calculating a substitutability penalty using the existence of substitute materials, minimum substitution costs, and substitution efficiency, and then calculating a comprehensive supply and demand risk index for each material by weighted combination of the probability of a shortage, the lead time risk, and the substitutability penalty.

[0070] To examine this more specifically, first, in the data definition stage, the material supply risk assessment unit (123) can systematically collect and organize various types of data. Requirement data includes unit requirement coefficients for each process stage, planned production volume, and standard scrap rate, and can be provided by the production planning module and process management module of ERP and MES. Inventory and procurement data include current stock, order quantity, expected arrival date, safety stock, and procurement lead time, and can be provided by the material management device (320) or the ERP material module. Substitutability data includes a list of substitute materials capable of performing the same function, the substitution efficiency (quality, performance loss rate) and substitution cost (additional purchase unit price, switching cost) of each material, and can be obtained through the company's BOM management module or manager input. Prediction data may include the defect probability by process and its confidence interval calculated in Additional Example 1.

[0071] Next, in the step of calculating the expected value of additional requirements, the material supply risk evaluation unit (123) can calculate the basic requirements and additional requirements, respectively. The basic requirements are calculated as "unit requirement coefficient χ planned production quantity," and the expected value of additional requirements can be calculated by a policy function that combines the defect probability and the standard scrap rate. For example, the expected value of additional requirements can be defined as (defect probability χ standard scrap rate) χ basic requirements, and can be adjusted linearly, by segment weighting, etc., according to the actual field policy. That is, as the defect probability increases, a value can be calculated that adds to the material consumption by reflecting the increase in rework and scrap.

[0072] In the step of calculating the probability of shortage occurrence, the material supply risk assessment unit (123) can calculate available inventory by summing the current inventory and the quantity scheduled to be received within the lead time. The shortage margin is calculated by subtracting the total required quantity and safety stock from the available inventory, and if the shortage margin is negative, it can be determined to be in a state of shortage risk. The probability of shortage occurrence can be statistically calculated by considering the variability of the required quantity and the probability that the shortage margin will turn negative. For example, after estimating the standard deviation of the required quantity based on past demand fluctuation data, the probability of "shortage margin < 0" can be estimated based on a normal distribution or an empirical distribution.

[0073] Next, in the lead time risk calculation step, the material supply risk assessment unit (123) can calculate a greater risk the longer the lead time required to secure replacement stock when a shortage occurs. The lead time risk can be calculated by multiplying the probability of a shortage by the lead time, for example, or the risk can be non-linearly weighted if the lead time exceeds a specific standard (e.g., 2 weeks). Thus, the lead time risk can represent a numerical value of the "severity of production disruption in case of procurement delay."

[0074] At the same time, in the substitutability penalty calculation stage, the material supply risk assessment unit (123) may apply the penalty differentially depending on whether there is a substitute material. If there is no substitute material, the penalty is applied at the maximum value, and if there is a substitute material, the penalty can be calculated by reflecting the minimum substitution cost and substitution efficiency. For example, the penalty can be defined as (additional cost ratio + performance loss rate), and this value can numerically express "how difficult it is when the relevant material is in short supply."

[0075] Finally, in the comprehensive supply risk index calculation stage, the material supply risk evaluation unit (123) can derive a final risk index by integrating all previously calculated elements. The comprehensive risk index Rm for material m can be calculated by the following mathematical formula 2 by weighting the expected value of additional requirements, the probability of shortage, the lead time risk, and the substitutability penalty.

[0077] [Mathematical Formula 2]

[0078]

[0080] Here, the weights (γ1–γ4) are pre-set according to the importance of the material group or product group and may be adjustable according to administrator policies. The final index is normalized to a range of 0–1 or 0–100 and can be mapped to "Safe / Caution / Warning / Danger" grades.

[0081] In the output and utilization stage, the communication unit (110) of the main server (100) can transmit the final risk index and grade to the manager terminal (200). For each material, recommendation tags such as "urgent ordering required," "review alternative materials," and "production plan adjustment" are displayed together to support the manager's rapid decision-making. Both the input and output results are version-controlled and can be utilized for post-verification and auditing.

[0082] The technical significance of this embodiment lies in the fact that, first, it can provide a differentiated effect by calculating the expected value of additional material requirements based on defect prediction results; second, by reflecting the possibility of stock shortages, lead time risks, and substitutability penalties together; and third, by enabling proactive and quantitative material supply and demand decision-making beyond simple inventory monitoring. Through this, manufacturers can predict the risk of material shortages caused by defects in advance and establish efficient material management strategies.

[0084] In conjunction with the material supply risk assessment described above, according to another embodiment of the present invention, an unstructured text analysis unit (122) within the processor (120) of the main server (100) can match unstructured text, such as equipment alarm messages, worker comments, and customer claims occurring at the production site, in units of time and process, extract clusters of similar types, and then quantify the severity, frequency of occurrence, duration, and sensor consistency of each cluster to calculate a text risk score. The score can be combined as an input to the defect probability calculation step described above to improve overall prediction accuracy.

[0085] Specifically, in the process of the processor (120) of the main server (100) structuring field record data to analyze unstructured data, the process may include the steps of aligning the text with a process time window and converting the text into normalized text, vectorizing the normalized text to perform clustering, calculating the frequency of occurrence, severity, duration of occurrence, and sensor consistency for each cluster, and calculating a text risk score by applying a weight to the frequency of occurrence, severity, duration of occurrence, and sensor consistency for each cluster and summing them.

[0086] To examine this more specifically, first, in the data definition and time alignment stage, the unstructured text analysis unit (122) can systematically collect and classify various types of text data. Equipment alarm data consists of an equipment identifier, a time of occurrence, and the original text of the alarm message, and worker comment data may include a LOT identifier (if available), a process stage (if available), a time of recording, and the original text of the comment. Customer claim data may consist of a product code, a time of receipt, and the original text of the defect description. Each text item can be aligned to a LOT / process according to the time window [ts, te] and the extension permission window [ts-δ, te+δ] of the process stage. Text that does not directly contain LOT or process information is connected to the nearest process window using time, equipment, and line information, and the alignment reliability can be internally displayed.

[0087] Next, in the preprocessing and term normalization stage, the unstructured text analysis unit (122) can convert the collected text data into a form suitable for analysis. The text can be converted into a normalized token sequence through unit and symbol standardization (degrees, %, mm, etc.), abbreviation interpretation (equipment, process code dictionary), processing of negation and emphasis expressions ("no alarm", "hard fail", etc.), and typo correction. Considering a mixed Korean and English environment, morpheme-based tokenization and character n-grams can be used in parallel, and a weighted dictionary can be applied to process and defect keywords (e.g., temperature overheating, vacuum leakage, surface foreign matter).

[0088] In the vector representation and clustering stage, the unstructured text analysis unit (122) can convert normalized text into numerical data and group texts with similar meanings. The normalized text can be quantified using word frequency-inverse document frequency (TF-IDF), and if necessary, the noise dimension can be reduced using principal component analysis (PCA). Subsequently, similar sentences are grouped using K-means clustering (K-means), and the number of clusters (k) can be determined by considering the silhouette coefficient and the internal fit indicator together. Each cluster can be automatically labeled with a top weighted keyword (e.g., "heater drift / overheating," "surface scratch," "foreign object detection").

[0089] Next, in the quantification stage of severity, frequency, duration, and consistency, the unstructured text analysis unit (122) can quantitatively evaluate the characteristics of each cluster. Severity can be determined by classifying text per cluster into critical / major / minor grades according to predefined rules, and mapping each grade to a numeric weight (e.g., 3 / 2 / 1). The frequency of occurrence can be calculated as the number of occurrences of the corresponding cluster in the target time window [ts-δ, te] (or the ratio relative to the LOT). Duration can be calculated as the ratio of the intervals in which the cluster appears consecutively when the time window is divided into a certain number of units (e.g., 0.6 when it appears in 6 out of 10 intervals). Sensor consistency can be calculated as the simultaneous occurrence rate of sensor-based abnormal events (exceeding the allowable range, sudden fluctuation, etc.) and the corresponding cluster in the same time window, and the degree to which the text and sensor are captured together can be calculated as a range of 0 to 1.

[0090] Finally, in the text risk score calculation step, the unstructured text analysis unit (122) can derive a final risk score by integrating all previously calculated elements. For a target LOT and process window, the text risk score (stxt) can be defined as shown in the following mathematical formula 3.

[0092] [Mathematical Formula 3]

[0093]

[0095] Here, the cluster set (C) represents a set of text groups with the same or similar meanings as a result of classifying normalized, vectorized text by a clustering algorithm (such as K-means). Each cluster (c) can be generated from equipment alarms, worker comments, customer claims, etc. The cluster weight (αc) is a coefficient representing the importance that a specific cluster has in the overall risk assessment, and can be set according to the manager's policy, learning results based on historical data, or the importance of materials or processes. For example, a high weight may be assigned to clusters containing critical equipment alarms, and a low weight to clusters of simple notification comments.

[0096] The frequency of occurrence (freqc) can be calculated as the number of times cluster c appears within a given time window [ts, te], or as the occurrence rate relative to the LOT unit. Examples of calculation formulas include the absolute value of the number of occurrences of text in cluster c or the relative value (number of occurrences of cluster c) χ (number of LOTs in the process). The severity (sevc) is a numerical value representing the semantic importance and criticality of the text included in the cluster. Text can be classified into critical (3), major (2), minor (1), etc., according to predefined rules, and converted into the corresponding numerical value. If necessary, an importance dictionary by defect type can be applied to enable automatic mapping.

[0097] Persistence (perc) is an indicator representing the degree to which a cluster occurs continuously over a specific time window. It can be calculated as the ratio of the intervals in which the cluster appeared after dividing the time window into fixed units. For example, if a cluster occurred in 6 out of 10 intervals, the persistence can be calculated as 0.6. Sensor consistency (cohc) is the co-occurrence rate of sensor-based anomaly events corresponding to the cluster within the same time window. For instance, it can be normalized to a value ranging from 0 to 1 by checking whether the actual temperature sensor also exceeded the allowable range at the time the "overheating" cluster occurred, and dividing the number of co-occurrences by the total number of cluster occurrences. A value closer to 1 may indicate a stronger match between the text and the sensor data.

[0098] This entire process can proceed in the order of normalizing and vectorizing the text, performing clustering, calculating freqc, sevc, perc, and cohc values ​​for each cluster, multiplying them by a weight αc defined by the manager, and summing them. The final stxt is normalized to a range of 0 to 1 and can be provided as input for the defect prediction and material risk analysis stages.

[0099] The technical significance of this embodiment lies in its differentiation in that it enhances the explanatory power and reliability of process prediction by quantifying unstructured text not merely based on the presence or absence of keywords, but also on clustering, severity, duration, and sensor consistency. Through this, potential risk signals that are difficult to capture with process data alone can be systematically analyzed, thereby contributing to the improvement of the prediction performance of the overall manufacturing supply chain management system.

[0101] The AI-based manufacturing supply chain management method and system (1000) of the present invention, which is completed through the process of structuring unstructured text data and repetitive type recognition described in detail above, can overcome the fundamental limitations of existing manufacturing information systems and provide comprehensive effects that can make a substantial contribution to the digital transformation of manufacturing.

[0102] First, the present invention can dramatically improve the proactive response capabilities of the manufacturing site through a predictive quality management system. By calculating a combined defect probability (P) that integrates a time-series-based defect probability (pseq), a category-material-based defect probability (pcat), a normalized correction value (g(a)), and a text risk score (stxt) through a multi-layered analysis method performed by the process defect rate calculation unit (121) of the main server (100), it is possible to make quantitative and highly reliable defect predictions that go beyond simple statistical approaches or individual model-based predictions. Through this, the manufacturer can take preemptive measures before quality problems actually occur and can significantly reduce production downtime and losses caused by defects.

[0103] Second, the present invention can simultaneously improve the precision and efficiency of material supply management. By having the material supply risk evaluation unit (123) calculate the expected value of additional requirements based on the defect prediction results and provide a material-specific supply risk index (Rm) that comprehensively considers the possibility of shortage, lead time risk, and substitutability penalty, the manager can make proactive and quantitative material supply decisions that go beyond simple inventory monitoring. This enables optimized material management that prevents production delays caused by inventory shortages while minimizing capital inefficiency caused by excessive inventory holding.

[0104] Third, the present invention can capture important quality signals that are easily missed in existing systems through the systematic utilization of unstructured information. By clustering equipment alarm messages, worker comments, customer claims, etc., and quantifying severity, frequency of occurrence, duration, and sensor consistency to convert them into a text risk score (stxt), potential risk factors that are difficult to capture with process data alone can be systematically analyzed and reflected in a prediction model. This approach can improve the explanatory power and reliability of predictions by integrating field empirical knowledge and intuitive judgment into a quantitative analysis system.

[0105] Fourth, the present invention can provide innovative functions while ensuring compatibility with existing manufacturing information systems. The communication unit (110) is non-intrusively linked with a plurality of external devices (300), such as a production management device (310), a material management device (320), a quality inspection device (330), and a field recording device (340), so that it can be introduced without conflict with existing MES, WMS, and ERP systems, and can support gradual and stable digital transformation. This can provide a practical solution that allows manufacturers to utilize advanced AI-based analysis functions while protecting their existing system investments.

[0106] Finally, the present invention can significantly improve the work efficiency of field operations personnel through intuitive visualization and decision support functions via a manager terminal (200). As complex analysis results are systematically organized and provided as defect rates by process, risk grades by material, and precautions based on unstructured data, the manager can make quick and accurate decisions, thereby enabling the implementation of an integrated management system that can simultaneously ensure the stability and efficiency of overall manufacturing operations.

[0108] It goes without saying that various embodiments of the present invention can be implemented as new embodiments by combining one or more of them.

[0110] An AI-based manufacturing supply chain management method according to various embodiments of the present invention may include: (a) a communication unit of a main server receiving at least one of process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data from a plurality of external devices including at least one of a production management device, a material management device, a quality inspection device, and a field recording device; (b) a processor of the main server preprocessing the data received in step (a) to align it with process-specific analysis units and then calculating the probability of defects in the process; (c) a processor of the main server calculating the supply risk for each material by considering at least one of the probability of defects, connection information between the process and the material, inventory, order, lead time, and substitutability information; (d) a processor of the main server structuring field recording data to analyze unstructured data; and (e) a communication unit of the main server transmitting the probability of defects, the supply risk for each material, and the unstructured data analysis results to a manager terminal.

[0111] According to another feature of the present invention, step (b) may further include: a process defect rate calculation unit of a processor performing a time-series-based analysis with time-series data matched to a process time window as input and calculating a time-series-based defect probability; a process defect rate calculation unit performing a categorical material-based analysis with product type, process code, equipment type, work shift, material lot attribute and inventory indicator as input and calculating a categorical material-based defect probability; and a process defect rate calculation unit performing outlier detection on a dataset integrating the time-series-based defect probability and the categorical material-based defect probability to calculate an outlier score and converting the outlier score into a correction value normalized according to the field distribution.

[0112] According to another feature of the present invention, step (b) may further include the step of a process defect rate calculation unit calculating a combined defect probability using a time series-based defect probability, a category material-based defect probability, a normalized correction value, and a text risk score calculated from the results of unstructured data analysis.

[0113] According to another feature of the present invention, step (c) may further include: a step in which a material supply risk evaluation unit of a processor calculates an expected value of additional requirements for each material based on process and material connection information based on a policy function reflecting a correction defect probability and a reference scrap rate; a step in which a material supply risk evaluation unit calculates a basic requirement as the product of a unit requirement coefficient and a planned production volume; and a step in which a material supply risk evaluation unit calculates a total requirement for each material by period by summing the basic requirement and the expected value of additional requirements.

[0114] According to another feature of the present invention, step (c) may further include: calculating available inventory by summing the current inventory and the quantity scheduled to be received within the lead time; deriving a shortage margin by comparing the available inventory with the total required quantity and safety stock; calculating the probability of a shortage using the shortage margin and past demand variability; calculating the lead time risk using the lead time required for procurement in the event of a shortage; and calculating a substitutability penalty using the existence of substitute materials, minimum substitution cost, and substitution efficiency, and then calculating a comprehensive supply and demand risk index for each material by weighted combination of the probability of a shortage, the lead time risk, and the substitutability penalty.

[0115] According to another feature of the present invention, step (d) may further include the steps of aligning equipment alarms, worker comments, and customer claims to a time window according to the time of occurrence and equipment identification information, converting text into normalized text, vectorizing the normalized text to perform clustering, and calculating the frequency of occurrence, severity, duration of occurrence, and sensor consistency for each cluster.

[0116] According to another feature of the present invention, step (d) may further include the step of calculating a text risk score by applying cluster-specific weights to cluster-specific occurrence frequency, severity, duration of occurrence, and sensor consistency and summing them.

[0117] An AI-based manufacturing supply chain management system according to one embodiment of the present invention comprises a plurality of external devices, a main server, and a manager terminal. The plurality of external devices include at least one of a production management device, a material management device, a quality inspection device, and a field recording device. The main server includes a communication unit that performs data transmission and reception with a plurality of external devices, a storage unit that stores process history, equipment status, inspection results, material inflow / outflow and movement paths, unstructured text data, and analysis results, and a processor that processes data stored in the storage unit. The processor includes a process defect rate calculation unit that preprocesses received data, aligns it to process-specific analysis units, and calculates the probability of defects in the process; an unstructured text analysis unit that normalizes, vectorizes, and clusters unstructured text data and calculates a text risk score using cluster-specific indicators and weights; and a material supply risk evaluation unit that calculates the supply risk for each material by combining defect probability with material-specific requirements, inventory, lead time, and substitutability information. The manager terminal can receive and display the process defect rate and the material-specific supply risk from the main server.

[0119] Although the present invention has been described above with specific details such as specific components, limited embodiments, and drawings, this is provided only to aid in a more comprehensive understanding of the invention and the invention is not limited to the above embodiments, and a person skilled in the art to which the invention belongs can make various modifications and variations from this description.

[0120] Accordingly, the scope of the present invention is not limited to the embodiments described above, and all variations equivalent to or equivalent to the claims set forth below, as well as the claims described below, shall be considered to fall within the scope of the concept of the present invention. Explanation of the symbols

[0122] 1000: AI-based manufacturing supply chain management system 100: Main Server 110: Communications Department 120: Processor 121: Process Defect Rate Calculation Unit 122: Unstructured Text Analysis Unit 123: Material Supply Risk Assessment Department 130: Storage section 200: Administrator Terminal 300: External device

Claims

Claim 1 In an AI-based manufacturing supply chain management method, the method comprises: (a) a communication unit of a main server receiving at least one of process history, equipment status, inspection results, material inflow / outflow and movement paths, and unstructured text data from a plurality of external devices including at least one of a production management device, a material management device, a quality inspection device, and a field recording device; (b) a processor of the main server preprocessing the data received in step (a) to align it into process-specific analysis units and then calculating the probability of defects in the process; (c) a processor of the main server calculating the supply risk for each material by considering at least one of the probability of defects, process-material connection information, inventory, order, lead time, and substitutability information; (d) a processor of the main server structuring field recording data to analyze unstructured data. and (e) a step in which the communication unit of the main server transmits the defect probability, the material-specific supply risk, and the unstructured data analysis results to an administrator terminal, wherein step (c) comprises: a step in which the material supply risk evaluation unit of the processor calculates the expected value of the additional requirement for each material based on the connection information between the process and the material, based on a policy function reflecting the corrected defect probability and the standard scrap rate; a step in which the material supply risk evaluation unit calculates the basic requirement as the product of the unit requirement coefficient and the planned production volume; a step in which the material supply risk evaluation unit calculates the total requirement for each material by period by summing the basic requirement and the expected value of the additional requirement; a step in which available inventory is calculated by summing the current inventory and the quantity scheduled to be received within the lead time, and the shortage margin is derived by comparing the available inventory with the total requirement and the safety stock; a step in which the probability of a shortage occurs is calculated using the shortage margin and past demand volatility; and a step in which the lead time risk is calculated using the lead time required for procurement in the event of a shortage.The AI-based manufacturing supply chain management method further includes the step of calculating a substitutability penalty using the existence of substitute materials, minimum substitution costs, and substitution efficiency, and then calculating a comprehensive supply and demand risk index for each material by weighted combination of the probability of shortage, the lead time risk, and the substitutability penalty, wherein the probability of shortage calculates the probability that the shortage margin turns negative, and the lead time risk calculates the probability of shortage by multiplying the lead time by the probability of shortage. Claim 2 In claim 1, the step (b) further comprises: a step in which a process defect rate calculation unit of the processor performs a time-series-based analysis using time-series data matched to a process time window as input and calculates a time-series-based defect probability; a step in which a process defect rate calculation unit performs a category-based material analysis using product type, process code, equipment type, work shift, material lot attribute, and inventory indicator as input and calculates a category-based material defect probability; and a step in which a process defect rate calculation unit performs outlier detection on a dataset integrating the time-series-based defect probability and the category-based material defect probability to calculate an outlier score, and converts the outlier score into a correction value normalized according to the field distribution, thereby providing an AI-based manufacturing supply chain management method. Claim 3 In paragraph 2, the above step (b) further comprises the step of the process defect rate calculation unit calculating a combined defect probability using the time series-based defect probability, the category material-based defect probability, the normalized correction value, and the text risk score calculated from the unstructured data analysis result, in an AI-based manufacturing supply chain management method. Claim 4 delete Claim 5 delete Claim 6 In claim 1, the above step (d) further comprises: a step of aligning equipment alarms, worker comments, and customer claims to a fixed time window according to the time of occurrence and equipment identification information, and converting the text into normalized text; a step of vectorizing the normalized text to perform clustering; and a system for calculating the frequency of occurrence, severity, duration of occurrence, and sensor consistency for each cluster, an AI-based manufacturing supply chain management method. Claim 7 In claim 6, the above step (d) further comprises the step of calculating a text risk score by applying cluster-specific weights to cluster-specific occurrence frequency, severity, duration of occurrence, and sensor consistency and summing them, in an AI-based manufacturing supply chain management method. Claim 8 In an AI-based manufacturing supply chain management system comprising a plurality of external devices, a main server, and an administrator terminal, the plurality of external devices include at least one of a production management device, a material management device, a quality inspection device, and a field recording device, and the main server comprises: a communication unit that performs data transmission and reception with the plurality of external devices; and a storage unit that stores process history, equipment status, inspection results, material inflow / outflow and movement paths, unstructured text data, and analysis results.The system includes a processor that processes data stored in the storage unit, wherein the processor includes: a process defect rate calculation unit that preprocesses received data, aligns it to process-specific analysis units, and calculates the probability of process defects; an unstructured text analysis unit that normalizes, vectorizes, and clusters the unstructured text data and calculates a text risk score using cluster-specific indicators and weights; and a material supply risk evaluation unit that calculates material supply risk by combining the defect probability with material-specific requirements, inventory, lead time, and substitutability information. The manager terminal receives and displays the process defect rate and material-specific supply risk from the main server. The material supply risk evaluation unit calculates the expected value of additional requirements for each material based on the connection information between the process and the material, based on a policy function that reflects the corrected defect probability and the standard scrap rate. The material supply risk evaluation unit calculates the basic requirements as the product of the unit requirement coefficient and the planned production volume. The material supply risk evaluation unit calculates the total requirements for each material by period by summing the basic requirements and the expected value of additional requirements, and sums the current inventory and the quantity scheduled to be received within the lead time. An AI-based manufacturing supply chain management system that calculates available inventory, derives a shortage margin by comparing the available inventory with the total required quantity and safety stock, calculates the probability of a shortage using the shortage margin and historical demand volatility, calculates a lead time risk using the lead time required for procurement in the event of a shortage, calculates a substitutability penalty using the existence of substitute materials, minimum substitution costs, and substitution efficiency, and then calculates a comprehensive supply and demand risk index for each material by weighted combination of the shortage probability, the lead time risk, and the substitutability penalty; wherein the shortage probability calculates the probability that the shortage margin will turn negative, and the lead time risk is calculated by multiplying the shortage probability by the lead time.

Images